Linux HW and SW RAID Benchmarked

An anonymous reader writes "A Norwegian site has written up an article with various RAID solutions benchmarked using both bonnie++ and dbench. The result shows a lot of surprises, especially when comparing low end sw RAID with high end hw RAID. The text is in Norwegian but the numerous graphs are self explanatory. It does look like a few kernel drivers need a little tweaking."

Comments

Norwegian?

[jack_call]

Too bad it's in Norwegian :(

Re:Norwegian?

[dm(Hannu)]

A møøse once bit my sister...

Re:Norwegian?

[ABCC]

I bet this was posted by a norwegian troll like these:

http://www.nordskip.com/vestfold/images/trolls.jpg

Re:Norwegian?

[Anonymous Coward]

Probably because she's a socalled "hælvetes sjarkhore"?

Re:Norwegian?

[fstanchina]

Beware, she could develop moositis and write toy operating systems for fun.

Re:Norwegian?

[Anonymous Coward]

Could be worse. Could be Glaswegian.

Re:Norwegian?

[Anonymous Coward]

write toy operating systems for fun.Why would she want to write Windows?

Re:Norwegian?

[Anonymous Coward]

Best start with Linux and once she has got a bit of experience maybe then take a shot at a real operating system.

Re:Norwegian?

[HG2]

Too bad it's down now :(

Re:Norwegian?

[dm(Hannu)]

No realli! She was karving her initials øn the møøse with the sharpened end of an interspace tøøthbrush given her by Svenge - her brother-in-law - an Oslo dentist and star of many Norwegian møvies: "The Høt Hands of an Oslo Dentist", "Fillings of Passion", "The Huge Mølars of Horst Nordfink".

(Come on, don't you remember our favourite Norwegian subtitles?)

Re:Norwegian?

[Anonymous Coward]

Mynd you, møøse bites Kan be pretty nasti...

Re:Norwegian?

[Anonymous Coward]

Dangit. I'm looking around for the (-1, Didn't get the joke) mod.

sure

[Anonymous Coward]

I dont understand them at all, so they mean linux fails there?no too efficent? as good as windows? as bad as windows?

you could explain it a little bit, please

I think that the results are obvious

[suso]

Surprising? No. Reading the results I can see that software raid is generally slower than hardware raid and that some of the SCSI drivers are not completely tweaked, probably because they can't get enough information from the manufacturer.

Re:I think that the results are obvious

[Frumious Wombat]

Drilling through the article with my utterly minimal norwegian (Prarie Home Companion + German + exposure to Danish coworkers), I think I've distilled the following:

Cache on the LSI RAID controller is 1/2 the adaptec. Performance is comparable, though not equivalent.

All of the controllers are 64-bit.

Adaptec SCSI is good for both hardware RAID and software RAID.

LSI has good hardware SCSI RAID only.

Don't use current SATA controllers (RAID or Otherwise) for best performance.

Does anybody with access to a good collection of modern hardware care to re-run this test in a language that Babelfish understands?

Re:I think that the results are obvious

[martok]

Ok, this comment is uninformed as I count myself among those unable to read the article. Would also consider myself a raid amateur.

I ran some benchmarks a while ago for my own server with four 15k scsi drives softraid5d on a dual channel aic7xxx card against an Adaptec hardware raid controler with write cache and 128mb of ram. Though the hardware did take load off of the cpu, read/write performance was much better with the software raid setup and since the machine was smp, the raid overhead wasn't noticable for our application.

After doing some reading on the subject, I realized this is normal but I was surprised at how much of a speed hit we would have taken if going with a hardware solution.

Re:I think that the results are obvious

[tota]

No they are not!
SiI 3114 did really well and it is cheap.

Never mind all these other posts claiming that SCSI beats the crap out of everything else, it does not!

SCSI is bloody expensive and only marginally faster in these benchmarks. Now, unless fast disk access is the only way to improve your systems performance, you are probably better off using SiI 3114 and having many more of those.

Now that does not cover issues like hotswap support, noise, MTBF, etc...

But still it was an interresting read (albeit in Norwegian..)

Re:I think that the results are obvious

[smartdreamer]

Jeg vil rette en advarsel til alle dere som skal ut å handle kontrollere etter dette. Sjekk_nøye_om kontrolleren er støttet av kjernen! Jeg bruker google for å se etter referanser til kortet på mailing-lister og lignende, men det hjelper fint lite når du sitter der med din Debian og det eneste du har tilgjengelig er binære RedHat-drivere som du ikke klarer å jukse inn i kjernen. Nå er det bare å løpe til konsollene og teste disksystemene deres. Denne testen gir ikke annet enn indikasjoner på hva dere bør velge. Så jeg tillater meg å komme med et lite råd helt til slutt: Kjør gjerne tester selv!

Having read this, I also think that the results are obvious... Unfortunately, I can't tell on which side.

And... what is a tilgjengelig? Never seen such thing! ;)

Re:I think that the results are obvious

[aaronl]

Actually, SCSI does beat the crap out of SATA, for many reasons. First is the reliability... they use better components on the SCSI drives, and they have better warrantees. Second is the higher spindle speeds. Third is SATA only lets you have *one* drive. SCSI lets you have 15. You also get real command queuing, more reliable controllers, longer cable lengths, external devices, hotswapping, and multipath I/O. You also get access to a massive amount of existing hardware, instead of waiting for SATA versions of things to come out. Look how long it took before a single SATA optical drive was released!

I never understood why the industry went with something silly like "Hey, lets take IDE and make a new connector. That'll fix ALL the problems!". They could've just chosen one of the SCSI variants out there, and called that the next big consumer thing. Would've saved tons of money in manufacturing and design. You could've leveraged your existing hardware lines to produce the drives and used existing chipsets to drive them.

I do agree though, SCSI doesn't beat the crap out of everything as a law. There are junk controllers and drives out there, and they perform badly. But if you buy a good controller and good drives, SCSI beats the crap out of most everything else.

This benchmark doesn't seem to have configured things very well, anyway. It looks like a very small number of drives were used, the minimum for the RAID type. This means minimum performance.

If SCSI had been used and you had like five or seven drives, it would've screamed! Also, different things do RAID different ways. On some setups a RAID1 will write to all discs simultaneous, and read from only one. You can seriously improve throughput by reading different chunks off each drive simultaneously. Same idea as RAID0, except it doesn't suck for reliability.

Re:I think that the results are obvious

[sp0rk173]

the SCSI drivers are not completely tweaked, probably because they can't get enough information from the manufacturer.

No, it's obvious from the results that linux coders suck. Like you said, not surprising.

Re:I think that the results are obvious

[chinakow]

Try disabling the cache on the raid card, it will making writing a lot of files faster, cache is good if you are writing less than the cache size to the disk on a regular basis, like small text files, but if you are doing CAD files or large photoshop files then the cache is really just a hinderance as the card fills the cache then starts writing to disk. disabling cache also saves you from needing a battery backup on the raid controller. so try disabling cache and trying again. and compare the difference, I think you will see the raid controller doing better in read/write performance. post and let me know if I am wrong(it wouldn't be the first time).

Re:I think that the results are obvious

[51mon]

"they use better components on the SCSI drives"

Many modern SATA drives are physically identical to the SCSI disks in terms of the disk components, the SCSI drives are more expensive because of the lower volumes shipped, and the more complex controllers.

Last study I read said that whilst the SCSI disks outperformed the SATA, pricing was such that they could buy many more disks (something like 3 to 1 ratio), and SCSI simply isn't that much better that it can outperform three times as many disks.

Besides for most purposes it is either total throughput (which isn't very different), or blocking writes to disk, the later bottleneck generally better met by doing the blocking writes to something that doesn't have to spin than to get more faster spinning things.

Re:I think that the results are obvious

[myowntrueself]

"Drilling through the article with my utterly minimal norwegian (Prarie Home Companion + German + exposure to Danish coworkers), I think I've distilled the following"

I'm sure there was something in there about lutfisk being particularly delicious. Though I always thought it was an ancient Viking recipe for cleaning dried blood from weapons and armor...

Re:I think that the results are obvious

[tap]

Second is the higher spindle speeds

You can get 10k SATA drives now. There are 15k SCSI drives, but they have small capaciticies and are very expensive. There are a small number of applications that actually want to use them. For most uses, a 300 GB 7200 RPM drive is better than a 73 GB 10,000 RPM drive that costs four times as much.

Third is SATA only lets you have *one* drive. SCSI lets you have 15.

You're off in two ways here. One is that your ignoring the topology difference between SATA and SCSI. SATA uses a star, you have a controller with multiple ports, with one drive on each port. SCSI use a bus, with a single port controler and multiple drives on each port. The star topology is a lot nicer in many ways, one of the reasons twisted pair ethernet with hubs replaced the bus like coax ethernet. Secondly, you can't connect 15 drives to a single SCSI bus and run at ultra-160 or ultra-320 speed. You're only allowed to connect something like 4 drives to remain in spec at those speeds.

A 3ware 12 port SATA card and a three port U320 SCSI card with four drives on each port both support the same number of drives. Except the SATA card will probably be 1/3 the price, the SATA drives will be 1/10 the cost per GB, and have higher transfer rates.

SATA does have real command queuing. There are real hotswap SATA drive bays. It's true the cables can't be as long, but since you only need to connect once device per cable instead of 4 or more, it's usually easier to connect. And believe me, I know my way around a SCSI cable.

Re:I think that the results are obvious

[tap]

Many modern SATA drives are physically identical to the SCSI disks in terms of the disk components,

Maybe this way true 10 years ago, but it hasn't been the case for a long time. If you look at the physical parameters, SCSI and ATA/SATA drives are completely different. Different spindle speeds, different number of platters, different capacities, different noise levels, etc. The only drives I know of that are the same are western digital's Raptor 10k SATA drives, which are the same (including cost) as some of their SCSI drives.

Re:I think that the results are obvious

[Cramer]

... and the fact that SCSI drives are actually tested. I have never once, in 20 years, had a bad SCSI drive right out of the box. I've had numerous IDE drives bad straight from the factory -- one hadn't even been programmed.

SCSI does bet everything else out there on all but one point: price. And that's really the only thing many people ever look at. If you care about the data you're putting on the drive, you don't entrust it to cheap hardware.

If they are, in fact, the same servo hardware, why is it that the SCSI drives last for decades without fault but the IDE drives rarely last beyond 2 years? I have SCSI drives that are now slightly over 20 years old and they still work as well today as they did in '85 (altho' hideously slow by comparison.) The oldest, still functional, IDE drive I have is 4-5 years old. Most don't last beyond 2 years. (it varies from manufacturer to manufacturer, but very few last much longer than the warranty period.)

Re:I think that the results are obvious

[Anonymous Coward]

I don't know what kind of drives you're using, but I've had equal amounts of IDE and hot-swap SCSI drives go bad. Considering we have about 10 times more IDE than SCSI drives, SCSI seems to have a worse track record... especially for hot-swap drives that cost four to eight times as much per GB.

"Most don't last beyond 2 years"

We have a slew of older IDE machines that have been running 24/7 (except power outages) for at least 4 years (long before I started working there) -- and are still running. The hot-swap SCSI drives that failed were no older than 2 years.

Granted, we have only had a few non-hot-swap SCSI drives go bad, but we don't have many machines so configured.

Re:I think that the results are obvious

[ookaze]

You can get 10k SATA drives now. There are 15k SCSI drives, but they have small capaciticies and are very expensive. There are a small number of applications that actually want to use them. For most uses, a 300 GB 7200 RPM drive is better than a 73 GB 10,000 RPM drive that costs four times as much.

Nonsense !
In no way a 300 GB 7200 RPM drive is better than a 73 GB 10 000 RPM drive, except in one way : the ability to lose 300 GB worth of data in one go when the disk crashes.
I have 36 Go SCSI 15 000 RPM drives. I paid them 170 euros (that is what you call really expensive I guess). Spindle speed is not even the best performance factor, the access time is. These "low cost" SCSI drives mops the floor of any SATA disk you put against them in REAL situation, on a REAL filesystem.
I have a small number of situation for you where the SCSI drives have no equivalent : swap, file transfers, system partition. When you have a machine that swaps like crazy without you noticing (except for the sudden noise the disk make), you will understand the difference with a SATA (or IDE) drive where everything just freezes several seconds.

Secondly, you can't connect 15 drives to a single SCSI bus and run at ultra-160 or ultra-320 speed. You're only allowed to connect something like 4 drives to remain in spec at those speeds.

Huh ? Perhaps that is because you will saturate the SCSI bus at full speed before being able to add other disks ...

A 3ware 12 port SATA card and a three port U320 SCSI card with four drives on each port both support the same number of drives. Except the SATA card will probably be 1/3 the price, the SATA drives will be 1/10 the cost per GB, and have higher transfer rates.

And I bet anything that the transfer rate of the SCSI disks will be much better. And you have strange facts too.
Because here in France, the Adaptec SCSI RAID card is 400-450 euros, the 3ware card you talk about is not available, but a 8 port 3ware SATA card is already 600 euros !!! And we're not even at the performance level of the SCSI card or disks. Even with 10k RPM SCSI disks, we are at a higher performance level than SATA. And we are talking of less than 2 times the price in favor of the SATA drives here (290 VS 165 ) !! So stop saying nonsense please !

SATA does have real command queuing. There are real hotswap SATA drive bays. It's true the cables can't be as long, but since you only need to connect once device per cable instead of 4 or more, it's usually easier to connect. And believe me, I know my way around a SCSI cable.distributions.

I agree with you, except on the easier to connect part I don't understand. Anyway, I agree that SCSI is more expensive, but don't tell me SATA is higher perf, that is just a lie. We did not even get into the reliability realm (though I don't think my low cost SCSI U320 drives can last more than 5 years working 24/24 7/7, one of my low cost U2W disk died after 4,5 years of this regime) ...

Re:I think that the results are obvious

[Shinobi]

I'm running two U160 disks from 1999 on this computer, as well as a Raptor on a Sil 3114. The U160 disks beat the crap out of the Raptor in real situations, such as handling around 10k different text files, or working with a 45MB image, or a large video stream.

The only situations where the Raptor wins is in benchmark programs, and only if write-caching is enabled. Disable the write-cache, and the Raptors performance sinks like a stone.

Re:I think that the results are obvious

[Anonymous Coward]

"Tilgjengelig" means available

Re:I think that the results are obvious

[TheLink]

"I have 36 Go SCSI 15 000 RPM drives. I paid them 170 euros (that is what you call really expensive I guess). Spindle speed is not even the best performance factor, the access time is."

170 euros? A 200GB SATA drive is about 100euros. I wonder if you limit your partition used to the fastest 18% of the 200GB (36GB) drive whether you'd have comparable access time. You'd be seeking from a small number of cylinders very close to each other. The odds you need to seek will also be less given the track densities. Of course you'd still have to wait 4ms on average for sectors to spin around. Assuming if you use the full 200GB it takes 4.5ms more on average to move the heads and read data, perhaps if you use only 20% of the drive you'd only need 1 to 2ms more to move the heads and that brings it close to SCSI territory for about half the cost.

Get two SATA drives do the same thing and RAID1 them and you'd have faster access time AND you'd have redundancy, for about the same cost.

Is a SCSI drive that much faster once you do that?

In fact if you don't mind sacrificing write speeds (and storage space) for faster read speeds, you could write the data twice or more times per track. For example, you write the same data, 180 degrees apart on a track (or every 120 degrees if you are doing 3 x). That way you effectively have twice or 3x the RPM for reads (but your writes are slow). I wonder if anyone has done that in practice though ;).

Norwegian

[bcmm]

Anyone know an internet translator that supports Norwegian? Or even a Norwegian? It would be nice to have a translation so we don't have to sit around making uninformed comments about what we can't understand...

Oh, wait...

Re:Norwegian

[wfberg]

Anyone know an internet translator that supports Norwegian? Or even a Norwegian? It would be nice to have a translation so we don't have to sit around making uninformed comments about what we can't understand...

I think "Jeg vil rette en advarsel til alle dere som skal ut å handle kontrollere etter dette. Sjekk_nøye_om kontrolleren er støttet av kjernen! " speaks for itself.

Re:Norwegian

[Anonymous Coward]

try with swedish, it's almost the same.

Re:Norwegian

[biglig2]

Let me help. Apparently, software RAID is slower than hardware RAID, and Linux SCSI drivers are of variable quality, and also setting a PC on fire degrades its disk performance.

Re:Norwegian

[Anonymous Coward]

try with swedish, it's almost the same.

That's clearly an insult to most norwegians! The union with Sweeden ended in 1905, which is 100 years ago!

Re:Norwegian

[TeknoHog]

PS - What the heck is nettverkskort, exactly?

Network card.

Re:Norwegian

[Novus]

"nettverkskort" = "network card".

Re:Norwegian

[Guanix]

nettverkskort = network card

Re:Norwegian

[Novus]

And in case it doesn't, it means "I'd like to warn everyone who's going to go buy controllers after this. Check _carefully_ that the controller is supported by the kernel."

Re:Norwegian

[EvilMonkeySlayer]

nettverkskort = A badly limping half eaten gorilla.

Those crazy Norwegians!

Re:Norwegian

[Anonymous Coward]

nettverkskort = "cheese hamburger with bacon"

Re:Norwegian

[hobbicik]

'Nettverkskort' is self-explanatory.
Try to guess what is 'innebygde'!

Re:Norwegian

[freqmod]

innebygde is "built in" in plural inne=inside (from in[n]) bygde (from bygd) as in buildt

Re:Norwegian

[cnettel]

I want to caution everyone who is going to buy a controller now. Check carefully if the controller is supported by the kernel.

Re:Norwegian

[lawpoop]

That's nothing. In Finland, one of the most high-tech countries on the planet, they still refer to magazines and newspapers as 'leaves' (lehtiä)!

Re:Norwegian

[Novus]

Swedish is quite similar to Norwegian, but the differences in spelling will almost certainly prevent all common machine translation systems with support for Swedish but not Norwegian from producing a meaningful translation.

Re:Norwegian

[Anonymous Coward]

I'm norwegian, and I can tell you, no one here is insulted. The swedes are slightly bitter at us because of our oil, but we like them just fine.

Re:Norwegian

[Anonymous Coward]

nettverkskort

Skorts are those half-dress/half-skirt thingies that girls wear so you can't get any good upskirt shots. Within the context of this article, it makes perfect sense to me.

Re:Norwegian

[G3ckoG33k]

I have used Systran for automatic translation of half a million web-pages from English-Swedish-English. It is a piece of crap as of today, producing non-indsfolloble f***wqx glfttbberish. Put Norwegian into that system, yes, and it would be even worse. Yet, phonetically the high forms of Norwegian and Swedish are more similar than Cockney and Texan.

Re:Norwegian

[Anonymous Coward]

Nettverkskort? Hm... I really really wonder if 'network card' would be a correct translation.

Personally, Norwegian doesn't seem very funny to me. Maybe it's due to the fact me not native Engleesh speaker.

Heh

[FlyByPC]

My bonnie++ was used by Norwegians, To see how fast my RAID could be, My bonnie++ was used by Norwegians, ...but was bonnie++ written in C?

Re:Heh

[Anonymous Coward]

Apologies to FlyByPC:

My bonnie++ was used by Norwegians,
To see how fast my RAID could be,
My bonnie++ was used by Norwegians,
...but was bonnie++ written in C?

Re:Heh

[Anonymous Coward]

Zelo confirms it, Bonnie is dying.

Re:Heh

[Anonymous Coward]

Or as RMS wrote:

My Ronnie lies over the radio.
My Ronnie lies over TV.
My Ronnie lies over and over.
What has he done to my country?

Re:The graphs are not self-explanatory!

[Anonymous Coward]

I agree wholeheartedly. An utterly pointless article. Editorial standards on the decline once again.

M.O.D. . P.A.R.E.N.T. . U.P. !

[Anonymous Coward]

What he said.

Worst. Story. Ever.

Better Link

[XanC]

here

Re:Better Link

[notanatheist]

how is that a better link? It's still Norwegian! Who cares if it's all on one page. Who modded that guy up? So uncool. Mod the first english translation up. Sheesh.

Damn...

[broody]

Damn. I've been a geek too long. After all these years I know understand how my pointy haired boss feel when attempting to read a technical article.

Re:The graphs are not self-explanatory!

[bcmm]

It's completely in Norwegian, and a link to a Babelfish version isn't even included.

Babel doesn't do Norwegian.

Coral cache

[tyler_larson]

http://www.hwb.no.nyud.net:8090/artikkel/15307/5

The page with the pretty pictures.

Re:Cmdrtaco suger eselballer.

[Anonymous Coward]

bésame el culo nen

Save yourself the pageloads

[Anonymous Coward]

The graphs in question don't actually appear until page 5. So unless you're technically literate in the norwegian language, you may want to just skip ahead. Unfortunately, despite the OP's comment, I don't think the graphs alone are terribly helpful or lead to any meaningful conclusions about the state of specific drivers.

Translation

[FlyByPC]

http://www.translation-guide.com/free_online_trans lators.php?from=Norwegian&to=English
Not that it's really useful. It's a *little* more readable than the original. I think.

Re:Translation

[Anonymous Coward]

Tried that, the translation looks more like yoda-speak than english

Re:Translation

[notanatheist]

I'm sure it would be helpful if translation service didn't get SLASHDOTTED!

Re:The graphs are not self-explanatory!

[Anonymous Coward]

so the link isn't included

Time to troll

[Afrosheen]

Ok, who the fuck allowed this submission to go through? A whole 2% of Slashdot readership will probably be able to read this, the rest of us are left in the dark. Are longer bars better, or worse? WTFOMGBBQ?!

Re:Time to troll

[Anonymous Coward]

2%? you're so optimist

Re:Time to troll

[Novus]

"Større er bedre" means "bigger is better", while "mindre er bedre" means "smaller is better". That should help a lot with the bar graphs.

Re:Time to troll

[Anonymous Coward]

If we believe past slashdot polls, 30 % of the readership is European. Since all scandinavians and a good proportion of Finns should be able to understand Norwegian, it isn't that far-fetched that approximately 2 % of the slashdot readership would be able to understand the article.

Re:Time to troll

[harkabeeparolyn]

WTFOMGBBQ?!

?

"What the fuck, O my God, BARBECUE?!"

Re:Time to troll

[Afrosheen]

Yeah it's something poking fun at chatters and it's been around for awhile. Google is full of examples. I was going to end it with a WTF but went all the way.

Re:Time to troll

[BiggerIsBetter]

Hmm. StørreErBedre. Time for a new nick?

Kernel 2.6.8?

[erasmix]

I dont speak norweian, but I noticed the benchmark seemed to be done on Kernel 2.6.8. But the latest kernel release is 2.6.11

Re:Kernel 2.6.8?

[Anonymous Coward]

i wonder if its like dates where 4.5.98 could be either april 5 or may 4 :P

Re:Kernel 2.6.8?

[rubycodez]

a couple major "commercial" distros use the 2.6.8 kernel

Re:Kernel 2.6.8?

[ak_hepcat]

Says in the article that for driver compatibility, he had to use a specific red-hat kernel.

That's adaptec for you. sodding bastards. I unfortunately bought a 1210SA from them and I've regretted it ever since.

Why, yes, it does come with an outdated kernel driver, and no way to recompile the driver for new kernels because they've kept it purposely closed source.

Probably because most of the raid work is being done in software anyway, and they just don't want people to find out how badly written their buggy drivers really are.

Me? Bitter? nah. I *like* my linux box stuck in the fricken stone age.

Re:Kernel 2.6.8?

[daclink]

Is it not supported by the aacraid driver in the kernel? I have an adaptec sata raid 2810sa card and it works fine under gentoo. DaClink

Re:Kernel 2.6.8?

[ak_hepcat]

Hmm. Depends on your level of supported. For my card, i can see individual drives, but the Mirror isn't seen. For me, that means broken.

I haven't tried any of the .12-pre kernels yet, though, so perhaps... I doubt it, but perhaps.

I'd be interested in knowing a bit more info about your setup tho -- message me privately.

My grasp of the article...

[Andreas(R)]

Jeg skal ikke gå så langt som å si at man burde satse på verken SATA, billige kontrollere eller software-RAID.

In english; I will not go as far as to recommend SATA, cheap controllers or software-RAID.

Seriously, is this frontpage news on Slashdot? I'm a native speaker, and the article did not impres s me much. In fact, there is nothing newsworthy about the article, and the author admits it in the conclusion. Not very insightful, the article is crearly written by an amateur. In fact, in my opinion, the only reason this was submitted to Slashdot, is because hwb.no is a new site, which is trying desperately to get visitors.

/cynical

Re:My grasp of the article...

[Anonymous Coward]

Akkurat det jeg tenkte når jeg så den. Har en følelse av at noen hos hwb tenkte "Hmm.. hvordan imponerer man komplett med masse pageviews så de gir oss masse reklamepenger? Slashdot!"

Men det virker som de virkelig prøver å lage litt lengre og bedre artikler enn hw.no har, så jeg er forsiktig optimistisk.

By the way slashdot, I'm Norwegian, and I was going to try to translate the article, but it's *way* too long. Sorry.

Re:My grasp of the article...

[caluml]

Jeg skal ikke gå så langt
I shall not go so far

It's almost readable.

Re:My grasp of the article...

[Anonymous Coward]

Bra med skepsis. Likevel, det var jeg som la den inn, og jeg er ikke på noen måte assosiert med HW eller HWB.

And that concludes todays course in Norwegian.

(In the interest of disclosure: I submitted the article but am not in any way associated with neither HW nor HWB. A closer view shows that sw RAID is in some cases as fast as hw RAID but the pattern is sufficiently interesting that I believ this set of benchmarks shows some kernel drivers would benefit from some tuning.)

Re:My grasp of the article...

[laa]

Then again:
"Billigkontrollere" har i testen vist seg som verdige diskkontrollere, samt at software-RAID faktisk yter meget bra.

I haven't read Norweigan since high school, but if I'm not mistaken, it means that "cheap controllers are worthy disk controllers and SW-RAID actually does a good job".

The author says he was surprised himself, but at least I can't see why, I think the conclusions are a bit messy. You are right, the article isn't very informative.

yeah . . . .

[failure-man]

The text is in Norwegian but the numerous graphs are self explanatory.

Maybe if you read Norwegian or are, y'know, smart it makes sense. Me however, I can't tell what the hell is going on. Will somebody please think of the idiots?!

Re:The graphs are not self-explanatory!

[Anonymous Coward]

Babel doesn't do Norwegian.

Sure it does, the problem is you need someone to read it out loud to be able to understand it.

Trollage i guess

[testednegative]

norse--
If only I had a "Learn Norse in 30days" book to advertise about now, i'd be rich

go Tiger!

[Crimson Dragon]

And, for the first time ever, I have used my translator in Dashboard!!! :D

On a serious note, it seems to me to be readily apparent the best RAID setup on the face of the Earth can be absolutely evicerated by a poor driver. There seems to be no profundity in saying that, and these benchmarks seem to be a result of this.

I am open to correction on this one, of course.

Re:OT: SlashNOT

[grazzy]

You reload slashdot to much dude.

Uninformed comments...

[Man In Black]

It would be nice to have a translation so we don't have to sit around making uninformed comments about what we can't understand...

Somehow, I don't think a translation would keep them away.

Surprises?

[EvilMonkeySlayer]

How is it surprising that software RAID is not as fast as hardware RAID?

Software RAID has to share the systems bus, that means transferring the data from n number of drives across the PCI (whatever) bus whilst with hardware RAID it's all kept to the RAID card itself.

This is not surprising in the slightest, the only people who use software RAID are the ones doing it on the cheap.

Speaking from experience, i've used both hardware and software RAID. I don't think there is a single person here who doesn't understand the disadvantages of software RAID.

Re:Surprises?

[Anonymous Coward]

The suprise is that some of the software raid setups are significantly faster than some of the adaptec hardware raid setups.

It's well known that some "hardware" raids are nothing but regular controllers, with all the raid functionality contained in the closed source and usually huge binary drivers. They are worse than pure Linux sw raid.

Re:Surprises?

[Jeff DeMaagd]

I think the real surprise for me is that SW RAID is 95% as fast as HW RAID without the pricey board, not that SW RAID is slower.

Also, another surprise is that a SATA RAID (speed) performs about as well as a SCSI RAID. Whether SATA drives are as reliable is a different matter, but with the cost savings, it is easier to have more spare drives on hand.

From a system bus bandwidth perspective, it would seem that the chief difference between HW and SW RAID would be that SW RAID requires some more housekeeping bits, the biggest one being the data from the parity drive goes over the system bus for SW, but it stays local to the RAID controller for HW.

From a CPU perspective, for SW, the CPU would have to compute the XORs rather than offloading them to the dedicated hardware, which are compute cycles and pages that could be done for other tasks in a HW setup.

For me, the speed difference is kind of moot though. If I want RAID, it would be for the redundancy and spanning multiple drives, not speed. Also, I have systems with 64/66 PCI and a system with PCI-X, so that bus isn't an issue.

Re:Surprises?

[Homology]

Speaking from experience, i've used both hardware and software RAID. I don't think there is a single person here who doesn't understand the disadvantages of software RAID.

Erh, I'm pretty sure that the majority here is only using IDE/SATA and believes that RAID0 is a huge speedup on a gaming machine.

Re:Surprises?

[EvilMonkeySlayer]

I was more referencing the disadvantage of RAID-5 with lots of drives which tend to flood the system bus.

If it's just a couple of drives in RAID-0 you don't have to worry about any kind of performance hit.

Re:Surprises?

[toddbu]

"RAID 5 you say?", he says, attempting his best Yoda immitation. "Disk space still so expensive that RAID 5 you must still use? Beware of the dark side my young padawan."

Am I wrong in thinking that unless your concerned about the physical disk space or power requirements or few extra $ to go with RAID 1, that there's no longer much benefit to RAID 5? I hear comments like "I can get data off of a striped set faster than a mirror", but as you mention there's a lot of processing that needs to be done to put all the data back together and I think that that tends to offset the gains in reading from multiple drives at once. Given the fact that the RAID 1 is simpler than RAID 5 and therefore by definition more stable, is there any reason any more to use RAID 5, at least for small (<1 TB) installations?

Re:Surprises?

[Crispy Critters]

"Also, another surprise is that a SATA RAID (speed) performs about as well as a SCSI RAID."

That is a surprise. As far as I can tell (not being able to read the article) he is comparing 7200 rpm SATA drives to 10,000 rpm SCSI drives (see page Testoppsett in TFA). Rule of thumb people use is that read/write speeds are proportional to the disk speed, so that is an advantage for the SCSI setups. He could at least compare 10k rpm SATA to 10k rpm SCSI "Harddisker."

"If I want RAID, it would be for the redundancy and spanning multiple drives, not speed."

For spanning multiple drives I think you want LVM.

Re:Surprises?

[rale, the]

"If I want RAID, it would be for the redundancy and spanning multiple drives, not speed."

For spanning multiple drives I think you want LVM.


That wont do much for redundancy, tho. With LVM spanning multiple disks, the loss of any single disk means the loss of all data on all disks. If you value your data at all, sw raid-5 is a good way to make sure it doesnt all get lost from a single failure. Oh, and for anyone looking to set up a linux software raid, evms is the tool to use - its much easier then the old mess of config files and command line tools.

Re:Surprises?

[Velox_SwiftFox]

RAID 1? Foo. Get the best of both worlds and go RAID 10. Reliability is optimal, even with 4 drives, a 2 out of 3 chance that a second failure will cause no data loss.

But when drives are cheap, the more the better - I tested a RAID 10 array of 12 400G Seagate SATA drives on a 3ware 9500S controller, and created one 48G partition to run SuSE 9.2 from. With 12 actuators and only the fastest 2% of the platters to move them over, I/O was gratifyingly fast.

Unfortunately, I was rather put off by the 3ware release notes; requiring me to remove half of the 4GB of RAM in the SuSE system, and:

Linux (all versions) and usage of Ctrl+c
The use of Ctrl+c to abort operations is not recommended while having I/O to the
3ware controller. The following are know incompatibilities:
1. Create a RAID-0 unit
2. Migrate the unit to another RAID-0 using CLI
3. Format and perform I/O on the array
4. Delete the unit using CLI
5. Hitting Ctrl+c causes a kernel panic

"Following are know incompatibilities"? I just can't abide spelling errors in technical documents.

Re:Surprises?

[Crispy Critters]

"With LVM spanning multiple disks, the loss of any single disk means the loss of all data on all disks."

According to the docs, that appears to be true only if you are using striping, in effect making the LVM a RAID-0 system. I think it is true that someone who wants to span multiple drives but doesn't need redundancy and doesn't care about the speed increase from striping would be better off using LVM than RAID, because LVM is a lot more versatile. So change the sentence to, "If I want RAID, it would be for the redundancy."

Re:Surprises?

[51mon]

He does indeed very odd.

The Hitachi SCSI drive gives you 73GB for 132.16GBP at 10,000 RPM (180p/GB) and 8MB cache.

The SATA150 Maxtor gives you 250GB for 95.23GBP (38p/GB) at 7,200 RPM with 16MB cache built in. So three times the space, and the software RAID performance looked pretty comparable when using the Adaptec controller.

The Maxtor Diamond Max Plus 9 at 32.95GBP is probably the suitable SATA150 drive to put up against an 80GB SCSI drive. And the saving of 100GBP probably won't make a huge impact in performance (2MB cache I believe), if it does, buy 4 of them to get 8MB of cache back ;)

You pay a lot for the extra cache in the drives, but the RAID controller if it is good will switch this off for write-back and use its own for reliability reasons. I wish I understood enough Norweigan to see what he'd done with write-back caching.

Even if you have 32MB or 64MB of cache on each drive, it is still small (and expensive) compared to both the disk drive size, and the memory used for caching on modern servers, as a result the disks own cache will have a fickle impact on performance, possibly improving some unrealistic benchmarks better than it will actually help in real life. My guess is most people are better off saving the money and buying RAM, which is probably cheaper per MB, and has a more general utility beyond just the I/O intensive operations.

Perhaps what we need is a competition to design and build servers for particular application purposes; email server, database server, and see what people can do for a specific amount of money.

Prices quoted are list at DABS.COM excluding VAT.

Re:Surprises?

[Gumber]

I've just built my first linux raid server for my home network (after using hardware and software RAID on WinNT/2k for years). I'm using mirroring for now to keep initial costs low, but as time goes on, I'll pick up cheep disksand move to raid 5.

I figured that at some point, I would go with a hardware RAID card, but it occured to me that one of the advantages of software RAID is that you can grow by buying the lowest-cost reliable hardware controllers and reconfigure things without regard for the vendors raid implementation.

Also, with Hardware RAID, if the controller dies, I'll need to get a controller from the same vendor if I want to drop in a replacement without tearing down and restoring from a backup.

Re:Surprises?

[Anonymous Coward]

Ahh.. but with RAID 5 you get redundancy spanning multiple drives :)

Re:Surprises?

[Jeff DeMaagd]

Uh, yeah, as others said, I would want to be doing RAID 5, not just stupid spanning. RAID-5 gives both redundancy and spanning multiple hard discs. RAID 0 is just spanning, which has no redundancy, which is why the number is zero, because it fails the "R" (redundant) part of RAID. RAID 1 is just mirroring, no spanning. RAID 0+1 mirrors and spans, but a 4 disk RAID-5 array has 50% more space than a 4 disk RAID 0+1

I don't know what the problem was that made SATA comparable to SCSI. There may very well be a limitation in the computer chosen. Also, the increased areal density of a large capacity 7200RPM drive allows the drive bandwidth to be competitive with a potentially lower density 10kRPM. The high RPM drives mostly have a benefit in seek times because the discs are about a third smaller in diameter

Re:Surprises?

[Cramer]

Except, one cannot direct boot to an LVM volume. It takes some userland toolage to discover and activate all the volumes.

The only way around this is with an initrd, which I think is an ugly, hard-to-manage, hack. The other option is to not use LVM for the rootfs.

LVM is a handy thing. But it's NOT RAID; don't treat it like it is. And it's infinitely more complicated than RAID, managed by mdadm.

Re:Surprises?

[dtfinch]

The software RAIDs kicked ass in all of the random access benchmarks.

Re:Surprises?

[pogson]

EvilMonkeySlayer wrote:

the only people who use software RAID are the ones doing it on the cheap.

Not necessarily. A decent PC is 64 bit 3000MHz with 1gb RAM and a few hundred gigabytes of storage. What more does anyone need? This isn't cheap, it is practical. If you want redundancy use software RAID 1 or network a bunch of boxes. If you need more speed, use software RAID 0 or 0+1. Who cares if a few CPU cycles go in the process? I have a system on my desk that serves eight clients as thin terminals. Nine of us use the thing simultaneously and it never bogs down. I use RAID 1 so I have redundancy and two users do not have to wait, but i have so much RAM, the drive light rarely comes on. This system cost $1000. It has 400 gB storage.What advantage would there be to hardware RAID?

hdparm -tT /dev/md3
/dev/md3:
Timing cached reads: 1728 MB in 2.00 seconds = 863.27 MB/sec
Timing buffered disk reads: 174 MB in 3.03 seconds = 57.47 MB/sec

hdparm -tT /dev/hda2

/dev/hda2:
Timing cached reads: 1716 MB in 2.00 seconds = 857.27 MB/sec
Timing buffered disk reads: 186 MB in 3.04 seconds = 61.27 MB/sec

Software raid has a role. It is the right thing to do in many cases.

Re:Surprises?

[241comp]

Of course, software RAID is only 95% as fast as hardware raid if you're doing nothing with the data. If your primary CPU is 100% in use by a high priority task, you may find that the HD speed suffers under SW RAID.

Re:Surprises?

[drsmithy]

This is not surprising in the slightest, the only people who use software RAID are the ones doing it on the cheap.

Not true at all - there are flexibility and robustness advantages to software RAID. We often use a mix of hardware and software RAID on single systems to get the best of both worlds.

Speaking from experience, i've used both hardware and software RAID. I don't think there is a single person here who doesn't understand the disadvantages of software RAID.

Funny, from the comments I read in the typical RAID thread on /., I'd say there's only a tiny minority of people here who have even the vaguest notion of not only the inherent advantages and disadvantages of the various RAID configurations, but how to actually go about comparing those relative merits to individual situations.

*Both* software and hardware RAID have their place in _all_ environments.

Re:Surprises?

[Fweeky]

"RAID 0+1 mirrors and spans, but a 4 disk RAID-5 array has 50% more space than a 4 disk RAID 0+1"

On the other hand, if a disk fails on your RAID-5 array, you're liable to see a significant drop in performance, since it starts having to rebuild data from parity spread across all the disks. Have a disk fail in your RAID-10 array and the performance drop is going to be much less significant since it's just a single degraded RAID-1. The RAID-10* array also has a significantly better chance of surviving if two disks die.

This is critical if you're going for speed and redundancy (say, on a busy database server) over redundancy and space (like your network file server).

* That is, 2 mirrors, and a stripe across them; the other way around is Very Bad[tm].

Re:Surprises?

[Jeff DeMaagd]

I really don't see a big issue with performance that way. I don't expect to run a RAID-5 with a failed drive for a long period of time. Given how rarely drives die, I don't expect that two of them would go down at the same time. I would have a backup for that circumstance anyway, as RAID is no substitute for a backup.

Re:Surprises?

[Crispy Critters]

"I really don't see a big issue with performance that way."

Your message shows that you are assessing the various possibilities only in terms of your specific needs. Every setup has its own requirements and its own performance criteria. There are situtations where having the disk access speed drop by 50% could cost someone from 10's of thousands of dollars to order of a million dollars.

Then there are people who are trying to maximize their value in disk space per dollar, and the difference between N/2 (RAID 1, RAID 10) and N-1 (RAID 5) disk space is significant.

Similarly, what is the risk of two dead drives? Higher than you think, because failures of drives are not uncorrelated, often corresponding with power supply issues or rebooting. Higher also if you have a system with ten drives in it.

How useful and valuable is the backup? How valuable is whatever went on the drives since the last backup? If you have the work of 250 people being stored, losing one day's work for everyone is like wiping out a year's work from one person, so that's $50-100k. The value of redundancy is really going to vary, depending on the purpose of the system.

Guide to graphs

[johnjaydk]

The caption: "Større er bedre" means "Bigger is better" (yeah right).

The caption: "Mindre er bedre" means "Smaller is better" (even more yeah right)

Whoever approved this article in "non-english" should be trambled to death by a mob of angy penguins.

Re:Guide to graphs

[wisdom_brewing]

trambled by angy penguins... i can only hope that that was intentional

Norwegian

[magarity]

From TFA: To innebygde gigabit-nettverkskort

That is just the coolest; I am hereby recommending everyone refer to networking as 'nettverkskort'. It might be cold in Norway, but they have some awesome sounding linguistic constructions!

PS - What the heck is nettverkskort, exactly? 'Networking'? 'Network Adapter'? Heck, I don't know what it is; I just know I like it.

Regarding the Kernel Used

[Anonymous Butthead]

This is not a very good review, they have used kernel 2.6.8, 2.6.11 has many fixes upon previous releases in regards to RAID and md (software raid) drivers.

Lets get a review that uses 2.6.11, then lets see where we are.

Re:Regarding the Kernel Used

[Dan Ost]

I don't know this to be true, but since the commercial vendors backported
patches for the 2.4 series, I would suspect that they would backport patches
for the 2.6 series. If that is indeed the case, then fixes in the 2.6.11
kernel have probably been incorporated into the 2.6.8 kernels used by
Red Hat, SUSE, etc.

So, the question is: did they benchmark using the vanilla 2.6.8 kernel
or a heavily patched 2.6.8 kernel from one of the commercial distros?

Re:Regarding the Kernel Used

[cibus]

[QUOTE]
Testen ble kjørt på Ubuntu Warty. Etter at testen ble påbegynt har Hoary kommet. Versjonen skal ikke ha noen innvirkning på resultatene. Både LSI- og Adaptec-kontrollerene ble kjørt med siste versjon av driverne. For å se på RAID-egenskapene i kontrollerne vi har testet, så har vi også kjørt alle kontrollere med kjernens eget software-RAID.

Kjernen benyttet under testingen var 2.6.8.1-5-amd64-k8-smp. Det ble brukt x86_64-versjonen av Warty.
[/QUOTE]

Translation:
The test was ran on Ubuntu Warty. After the test-project was started Horay was released. The Ubuntu version should not have any effect on the results. Both the LSI- and Adaptech-controllers was used with the latest versions of the drivers. To measure the capabilities of the controllers we have also tested all the controllers with the Linux kernels internal software RAID.

The kernel used for this test was 2.6.8.1-5-amd64-k8-smp. The x86_64 version of Warty was used.

Bad trolling.

[Anonymous Coward]

C'mon, do it right!

If this was "Slashdot. News that USian nerds can understand." it would be EMPTY!

Re:We tried working with Linux RAID

[Anonymous Coward]

Nice troll attempt, but this was done yesterday.

HTH. HAND. STFU

Re:OT: SlashNOT

[Anonymous Coward]

That's right, but what should an ugly fat unemployed useless fag like me do else?

Re:We tried working with Linux RAID

[Anonymous Coward]

Actually, it's been done twice today, too. Please try to keep up.

Re:The graphs are not self-explanatory!

[Anonymous Coward]

What exactly didn't you understand? The "st0rre er bedre" or the "mindre er bedre"? I wanna know, maybe I can help...

My take...

[Anonymous Coward]

It seems to me that hardware SCSI RAID stomps the crap out of any other RAID solution! This is hardly a surprise to me. Though I suspect it does come as a surprise to those that always argue that SCSI offers no advantage over IDE.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Try this experiment

[hey!]

Take the following, randomly selected paragraph from the article:

LSI MegaRAID SCSI 320-1.
Low-profile kontroller. Har både intern og ekstern kontakt for tilkobling av SCSI-kabel. Bruker også en Intel GC80302- prosessor. Støter RAID 0, 1, 5, 10 og 50.

Now, give it to a non-Norwegian speaking geek, and a non-Geek Norwegian speaker.

Who do you think will have more luck making heads or tails of it?

Re:Try this experiment

[nickptar]

Low-profile controller, connects to internal or external SCSI devices, has an Intel GC80302 processor, and supports RAID 0, 1, 5, 10, and 50.

Right?

(I'm a non-Norwegian-speaking geek, of course.)

Re:Try this experiment

[Homology]

Indeed, that is correct (I speak Norwegian natively).

For a supposed geek site, many /.'ers shows an alarming ineptitude to find/verify information. For the controller in question, he could quite simply use Google, or go to the LSI home page. The tools used are standard, the controllers/hardware are standard, which Linux kernels used should be apparent. Understanding the conclusions, of course, means understanding Norwegian, but he should be able to interpret the output himselves.

Re:Try this experiment

[caluml]

non-Norwegian speaking geek

I'm a non-Norwegian, speaking geek - is that close enough? :)

Re:Try this experiment

[leakingmemory]

Actually, there is a typo in the article:

Støter RAID 0, 1, 5, 10 og 50.
should be:
Støtter RAID 0, 1, 5, 10 og 50.

Re:Try this experiment

[G3ckoG33k]

Where "Støtter" means "stöder" (or "stödjer") or "supports" in more widespread languages. ;)

Re:Try this experiment

[hey!]

Which only shows that we should speak in Lisp. The parens would completely disambiguate the negation operation.

Re:We tried working with Linux RAID

[Anonymous Coward]

Sincere apologies kind troll, I shall endeavor to keep abreast of trolling activities in future.

Some useful phrases:

[Ruie]

• storre er bedre - larger is better
• mindre er bedre - smaller is better
• Neste side - next page

Post more guesses below

Tweaking in Norwegian

[barfy]

involves those little hex drivers, and of course there is always one nut left over....

What are we talking about?

Re:Tweaking in Norwegian

[Bazzalisk]

I think you are confusing Sweden with Norway, which is generaly considered a bad move. Like confusing America and Canada, or New-Zealand and Australia.

Re: Why Does SlashDot Probe My System's Ports?

[Anonymous Coward]

More info here:
http://slashdot.org/comments.pl?sid=108060&cid=918 9489

No they didn't!

[Anonymous Coward]

They knew that 3ware was total crap and therefore, there was no point in including 3ware in the test. Including 3ware in tests against LSI and Adaptec would have been like including Yugo in a Ferrari vs. Lamborghini shootout.

In short 3ware sucks!

Re:No they didn't!

[Anonymous Coward]

Troll. Flamebait. Idiot. Take your pick.

Re:No they didn't!

[Anonymous Coward]

Tool, shill, zealot. Take your pick.

3ware sucks ass. The grand parent post was correct.

Re:Can't read the article

[Anonymous Coward]

i understand you are from USA, this is why you say something like "nobody speaks".

If you cannot speak it, it doesn't mean it is not used

Translation of their conclusion:

[Nichotin]

Ok, here is a rough translation:

I have wanted to test some real SATA controllers against SCSI controllers for some time now, to see how good SATA has become. I once thought that cheap controllers like Sil 3114 is cheap crap that manufacturers put on their boards simply to provide SATA-support, and that software RAID was a cheap, but insufficient solution, since I have followed the principle that hardware does the job best. "A more expensive controller, means more hardware", was my initial guess, but it seems that even the cheap controllers are worthy. Software RAID also performs very well. SATA is no longer some gag for disk systems that are supposed to perform well, and many myths have been dispelled by my test.

I will not go as far as to say that you shall place your bet on cheap controllers or software RAID. The reason is simple, in a expensive controller, there is much more functionality, that a cheap controller can just dream about. Functionality like hot-spare drives and hot-swap, just to mention some. I do not want to recommend SATA over SCSI in a while either. The lifespan of a SCSI drive is in most casese many times as long as a vanilla SATA-disk. When you choose a solution, it should last. If you have machines that has a big fat controller, RAID50, then SATA might be something for you. If you have a machine that needs redundancy on the internal drives, but where changing controllers, or even buying them in the first place has been in the way, then software RAID might be the solution for you.

I shall be careful to mock the LSI controller, as I think there might be a problem with the way the test machine talks to it. I think the new Megaraid driver in the kernel might be the problem. Either it needs to mature, or it is simply that it does not like 64-bit Linux. I have not tampered too much with the default settings, but it runs superparanoid verification algorithms when it sends and recieves data. I have not fleshed the BIOS on any of the controllers.

Adaptecs controllers do very well. Everything was not perfect with them, and the aacraid driver in the kernel was too old for both of the controllers. From their website, I found something that looked like source code (Adaptec seems to rely on 100% RPM based distros), and I could bouild my own module. After that, no problem. A little minus is that the aacraid does not report how long the controller has gotten in building the array after you have set up a RAID. By looking at the SCSI-BIOS after some hours, I got to verify that the array was built.

I want to warn everyone that is going to buy a controller. Carefully check that the controller is supported in the kernel! I use Google to check for references to the card on mailing lists, but that does not help much when you have Debian, and all that exist is binary RedHat drivers.

Now, run to your console and test your disk system. This test does only give you indications on what to choose. I allow myself to give you one final advice: Run tests for yourself.

Parent Post Resulted in These Scans:

[Anonymous Coward]

port scans that occurred while posting the parent (I've X'ed out my IP address):

2005/05/22,12:30:06 -5:00 GMT,66.35.250.150:33752,XXX.XXX.XXX.XXX:80,TCP
20 05/05/22,12:30:08 -5:00 GMT,66.35.250.150:33755,XXX.XXX.XXX.XXX:444,TCP
2 005/05/22,12:30:10 -5:00 GMT,66.35.250.150:33756,XXX.XXX.XXX.XXX:1080,TCP
2005/05/22,12:30:12 -5:00 GMT,66.35.250.150:33757,XXX.XXX.XXX.XXX:3127,TCP
2005/05/22,12:30:14 -5:00 GMT,66.35.250.150:33760,XXX.XXX.XXX.XXX:3128,TCP
2005/05/22,12:30:16 -5:00 GMT,66.35.250.150:33761,XXX.XXX.XXX.XXX:6588,TCP
2005/05/22,12:30:18 -5:00 GMT,66.35.250.150:33763,XXX.XXX.XXX.XXX:8000,TCP
2005/05/22,12:30:20 -5:00 GMT,66.35.250.150:33764,XXX.XXX.XXX.XXX:8080,TCP
2005/05/22,12:30:22 -5:00 GMT,66.35.250.150:33765,XXX.XXX.XXX.XXX:81,TCP
20 05/05/22,12:30:24 -5:00 GMT,66.35.250.150:33766,XXX.XXX.XXX.XXX:1026,TCP
2005/05/22,12:30:26 -5:00 GMT,66.35.250.150:33767,XXX.XXX.XXX.XXX:3124,TCP
2005/05/22,12:30:28 -5:00 GMT,66.35.250.150:33769,XXX.XXX.XXX.XXX:3382,TCP
2005/05/22,12:30:30 -5:00 GMT,66.35.250.150:33771,XXX.XXX.XXX.XXX:7032,TCP
2005/05/22,12:30:32 -5:00 GMT,66.35.250.150:33772,XXX.XXX.XXX.XXX:8002,TCP
2005/05/22,12:30:34 -5:00 GMT,66.35.250.150:33774,XXX.XXX.XXX.XXX:8090,TCP

This has occurred for at least a year, every time I post as AC.

Re:OT: SlashNOT

[grazzy]

write comments as a anonmyous coward?

Interesting

[Anonymous Coward]

It appears to be a test for a proxy. perhaps they are trying to prevent posts form open proxies. But, if that is what they are doing, it is completely ineffective. People post from open proxies all day long with impunity.

Chalk it up to a stupid Tacoian idea that can't hurt you.

Re:OT: SlashNOT

[Anonymous Coward]

;) I think it can be said that he is FLAMMING

Re: Why Does SlashDot Probe My System's Ports?

[pomac]

Looks like a proxy scan.

Ie, noone is allowed to post from a open proxy.

Sometimes SW RAID is much faster than HW RAID

[Anonymous Coward]

A few years back I was responsible for benchmarking potential RAID solutions for a major computer company. We investigated both software based and hardware based solutions.

The conclusion we reached; software RAID gave greatly superior performance than the hardware RAID solutions available at the time, but the hardware RAID solutions had better feature sets and usability.

The superiority in performance that the software raid solutions showed was due to a quirk in what was then state-of-the-art in RAID and systems design.

Most RAID controllers at that time contained embedded Intel i960 processors running at around 100 MHz, and had caches that topped out in the 128 MB range. Meanwhile, systems contained 2-4 CPUs in the 1.2 GHz range, and 2-8 GB of memory. There was simply no way that the embedded processor and cache on the RAID card could manipulate the data as quickly as the primary system resources could, and the benchmarks showed it.

The "exception" to this performance was when RAID-5 was used. Because RAID-5 requires computational resources above and beyond simply moving data back and forth in order to calculate parity, the host-based RAID solutions couldn't always keep up.

It was the fact that RAID-5 required additional computational resources that led fairly directly to the "ROMB" (RAID on motherboard) solutions that some vendors today. The ROMB chip is often nothing more than an XOR engine, to accelerate parity calculations.

The major, major, shortcoming we found with software RAID solutions was that they did not work with our customer's software, if that software ran outside of an operating system that had drivers for the solution. With hardware RAID, the physical disks were completely abstracted away, and you could run in any possible environment and still be able to read/write from your RAID volumes.

All of the above commentary about hardware vs. software performance is meant to apply to a specific point in time. I wouldn't try to extrapolate those results to current technology without rerunning the experiments today.

Re:Sometimes SW RAID is much faster than HW RAID

[drsmithy]

The "exception" to this performance was when RAID-5 was used. Because RAID-5 requires computational resources above and beyond simply moving data back and forth in order to calculate parity, the host-based RAID solutions couldn't always keep up.

RAID5 isn't slow because of the "computation overhead", it's slow because of the additional disk seeking. Even a paltry 300Mhz P2 has checksumming speeds near a gigabyte a second.

Hardware RAID5 may have outperformed software RAID5 in your comparison, but unless your systems were constantly pegged at 100% CPU usage, it didn't have anything to do with "computational overhead".

Nøth1n6 tø s33 h3r3 3xc3t å) @nd

[Anonymous Coward]

üb3r 3l1t3 t@Lk....

å) -- caucasian with a black eye!

The benchmark should be readable even if you don't understand the content. The magic of numbers... universality.

Re:Can't read the article

[Anonymous Coward]

There are 4 million speakers of the Norwegian language worldwide. The population of the earth is about 6.5 billion people. For all intents and purpouses, that *is* a language nobody speaks you ignorant, smelly euro-trash.

no need to guess

[Erris]

Why guess when translation is available. If you don't parle Google Toolbar Translation, you can go here: http://www.translation-guide.com/free_online_trans lators.php?from=Norwegian&to=English

While not perfect it backs your mindre as small.

If that's not good enough try deductive reasoning, such as:

• K/sec is good, therefore "storre et bedre" is something like "bigger is better".
• "mindre" is not "storre" and must mean something different. "CPU last" is most likely CPU load, read all about it here: bonnie docs.

What's clear to me is that ordinary SCSI kicks SATA ass, software or hardware. If you want to be cheap, buy used equipment from ebay or pricewatch. Works for me.

Re: Why Does SlashDot Probe My System's Ports?

[Bert64]

Looks like its checking for open proxies, kinda makes sense to do that especially if your posting as AC, as trolls will often flood open forums like slashdot with garbage posts via proxies.

Page 1 translated, paragraph by paragraph:

[jzono1]

What should one look for in regards to controllers & setups in a new computer? Do one need a controller in the 200-500$ range to ensure uptime & reliability without sacrificing raw power? Is SCSI still the only viable option for high performance computers? In this test we will look at some of the options and the results that they give when it comes to storage.

Most motherboards aviablee today have somo sort of S-ATA controller builtin. This is true both for low-end desktops & expansive servers.One of the Most common controllers seen today is the Silicon Image 3114, which is a four port S-ATA controller with stated raid abilities.

In almost everyone of these cheap S-ATA controllers they advertize with RAID. This is actually only half the truth. These controllers can, like mostly every other controller, drive Raid arrays, but most of the RAID happens in software, in windwos this means driver-level RAID. Rumors say that one should look for RAID5 supporht to get hardware raid. If that is true is uncertain, since GNU/Linux does this in software.

( Is performance = size * Price? )

I denne testen så skal vi se på noen ulike typer kontrollere. Den nevnte SiI 3114 er en av disse billige kontrollerene uten ekte hardware-RAID. Hvordan yter denne mot et mye dyrere SCSI-oppsett? Er SATA blitt så bra at dyre SCSI- løsninger bare er for spesielle maskiner?

Big thanks to Nextron for borrowing the test rig, multiple controllers and other equipment for this test. Otherwise, thanks to MPX for borrowing the adaptec controllers.


I may do the other pages too... Too boored now...

The results are obvious

[Anonymous Coward]

Yes.

First off, most SATA controllers are NOT hardware RAID although they support hardware raid options.

This is called BIOS raid and esseciantly uses software drivers in a similar fasion that Winmodems or software modems use software in their drivers to emulate hardware.

Dedicated hardware RAID devices are much more expensive, up in the hundreds of dollars for the controller. These devices use a embedded style cpu running around 200-400mhz that is specially designed for doing work like this.

For Linux MD, software raid, you generally have a 2000mhz to 3000mhz+ cpu backing it up.

So you guess which is faster.

It realy makes sense once you realise what is going on here.

There is no 'driver' tweaking needed. Linux software RAID is the bomb, it's better then FreeBSD/Windows/OS X/etc, it's something that Linux has strength in.

Supports almost any RAID configuration. 0, 1, 4, 5, 10, whatever you want. It supports many different ways to get raid using any block device...

I use 1 PATA drive on my onboard then I have 2 SATA drives on a PCI card in RAID 5 configuration. It's very nice.

In contrast those cheapy SATA 0/1 raid setups you get on cards are ASS. Always. ALWAYS, linux MD raid is faster.

What you want in hardware RAID in in Linux is RELIABILITY, not speed. If you want speed you can go with a Linux cluster using Lustre' or Redhat's GFS and get much more impressive results at a cheaper price.

Hardware RAID allows advanced error detection, working hotswapability, and other features that go beyond mere I/O.

Re:The results are obvious

[StillAnonymous]

"Dedicated hardware RAID devices are much more expensive, up in the hundreds of dollars for the controller."

I've never really understood why this is. Other than the features you mention at the bottom of your post (advanced error dections, etc) the actual RAID processing is extremely simple, no? It's just a quick parity scheme. XOR is a very basic function that is trivial and inexpensive to implement with a few gates. Why is it not used on every "BIOS" RAID implementation?

Even hotswap is becoming more commodity now, with SATA II specs requiring it. It's more of a hardware issue than a software one.

Re:The results are obvious

[VON-MAN]

I need to agree with you here. But when looking at the Adaptec results, it looks like the company does deserve its bad name.

Re:The results are obvious

[jericho4.0]

Because the people who really need it, can pay for it. Much like why Cisco hardware is so overpriced.

Re:The results are obvious

[norton_I]

To get even acceptable raid-5 performance, you need large cache. Linux can use system memory, but hardware raid solutions need their own cache, on some cards 1GB or more, sometimes battery backed so that data can be written out t. That, and the processor are not cheap (though they are not terribly expensive). Combine that with relatively limited market and lack of economies of scale, and you have somethings costing a few hundred dollars.

Plus, the market leaders have a competative advantage in percieved reliability and trustworthiness that is particularly important to their customers, which allows them to charge more than a potential competator would be able to.

Re:The results are obvious

[hartz]

RAID 5 in hardware needs to do a pretty bit more than a few XORs to calculate parity.

The process involves first a read of the stripe after determining what blocks on the disks is to be affected, into a buffer, then it needs to know which disk (column) contains the parity bits for the current operation, then it updates the data in the buffer, then re-calculate the parity bits, and then write back the whole stripe.

Some of this is avoided if the size of the data to be written matches the stripe width and aligns with it, but it still needs to do a whole lot more than just calculate parity. And if the data does not align with the raid stripe, the process may need to be repeated multiple times (for each affected stripe row), so the picture becomes much worse as the controller must also break up the request into parts which does align with the stripe!

RAID on the Desktop

[Master Bait]

Maybe software RAID is fine for desktop machines (as tested in the article) and maybe it is fine for dedicated fileservers. But hardware RAID and hardware SCSI RAID in particular are leaps and bounds more useful for anything but those two mentioned situations.

You're right that four drives in a software RAID setup will incurr four times the interrupts as a hardware RAID setup. Also, people tend to ignore seek times of SCSI drives which translate to greater I/Os per second -- something actually desireable in a multi-tasking, multi-user environment.

RAID is a waste of money

[Theatetus]

I know that's tech heresy but I think RAID isn't cost-effective. Spend the money you sould spend on RAID improving your backup and restore solutions.

Yes, 3 times out of 10 you can hot-swap a failed drive. The other 7 times, the controller itself goes, and 2 out of those 7 times it takes one or both drives with it.

Re:RAID is a waste of money

[Anonymous Coward]

All of this assuming you can afford to take the system offline to do the restore, of course.

Re:RAID is a waste of money

[the eric conspiracy]

The other 7 times, the controller itself goes

That's why you have redundant controllers and a dual channel architecture.

Re:RAID is a waste of money

[Kent Recal]

In my expirience it's more like 9 out of 10 times can you swap (or even hot-swap) a failed drive.

Out of 5 drive failures that I have expirienced only one brought the machine to a halt (assume the failed drive did something strange to the ide bus).
In all other cases the raid degraded gracefully and the machine could be shutdown cleanly to swap in a new drive.

It *should* be even better with server-grade SCA (hot-swap SCSI) drives because since these are built for hot-swap they are even less likely to confuse the bus when they go down.

So, as you might have guessed, I don't agree with you.
Swapping in a new disk is usually much less hassle than restoring from backup.
Even if the machine halts when a disk fails - at least you still have your data.

Re:RAID is a waste of money

[Anonymous Coward]

I use RAID for snapshot backup systems. It's very cheap now to use a RAID array to snapshot half-a-dozen different core servers, and do your tape backup from that.

The only times I've had controller problems have been with Promise, which is admittedly the worst RAID manufacturer I've even heard of.

Re:I translated this into Finnish

[Fizzl]

Haesta paska suatanan peikko.
Eihä tua näytä suamelta etes murteella.

Perkele.

Say it out loud

[Colin Smith]

Really. It looks really difficult when written, but spoken Norwegian is *almost* understandable to English speakers. Much like Geordie in fact.

Re:Say it out loud

[nietsch]

Unless spoken in norwegian, that is.
But you are right about the word similarities. This is where you can see some linguistic remnant of the language spoken by the saxons. The saxons came from borthern germany/soth denmark and migrated on a big scale to england.

As dutch speaker you see the same when traveling though scandinavia. The writing is rife with spelling errors, but you can quickly get the hang of it and read/guess newspaper headlines. That is untill you get to Finland, as finnish has a complete different ancestry.

Re:Say it out loud

[lawpoop]

A funny anecdote:

I was an exchagne student in Finland when I got out of high school. I was struggling to learn Finnish, which is totally unrelated to any European language. The part of Finland I lived in had a significant number of Swedish speakers, and my exchange student buddies who were learning Swedish were practically fluent already.

Every day on my way to school I passed by an AMT. The sign above of stuck out of the wall. One day I glanced up at it and it made sense -- "Gold mint" -- 'money mint'? I was elated that I was finally picking up the language. That same day I was returning home from school and I looked at the sign. It was as cryptic as ever -- it read "kultarahaa". Then, as I passed under the sign, I realized that there was easy to guess Swedish on one side -- "Guldmynt", and Finnish on the other -- "Kultarahaa".

I'm proud of my struggle to learn Finnish.

Re:Say it out loud

[Anonymous Coward]

> That is untill you get to Finland, as finnish has a complete different ancestry.

Some Finns describe themselves as the result of interbreeding Russians with humans. Which is not right. Entirely.

Re:Say it out loud

[Anonymous Coward]

Of course both British Saxons and Norwegians originated from the same Germanic roots.
But try starting with how the Vikings invaded and settled in Britain. That's how the languages came to resemble one another. In 1066 Norwegians and Britons spoke the same language - afterwards it diverged. Later the Normans - also Scandinavians but living in Normandie (Norman-country) France - came and conquered the British Isles again. Voila. French words entered the language.

I'm not sure I like this SATANIC RAID

[Anonymous Coward]

IN all but all these approve SATANIC- supervising the accumulations advertise facts along with RAID. This am actual a halvsannhet. These supervising able, as all but everybody else, counter in RAID, but mesteparten at RAID- characters happen to in the shareware in OS- a. IN Windows am this as a rule in carry on. Rumours say that they do be about look after RAID5 for å see about it is a genuine RAID- be in control of( as in that it facts transparent in hardware). About this figures am uncertain afterwards GNU/Linux has in shareware. Autobahn accordingly ought abrupt.

Yeah, real helpful there...

They've messed up somewhere

[tap]

At my previous job I built a number of RAID systems, from hardward SCSI to software ATA to hardware SATA.

I can't tell how many drives they are using (4?) or what raid level, but their benchmark results just aren't correct. They should be able to get bonnie++ read bechmarks in the 200 MB/sec range. They're getting in the 8 to 60 MB/sec range. The single character I/O benchmarks don't make sense either, they should be nearly the same with CPU usage at 99%. For some reason their disks are running much much slower than they should be. Just connecting two of those maxtor drives into the motherboard IDE controller and using linux software raid should be able to beat all their benchmarks.

Re:They've messed up somewhere

[notanatheist]

"hdparm" is your friend. Perhaps they're linux n00bs. Although, this does make me curious enough to warrant a better review. HW & SW RAID, linux distros with 2.4, 2.6, and proprietary drivers like Promise's, and throw in the BSD's for a well rounded review too. Don't forget filesystems too. There's been plenty of those though with EXT3 vs Reiser vs XFS vs etc...

Go work on a system with 24x7 req'd availability

[Anonymous Coward]

Stop working on toy computers and go to the real world systems that need to be up 24x7 with no excuses. I've seen companies spend literally millions of dollars when they discover their system has a design flaw that must be fixed because it results in a single point of failure that drops availability from 99.9999% to 99.9%.

Run your MTBF numbers when you have terabytes and terabytes of storage that must be online 24x7, but without RAID being used. Then compute the same thing with RAID.

Put this in your "lessons learned" part of your brain.

Rough .no-translation by a Norwegian

[zokum]

My own comments are inside []-brackets.

Bolded text:
What controllers should you look for for a new machine? Do you need one costing 2-4000 NOK (300-500USD) to maintain uptime and data integrity without losing speed? In this test we will look at some of the options and the results when building a good system.

Published May 13.
Most modern motherboards has some form of S-ATA. Both desktop and servers. One of the most common is the Silicon Image 3114. This is a 4 port SATA with alledged RAID-capabilities.

In almost all cheap SATA-controllers they tout raid capabilities. This is a half-truth. These controllers can as just about ever other controller be used in a RAID-array, but most of the work is done in the OS. In windows, mostly in the driver. Rumor has it one should look for raid 5 capabilities if one is looking for a true hardware raid solution (as in transparent to the rest of the hardware). Whether this is correct is not known since GNU/Linux has RAID5 support in software. The road ahead should be short. [idomatic expressiob, doesn't make much sense in Norwegian either in this context.]

[2 controllers' pics]

In this test we will look at different controllers. The aforementioned Sil 3114 is one of those cheap ones with fake hardware raid. How well does it do compared to a much more expensive SCSI setup. Is SATA so good that expensive SCSI setups is only useful in special cases?

Thanks to Nextron for a machine, several controllers and other equipment. And also thanks to MPX for the loan of several Adaptec controllers.

[Next page]

[I will skip most of the redundant translating] Fire diskport -> four disk ports.
[The comment about the 1.5 GiB memory is about finding a faulty chip.]
enhet -> unit
[long text]
This pretty server has almost all one needs in its small cabinet. It comes with "speed-couplings" [hard to translate] for SATA-disks, so the test with the SCSI controllers is done with an external SCSI cabinet and a PSU. The barebone system kan be delivered with SCSI if needed or one can add this oneself.

With it's 6 angry [slightly different conotations in Norwegian] and tiny little fans I would recommend being in the same room. Noicy like a small machine room. [as in say a boat].
[Next page] David. This chip has several Goliaths to fight.
* SiL 3114
On most controllers one sees this one or it's little brother. 3112 is often used as an interface to the disks. Simple controller with no RAID caps in HW.
* Megaraid. 150-4
This one has two 3112 chips for the SATA part and 64MiB ECC cache and an intel processor. It is not low profile but has a nice space saving design. Supports Raid 0,1,5 and 10.
* Megaraid 320-1
Low profile SCSI, internal and external connector. Has the GC08302 procsessor. Supports RAID 0,1,10,5,50.
* Adaptec 2130SLP.
Low profile. Internal and external connector. Has a staggering 128 MiB DDR Cache. Supports RAID 0,1,10,5,50 and JBOD.
*Adaptec 2410SA
Low profil SATA with two 3112 chips for SATA support, comes with 64 MiB cache. Supports 0,1,5,10 and JBOD.

[Rant about "true" RAID and level 0 and JBOD with link to a guide.]

The different controllers has support for various functions. LSI controllers tout their "on the fly" changes in the array, changing of raid-type without losing data and similar. Adaptec focuses on SNMP and a lot of the same as LSI. What one needs is up to the reader. The four "external" controllers come with various cables, manuals and CDs.

[Next page] During the test we used 50GB partitions. Sata disks were almost 3.5 times as big as the SCSI ones, and under this test the file system etc should change the results due to different physical size. It's not really possible to compare it directly, since the disks are quite different, we're looking for patterns in how te configs behave, not only if SATA can compare with SCSI.

For the test we used bonnie++ and dbecnh. [links]

Nonnie++ was us

Forgot HP SmartArray 6 too...

[Erik_]

In all the different RAID analysis on the internet, I never see people trying the HP SmartArray's. Those cards are the ones used by a lot Linux servers.

This article sucks.

[Scorillo47]

This is one of the worst behchmark articles I read in a long time.

What is missing is a systematic analysis of DISK performance in respect to various dimensions:
1) Pure disk read vs. pure disk write vs mixed I/O/
2) Sequential I/O vs random I/O vs. mixed
3) block size (512 bytes, 1K, ... 256K), or various requests combined.
4) Does it matter when you have multiple RAID volumes vs. only one? (this matters especially for SW RAID)
5) Disk/LUN size.

Also, the performance numbers should include:
1) The average and max read/write time. Distribution patterns.
2) The average and max queue length in various workloads.
3) The predictability of these numbers. (which matters for SW RAID)
4) RAID recovery time
5) Performace numbers when the RAID set is broken (one disk is bad).
6) Performace numbers during RAID recovery (set rebuilding - i.e. after you replaced the bad disk with a good one).

It is also interesting to find out various I/O patterns for real-world deployments, for example a database, a file system, etc. Isolating the I/O pattern for common file system operations, or database deployemnts, and map them to the figures above.

And, one more thing - an administrator will likely deploy RAID for its main advantage - relatively good read performance combined with its low downtime when one disk goes bad. But there is a drawback - if one of the three disks goes bad, the performance will be severy affected. On a badly designed system, the performance drop might render the server unusable if it cannot stand up the continuous flow of requests. So it is very important to know the particular behavior during RAID recovery. This data is also missing from the article.

Re:I translated this into Finnish

[rich_r]

Sadly, having lived with a finn for nigh on five years, the only word I understood was the last one.
I use it frequently when my normal vocab is undesired...

I dunno, man

[lorcha]

I have software RAID5 set up at home for my media server. Only had one disk failure, and the array dropped into degraded mode and I got an email alerting me that I had a disk failure. The next day I swapped in a new disk and rebuilt the array.

I'm a happy RAID customer.

2.4 is better than 2.6

[Vlad_Drak]

2.4.27 still provides better md performance than 2.6.9 says Neil, not sure if this go fixed in .11.

http://cgi.cse.unsw.edu.au/~neilb/

Re:2.4 is better than 2.6

[Anonymous Coward]

Absolutely! Here are test results from my computer, PIIXx driver controller attached
to two Samsung 120 GB disks, dual PIII:

2.4 read, no raid: 40920502 bytes/s
2.4 read, with raid: 58939326 bytes/s
2.6 read, with raid: 43712650 bytes/s

My milkshake is better than yours

[Vlad_Drak]

I use md so I can span PCI buses with multiple controllers to get better performance than a single HW raid card. Also, when my controller goes south I don't have to get the same controller. If I was really desperate I could use the onboard. I can upgrade my controller without backing up and restoring the array. I could get a SATA-II controller and slowly move my drives to SATA-II. I feel like I get more control with mdadm, too. At least I can inspect, alter or wipe the drive metadata while I'm up in the OS.

Re:My milkshake is better than yours

[NerveGas]

With a decent (133 MHz, 64-bit) PCI bus providing 600-800 MB/sec bandwidth, it's going to take a good number of fast drives in RAID 0 before you even come close to filling one bus. With any other RAID level, you won't even have to worry about filling that bus.

steve

CPU load impacting on softRAID

[yem]

I'm not familiar with the benchmark software, but do they stress the CPU to any degree?

In the Real World, CPUs with RAID storage usually don't just sit there spinning the platters. They're running high volume SQL database servers and application servers. These things have a habbit of hammering the CPU.

When the CPU is otherwise occupied, you'd think "software RAID" would take a big hit. Was this situation tested in these benchmarks?

Re:was results obvious, SATA Bus Mastering?

[Anonymous Coward]

When I started using SCSI, the thing I really liked was the
bus mastering (not all SCSI controllers have it of course.)
I could copy large amounts of data SCSI to SCSI and it wouldn't
bring the CPU to its knees. I bought a firewire controller
and adapter to make an old IDE drive look like firewire and
was pleased to see the same effect with that. But I haven't
tried SATA yet, and when I searched for info on SATA
on the net, I never could get a clear answer to the question,
does SATA do bus mastering?

Norwegian and English articles read equally often.

[Anonymous Coward]

You know, people read the fine article about as often in Norwegian as when in English.

Submission not stated that tests were performed using bonnie++ and dbench.

Re:I translated this into Finnish

[Aggrajag]

Maybe you didn't understand it because the grandparent used a dialect called 'Savo' in his message.

Re:I translated this into Finnish

[rich_r]

If only I knew enough finnish to make that distinction- a reflection both on the difficulty of the language and my profound inability to learn it!

Biggest probem with SCSI

[MiKM]

It also beats the crap out of your wallet. From what I know, the real advantage to SCSI over SATA exists in the workstation/server world, not the general user/gamer world. But I may be wrong.

Summary of issues

[Anonymous Coward]

It seems not everyone reads Norwegian. Here is a brief summary of the anomalies stated:

• sequential output per char:
  • hw and sw RAID mostly equally fast
  • however LSI SATA performed badly. In most tests LSI SATA performed sub par. This could be worth looking into.
• sequential output per block:
  • hw RAID shines, especially SCSI.
  • however sw RAID Adaptec SCSI outperformed the rest
• sequential input and output per char:
  • sw mostly outperformed hw RAID
  • However CPU use by Adaptec drivers hw and sw were strangely high

I hope this sumamry could be of use to the Kernel driver developer.

Comment removed

[account_deleted]

Comment removed based on user account deletion

this isn't helpful...

[dahlek]

Come on, people...what I want are solutions to make a kick-ass system, on the cheap.

Reading various man-pages and HOWTOs, you discover cute clues such as setting up software RAID1 on two similar drives each on it's own controller == _much_ better speed, at some risk (if either drive goes, data is lost).

I've also read that putting a swap on two drives will make parallelize the swap system and increase speed...

But I see no benchmarks or comparisons anywhere! From the way the RAID documentation makes it sound, I would think that people would be screaming to make a system with 4, 20 GIG drives RAID1-ed together just for the "extreme" speed boost. It seems perfect for a Myth type setup - maybe even allowing software capture on older hardware...

Where are the stats and benchmarks for this stuff??

Pay no attention to that parent poster.

[Anonymous Coward]

Before that text, there is some dispute as to where to install a 64bit PCI adapter. It reads

"As Koch in der poot ey seik mity pr0n sirber of Ahhhhhhhhhh"

As you can all agree, this was a dictation lost in translation. Some even go as far as telling that "Ahhhhhhhhhh" is a town...or the lost city of Atlantis! This is great news for nerds!

Unh? What?

[Anonymous Coward]

Software RAID has to share the systems bus, that means transferring the data from n number of drives across the PCI (whatever) bus whilst with hardware RAID it's all kept to the RAID card itself.

Excepting the slight extra overhead to send the additional CDBs, the data that traverses the PCI bus is identical for hardware and software RAIDs.

Think about it. If I write a file to RAID, each byte of the file goes across PCI once. It may not be in the same order for hardware and software RAIDs.

And what good would a hardware RAID controller that kept all it's data to itself be? I don't need hardware RAID to speed up my write only storage.

Re:Can't read the article

[Anonymous Coward]

Yet - you and everyone here want to read it, don't you?

Re:Can't read the article

[Anonymous Coward]

Yeah - because English and Norwegian have common roots. Both are germanic peoples. And Norway is still an entirely white nation - unlike the US. In the future Spanish will be numero uno in the US! We'll still have our language.

Re:I translated this into Finnish

[Anonymous Coward]

I will help you and translate that to proper Finnish.

Haista paska saatanan peikko.
Eihän tuo näytä edes murteelliselta Suomelta.

Perkele.

The perfect card for 8-drive SATA SW RAID ???

[Anonymous Coward]

Does anyone know how well these work with linux? They are both ~$100 a peice, and offer 8-ports on a 133MHz 64-bit PCI slot.

Supermicro's SATA-MV8 SATA controller provides 8-port SATA HDD support via 64-bit PCI-X bus interface with high-performance features. Built around the powerful Marvell SATA controller chip, it offers functionalities for server, workstation and network storage environments in a low-profile PCI platform.

http://www.supermicro.com/products/accessories/add on/DAC-SATA-MV8.cfm

Supermicro's SAT2-MV8 SATA controller (based on the Marvell Hercules-2 Rev. C0 SATA host controller) provides 8-port SATA HDD support via 64-bit PCI-X bus interface with high-performance features. Serving as a 2nd generation SATA storage card, the AOC-SAT2-MV8 offers double the data transfer rate of its 1st generation counterpart and offers functionalities for server, workstation and network storage environments in a low-profile PCI platform.

http://www.supermicro.com/products/accessories/add on/AoC-SAT2-MV8.cfm

Re:forgot 3ware....

[Terrasque]

Actually, they didn't forget 3ware.
The cards didn't fit in the cabinet.

Re:Go work on a system with 24x7 req'd availabilit

[Anonymous Coward]

Anyone who is looking to BUILD a system with these requirements, especially reading an amateur website such as Slashdot or the one linked to is going to regret it.

Put this in your "lessons learned" part of your brain.

Wow, no

[arete]

Wow, no. RAID 10 has a few niche applications, but...

RAID 10 is really not the best of any worlds. I'm all about 4 drives for redundancy, but RAID 10 is not optimizing _anything_.

A good RAID1 setup will read as fast as RAID0 - the controller or sw will read separately from each drive. The same is kindof true of RAID5 as long as the stripe/chunk size is large enough. So you're only talking about write speed. And most "speedy" applications are read/seek bottlenecked, not write bottlenecked.

If you need to maximize redundancy use 4 drive RAID1 - and make sure your bus has the bandwidth. Or use 3 drives, still improve your bandwidth, and buy more RAM.

But as long as your controller is fast enough to handle it RAID 5 will give superior speed AND superior redundancy on 4 drives compared to RAID 10. The only time I know of that that's not true is if you have software RAID and you are also CPU bottlenecked. To answer the GP's question - you still need RAID5 when you're not CPU/controller bottlenecked but you ARE write-speed redundancy bottlenecked. This is rare.

Other notes:
Cheap hardware RAID controllers ARE software RAID in the driver. Slightly better controllers offload the RAID processing - and might only be able to handle RAID 10.

The best controllers have substantial battery backed cache - meaning you can commit writes immediately without waiting for the seek. This is a HUGE improvement, especially on small files, and it can be seen using a high end RAID controller even with only 1 drive in a high IO environment. The other poster who said to turn it off is insane.

Re:The graphs are not self-explanatory!

[ozmanjusri]

Babelfish isn't the only translator on the net. Google found this one easily enough...
http://www.translation-guide.com/free_online_trans lators.php?from=English&to=Norwegian

Combat the slashdot effect...

[gods_design]

Write any page that might be interesting to slashdot geeks in a foriegn language that most of slashdots readers don't understand. They will all just hang out on slashdot and wait for a translation to be posted on someone elses servers! ;^)

just use binary, non-free LKM's

[Anonymous Coward]

Isn't that the Linux way? Who gives a fig about programmer documentation - afterall "just as long as I can use my system" is the only criteria Linux users care about, right?

Re:was results obvious, SATA Bus Mastering?

[jest3r]

Using IDE or (S)ATA, Bus Mastering is acutally called DMA. This is because unlike SCSI your IDE controller is not on the PCI Bus (hence Bus Mastering). However DMA has the same effect. There are actually two transfers modes that IDE can use - PIO and DMA.

PIO (Programmable Input/Output) data transfers use the CPU to control data transfers between the hard drive and RAM (very CPU intensive).

(U)DMA (Direct Memory Access) transfers do not involve the CPU, transferring data directly between the hard drive and RAM (very fast).

RAID + dual boot

[irw]

One particular advantage of hw raid over sw raid which I do not believe has been mentioned is the removal of complexity.

In particular, given a linux/windows (/anything else) dual boot arrangement, *sharing* a sw raid setup is going to be difficult, because each OS will arrange the raid layout on disk slightly differently.

I have hw raid5 on scsi with a large FAT32 data volume (best common FS between linux/windows), and no problems. I would not like to try to build this using sw raid (and I'd have to pay more for a server edition of windows rather than win2k pro, assuming windows dynamic volumes can *do* raid5).

On the other hand, if the raid board breaks, you're slightly screwed, unless it's recent and/or common enough that you can get a replacement (hint: buy a spare).

Re:I think that the results are not so obvious

[Anonymous Coward]

In no way a 300 GB 7200 RPM drive is better than a 73 GB 10 000 RPM drive, except in one way : the ability to lose 300 GB worth of data in one go when the disk crashes.

Luckily when your SCSI drive crashes, you only lose 73 GB of data. Hope that wasn't an important 73GB. IMHO, anyone who doesn't use RAID on important data is a fool, whether it's SCSI or IDE. Also in my personal experience, SCSI drives are just as unreliable as IDE. My guess would be due to the higher speeds necessary generating more heat and being more prone to problems.

I agree with you, except on the easier to connect part I don't understand. Anyway, I agree that SCSI is more expensive, but don't tell me SATA is higher perf, that is just a lie.

SATA is higher performance...if hdparm -t is your only "benchmark". (S)ATA drives are great at sequential throughput. But put them under load with many processes writing randomly and they start to look pretty bad compared to SCSI.

That said, I use ATA for my home storage needs. Why? I want the capacity. (S)ATA is "fast enough" for my needs. If I was running a database server that made me lots of money, I'd use SCSI arrays (as I do at work). But at home, it's not worth the additional expensive.

WD makes some SATA drives that appear to be identical to SCSI drives. Same capacity and spindle speeds (such as 73GB 10Krpm). It would be interesting to compare one of those to an identical specced SCSI drive. That would give a true indicator of how SATA as a bus compares to SCSI. My guess is they would be about even, that the big difference between the two is really in the drives.

The translation:

[hawk]

C'mon, it's quite obvious:

"Ten thousand XXXXXX slashdot editors
ran through the weeds,
chased by vun norvegian"

:)

hawk

Almost

[Hellkitten]

s/or/and/

asymmetric raid "5"?

[hawk]

You look like the right people to ask my question :)

Can raid 5, or something similar, be done with asymmetric disks? The particular application I'm thinking about is a central household server, and periodically replacing the smallest disk from time to time (or just adding more to the mix).

This would seem to give a reasonable safety level (if it works), while adjusting to the reality that backups, umm, tend to get overlooked "at times."

hawk

A new war!!

[hartz]

We used to have Mac vs PC, Linux vs Windows, Star Wars vs Star Trek ... Now we have SCSI vs SATA wars!