Slashdot Mirror
Latest SCSI Drive Reviewed
Sivar writes "StorageReview got their hands
on a Maxtor Atlas 10K V, the first SCSI hard drive in more than two years to double
capacity. Considering how quickly storage was improving just a few years ago, and other news like Intel's cancellation of the 4GHz Pentium IV despite AMD's lead you have to wonder if the traditional predictions of the end of Moore's Observation are actually beginning to come true."
the traditional predictions of the end of Moore's Observation
Thank you for correctly not calling Moore's observation "Moore's law". It's refreshing once in a while.
"A door is what a dog is perpetually on the wrong side of" - Ogden Nash
The other two models are 73 and 147 according to the article.
What the fuck do hard drive capacities have to do with "Moore's Observation," which was about transistors?
[
Welcome to the new /. where we just LOOK like we know what the hell we are talking about.
It's a doubling of the density of transistors every 18 months. It doesn't say anything about magnetic storage density or the clock speed of chips. Intel cancelling the 4GHz P4 was just admitting (and it's about time) that cranking up the clock speed is not the best way to improve CPU performance. There is no indication that will prevent Moore's Law from continuing
Jason
ProfQuotes
The article claims that hard drives are starting to clammer for 16 mb caches. It seems odd that no one has come out with a standard cache expansion kit.
A mother board with an ATA chipset that could plug in older dirt cheap SRAM or even newer DDR or better. Imagine a 4 gig cache of SRAM attached to your harddrives. A machine left on for a while would start to smoke.
I have some really highend SCSI raid controllers that allow 256 megs of cache...I wonder why there is a product out there to add cache to an existing ATA system. Obviously cost is an issue, but it seems like this sort of thing would give a big bang for the buck. High end games will pay anything for a 5% perf increase.
i wont be moving back to IDE.
First, Moore's Law has nothing to do with hard drive storage space. That said, hard drive capacities have been growing at a pace exceeding Moore's Law for several years now. If that rate slows down, it'll probably still be a pretty fast pace. Besides, these are fast SCSI hard drives. You have to look at IDE hard drives to really see storage space improvements.
Second, Intel cancelled their 4GHz CPU because of heat problems. It turns out that Intel's engineers just can't get the leakage current down to low enough levels. But again, Moore's law has nothing to do with clock speed... the metric is the number of transistors on the chip. In this regard, Moore's law is still on track. To counter the heat issue, logic designers will have to rethink their designs to do more work per clock cycle. AMD already does this with their chips. Intel is going down this route too with its Pentium M. Same with IBM's G5. The Pentium 4 is a horrendous example because Intel designed it to be inefficient so they could ramp its clock speed. Well now the consequences of that stupidity is showing.
You know, I've heard that the human brain operates at about a 10Hz frequency, has 100Bln neurons, and trillions of interconnections. Amazingly, its power dissipation is at around 40W. (And its MIPS rating is on the order of 10^15 instructions per second). Clearly mother nature got it right for efficient computation.
mean time between failure (MTBF)...
SCSI 1.2 million hours
SATA 0.6 million hours
That and SATA is still NO WHERE near the performance of higher end let alone mid-range SCSI drives.
SCSI drive capacities have stayed where they were while IDE drive capacities got bigger because for real-world RAID arrays (where SCSI drives are used) capacity isn't the goal. It's speed. If you need 1 Terabyte of really fast RAID storage it makes far more sense to put in 15 73gbyte3 SCSI drives (10K RPM, 15K RPM) than it does to use 4 300 GB IDE drives (7.2K RPM).
In the meantime IDE drives have begun to be used in RAID arrays, but usually where capacity matters and not performance. Admittedly the lines have blurred, especially for network-connected storage arrays where ethernet pipes are the limit and you cannot really tell the difference between a good IDE array and a regular SCSI array.
"You know, I've heard that the human brain operates at about a 10Hz frequency, has 100Bln neurons, and trillions of interconnections. Amazingly, its power dissipation is at around 40W. (And its MIPS rating is on the order of 10^15 instructions per second). Clearly mother nature got it right for efficient computation."
Try calculating PI on it.
Serial ATA will not be able to compete with SCSI in the server market until implementations of native command queueing are complete. Are far as I know, none of the current SATA drives on the market currently support NCQ.
It would be nice to see more hours out of the SATA drives, after all the big huff about the warrenty reduction by Maxtor and WD I picked up one of their "3 year" drives, it still shit the bed after 6 months. Yeah! 2 years worth of pr0n, Enterprise and Red Dwarf episodes gone. Guess you'll still have to role the dice on dependability.
We've said recently that as machines get faster, the software gets slower, so the work we have to do doesn't get sped up much (though the expectation for bells and whistles like fancy typesetting go up and up...), so would it really make such a big difference in our lives?
Here's one nifty thing that will break with Moore's Observation: the optimal slack time for large computations. If you're doing large computations, it would suck to see your slack time evaporate!
See what I've been reading.
SATA:
SCSI:
And yes, you can tell the difference, even as a "normal" user.
This doesn't mean SATA sucks. In fact, it's quite good for the target application: Low- to medium performance workstations. And it also doesn't suffer from bus termination-fu and other black arts of SCSI configuration. But there are still clear benefits to SCSI.
Schwab
Editor, A1-AAA AmeriCaptions
yes, I have OLD SCSI U160 drives and tried the new fast SATA drives for my Media array.
the old SCSI drives from 4 years ago kick the ever living crap out of the SATA drives.
this is non raid performance. When capturing RAW video from a TARGA 3000 card (A $7,500.00 professional video capture card) the SATA drives would drop frames and completely CHOKE after 5 minutes of capture at 40Megabytes per second.
the old SCSI drives with an even older 29160 scsi card had zero problems.
I hope that SATA will speed up eventually, but SCSI is drastically faster, even from ages ago.
I'm betting that SCSI U320 makes the fastest SATA stuff look like a complete joke.
Do not look at laser with remaining good eye.
I recently replaces an SATA array on a database server with SCSI disks. I had used WD Raptor disks which, from the specs, seem like a nice drive. But when I got the SCSI array put together, it was embarrassingly fast. There's something about SATA (and ATA) that makes it suck at synchronous, random loads.
I see 45-50MiB/s from my single 10k Raptor SATA. I know it's not SCSI, but I figure for open comparison. At work in a Dell 670 workstation with a RAID0 of Seagate 10k's I see around 120MiB/s. Not too shabby considering the parts involved.
wdd
SATA maximum spindle speeds are definitely not 7200 RPM. Western Digital's SATA Raptor drives are 10,000 RPM.
First of all, Moore said nothing about storage, only transistors in semiconductors.
If we assume there is a similar correlation with density on magnetic media, it still doesn't necessarily mean it's slowing down now.
AFAIK, drives had a major slowdown in the past around the 8GB mark and then suddenly 20GB->120GB appeared very rapidly, and then slowed down a bit then. I'd need to do alot of research and get some actual data before making a statement about exponential growth of magnetic storage density and whether or not it is feasible to continue or at what rate in the future.
Also, narrowing the comparison to just SCSI devices is foolish, as they are rapidly being supplanted by cheaper ATA based devices. Yes SCSI is superior, it always has been. Except in one place, cost per unit storage. And as they say, quantity has a quality all its own.
Also, lower costs disks such as SATA enable alternate means of increasing capacity and performance such as low cost RAID. SCSI used the RAID argument over mainframe SLED solutions to win in the market. Now mainstream SATA drives are using the exact same argument vs SCSI. The same principles that were true in the 80s and 90s are true now: more disks have inherant advantages, and can be flexibly arranged to provide whichever one you want whether it's performance, capacity, or reliability, in varying degrees. All for lower cost even with the added hardware overhead of the controller.
Finally, there's one more factor that can be causing the slowdown in disk expansion. The fact that file sizes do not expand at the same rate, so demand for larger storage is being outpaced by the increase in density. I'd be interested in seeing what the average webpage size is from 1994-2004. I'm sure it goes up really quick as features like image support and frames first come in, but then mostly levels off. Word processor documents, even bloated by modern office suites, are still not more than an order of magnitude larger than they were 20 years ago. People still put their school papers and resumes on (GASP!) floppy disks. And their rate of density increase has been zero for quite some time, discounting alternate formats such as zip and usb flash.
As storage continues to increase, we're seeing people actually have enough storage. I remember having to pick which games I could install on my 286 and 486. Now I just throw them on and by the time my disk fills in a year I just buy more disk as it's that cheap. My 105MB hardcard for my 286 cost ~$700 in 1989 or so. The 1.7GB fast SCSI-2 Micropolis HD I upgraded my 486 with the 525MB SCSI-2 Conner cost $900 in 1994. These days I could go grab a 200GB disk for $99 on sale. But the point isn't that the technology is better. In 1994 the biggest disk I could get was about 9GB and cost thousands. These days if I want the bigest thing on the block it's 400GB and costs under $400. What the average user gets in a new machine is much closer to the most advanced part in the market than it was 15 years ago when we had 340GB HDs in home machines and 4GB HDs in highend servers. Where did the highend disks go? RAID replaced them. These days if you want an order of magnitude more than what a major OEM ships as standard (Say, 160GB*10) you go for a RAID, either SCSI or ATA.
Once you're paying for RAID hardware you're getting performance levels in the enabling hardware that make SCSI irrelevant. SCSI has a 320MB/sec bus, command queueing on drives, and a dedicated CPU and cache on the host controller. A highend SATA RAID like 3Ware has 150MB/sec per drive non-shared switched bandwidth, command queuing on drives, and a dedicated CPU and cache on the host controller. Only the 3Ware setup will give you VASTLY more bang for the buck because you can buy more and larger disks to give whatever performance/capacity/reliability you want. A 12 drive SATA RAID10 is going to utterly destroy a 5 drive SCSI RAID5 in every possible way except for thermal output and physical space, which can be
Introducing the new Occam Fusion! Now with sqrt(-1) fewer blades!
Maxtor...
After losing a total of twelve DiamondMax drives to hardware failure, never again. Eight I had purchased, the other four were replacements for four failures.
I had four in two separate mirror configurations fail within minutes of each other. The original eight were bad within twelve weeks of purchase.
My local retailer honored the replacement warranties with more DiamondMax drives. I accepted on the first four failures and those died within 6 months.
Never, EVER again will I buy anything from Maxtor.
There's so little difference between politics and jihad lately...
I'm serious. Is there some way around the PCI bottleneck? Is it not as bad as I think it is? Should we all be using PCI-X anyway?
Sig:Why copyright isn't a fundamental human right
... how many Libraries of Congress is that?
This issue is a bit more complicated than you think.
1. 150MB/s is waaay more than a single drive can push, so it is more than sufficient.SATA is a point-to-point connection, one drive per channel. SCSI may be 320MB/s and support up to 15 devices, but that bandwidth is *shared* among all of them. By the time we have HDs that can actually deliver 150MB/s transfer rate, faster SATA will be available.
2. Maximum number of devices: that's a number you pulled out of your ass. You can have as many SATA devices as SATA ports. 3ware makes nice 12-port RAID controllers.
3. Spindle speed & seek time are the properties of the *hard drive*, not the *interface* (Do you understand the difference?). A SCSI and SATA HD with otherwise identical specs will have the same performance. Also, there are 10000RPM SATA HDs -- the WD Raptors, though they are not very cost-effective. If reasonably-priced 10K and 15K RPM SATA drives are released, they will totally kill SCSI market (which is, I suspect, the main reason they are not available).
___
If you think big enough, you'll never have to do it.
And it also doesn't suffer from bus termination-fu and other black arts of SCSI configuration.
that has been a non issue cince LVD scsi came into existance in 1998. active termination cables as well as active termination on the cards solve the issue completely.
Setting LVD scsi id is super easy (with 16 of them to choose from) and some newer drives will autoset their id.
anyways, SCSI can have multiple hosts on the chain. I can have a drive array with 2 computers accessing it at the same time. something that is 100% impossible with IDE.
finally, SATA may be able to handle a burst of that speed but it CERTRIANLY can not keep that data rate sustained. we tried for weeks to integrate SATA drives, the best we could buy, into the video capture workstation at work.
we went back to the stack of 12 32gig scsi drives and kept the SATA drives for a storage server use only.
SATA is very low performance to even the older SCSI equipment.
Do not look at laser with remaining good eye.
Then all those software companies making investments on more and more inefficient software are gonna take a hit big time. It would definatly be nice to see a good sine curve to moorse's law, whereas you get peaks of developement (meaning, progress is doubling every year or so) and drops (where tech is only gaining in 1.2-1.3 times capacity every 2 or so years). Gives technicians a chance to catch up and spend time unionizing, gives companies time to review their strategies and focus their designers on better materials and more feature filled hardware, and it forces software designers and especially their bosses to rethink their strategy of creating ultimatly trashy, inefficient, flashy software tools.
As for moors law coming to an end, we'll have to see. There's been an auful lot of new stuff on the horizon, and I think we've gotten to the growing pains number 4, where major hardware changes are occuring; the first started with the 80386 and 80486, virtual mode, simm memory, EISA, IDE, and AT standards. The second with the pentium, EIDE, PCI, AGP, MMX, 3dNow, widespread modem use, and CD-rom's with the ATX standard. The third with the pentium 4/ddr/qdr, DVD-rom drives, PCI taking off into never never land (how many different kinds of cards is that?), LANing PC's together via DSL lines. Now we're in the 4th generation, where we've got 64 bit datapaths, new instruction set additions, SATA, PCI-X and PCI-express, DVD burners, Gigabit ethernet, usable, pretty linux, mini-ITX standards.
The first set of changes turned the PC into a mutli-user inexpensive platform. The second gave it internetworkability and spurred the internet, as well as drove it into some multimedia stuff. The third added 3d gaming to the platform, perfected the networking aspect, and added a lot more data features and especially, and most importantly, stability. Now, we're getting into the most significant of those stages; making machines a *lot* more powerful and easier to configure. Just look at some of the newer 3d games coming out, I remembered watching some Cutscene's from old FF games as well as some old computer games, and Doom3 blows their socks off. Again, after these changes have occured, we'll move into another term of relative peace.
The 5th generation tech I fully expect to come in around 2007-2008, and will be centered around public wireless networks (more or less, people leaving open wifi all over the place), porability, altered reality (think virtual grafitti, waypointing your friends, ect). It'll also be marked by a major freedom vs corporatism; DRM vs the internet, for example; DRM will probably seek to segment the internet into trade zones, or as the companies will call them, "trustworthy zones"(example message: You are leaving the safe zone, if you leave the safe zone, you will be subject to viruses, trojan's, malware, and bad stuff. Do you wish to continue?"). As malicious software becomes more prevalent and voracious, we'll see the open source movement gaining a lot of steam considering these corps will begin digitally enslaving people. Why spend a billion on advertising when you could simply serve it to people off of their own computers?
So, within the next few years, we're going to see a lot of bad and good things happening, and most likely, some people's lives turning to hell, namely, those who don't care. Those who choose to fight it out will probably be persecuted; breaking DRM is, afterall, against the DMCA, and if MS gets angry, they can pull strings to have your linux-coding monkey ass assassinated or thrown into jail as a terrorist. Things'll get interesting, to say the least.
Candy-Coated Knowledge
> some newer [LVD] drives will autoset their id.
Actually, I haven't seen a disk in YEARS that won't do this.
Of course, it may have something to do with the equipment I'm buying; I was kind of surprised to see that there are still 7200 RPM disks out there when I read the OP's post. Most of my disks are 15,000 RPM with the older ones being 10,000 RPM.
> we went back to the stack of 12 32gig scsi
> drives and kept the SATA drives for a storage
> server use only.
If you want to ditch the cable nest, think about switching to FC_AL. I'll bet you could talk a Sun A5000 into talking with a windows or linux box without too much grief. These enclosures are going for pretty cheap on eBay these days because they only hold 14 disks and are getting near EOL (FWIW, A5200 holds 22). Then populate with FC_AL disks as you see fit. These are basically SCSI disks in terms of everything but drivers, HBAs, and connectors.
> SCSI can have multiple hosts on the chain.
Which is a damned useful feature when setting up redundant (HA) servers. FWIW, you can do the same thing with FC_AL, and the A5x00 arrays can actually have four separate GBICs on each enclosure. Two GBICs per IB gives you a theoretical max of two loops at 100 MBytes/sec (each FC_AL disk has two interfaces as well). Then you can run two hosts (or more if you want) off each loop.
Do daemons dream of electric sleep()?
nah, cpu and ram will be used up by higher and higher levels of programming languages. Just look at .net, easy enough but a simple hello world has a 40 meg footprint
This is a joke. I am joking. Joke joke joke.
... I can guarantee you most DB guys I know would shit their pants in joy if they could get 15k RPM 9GB drives in bulk. I know of DBAs that buy 18g drives and only use half of them. In theory you only use the inner cylinders, but internal geometry these days is largely divorced from logical geometry.. DBAs who deal with random small writes want lots and lots of spindles striping using lots and lots of hardware RAID adapters.
The super exciting thing about the 2.5" drives IMHO for SCSI is the possibility of boosting rotational speed thanks to reduced media weight. If you could get 1" 20-40kRPM 9GB SCSI or SAS drives and join together 100 of them that would be unbelievable.
What the fuck do hard drive capacities have to do with "Moore's Observation," which was about transistors?
Other than the curious observation that both IC density and magnetic storage density happen to be ceasing to scale up at the same time?
All operating systems do this anyway with your system RAM.
The memory on the drive is just there as a holding pen for pending reads and writes so that it can give the drive head a chance to get to where it needs to be, perhaps killing multiple birds with one stone.
At a certain capacity you start needing more cache because you'll be dealing with potentially more complex access patterns (more disparate regions to access data, larger transfer units per track)
It is not a substitute for a file-system/block cache.
THIS THING CAN TURN ON A DIME, MACROSSZERO STYLE ALSO FUCK BETA, ~NYORON
Had nothing to do with hard drives, or even processor speed. It merely stated that technology was advancing such that the number of transistors on a chip doubled every 18 months.
https://www.eff.org/https-everywhere