IBM Launches New Product Line

An anonymous reader notes that "IBM has launched its new product line of storage devices: the DS6000 and the DS8000. The results are quite impressive, with the DS6000 being rack mountable, 3U, and ONLY 125 pound storage device that will hold up to 67.2 TB! The DS8000 is equally impressive, with 6x performance of ESS 800 (Shark), making it the most powerful storage system to date. "

Comments

Hot Damn, now I can finally ...

[bushboy]

download the whole internet !

Re:Hot Damn, now I can finally ...

[moro_666]

if you are strong enough, get also the ds6000/8000 backpack from ibm and carry the whole internet around with you :)

Re:Hot Damn, now I can finally ...

[slarshdot]

Then you need another wad of cash to buy something to back it up!

Re:Hot Damn, now I can finally ...

[Emugamer]

download the whole internet !

not according to your sig

Re:Hot Damn, now I can finally ...

[mirko]

Your comment actually drew me to some conclusion : if it's cool to have punctual storage units that "huge", I think I'd like better some distributed secure raid over ip devices because I still think it's not the size that matters but it's the interaction and the evolutivity.

Re:Hot Damn, now I can finally ...

[earthstar]

But Youd still need Google to access the content.

Seriously,If google intrduced a hard disk search, i think it would be faster than the current speed with whick we search.
[oh ok.i know Longhorn is tryin this.Whether it will succeed..God knows.]

Re:Hot Damn, now I can finally ...

[bushboy]

haha - I was waiting for someone to comment on that ;)

Re:Hot Damn, now I can finally ...

[Cygnus78]

download the whole internet !

Do you have a torrent ?

Re:Hot Damn, now I can finally ...

[moyix]

I believe you mean you can download one of the "internets".

Re:Hot Damn, now I can finally ...

[Anonymous Coward]

lol, i was waiting to hear a joke about george bush's great knowledge :D

Re:Hot Damn, now I can finally ...

[jaephu]

"if you are strong enough..."

unfortunately, not many of the people who read this are...

Re:Hot Damn, now I can finally ...

[bdcrazy]

If google introduced a hard disk search, it would be remarkably similiar to disk indexing in windows (i'm sure there are other utilities that do as well and prolly work better). Google is fast because it indexes and caches the data for easy retrieval. So you spend hard disk space and processor cycles for easier retrieval. Wow that is what google must be doing :)

Re:Hot Damn, now I can finally ...

[simcop2387]

nah we've been able to fit that on 5 1/4" floppies for decades

With all that storage...

[Ammishdave]

Imagine the pr0n... and the movies... and the music... and everything else, for that matter heck, why not just download the portion of the 'net I go to on to one of them. If only I had the money

Re:With all that storage...

[Mjlner]

If only I had the money

RTFA! It's only 125 pounds! (Sterling, I'm sure.)

Re:With all that storage...

[Anonymous Coward]

No, not Sterling... flesh.

Re:With all that storage...

[Antique Geekmeister]

Well, yes. But that "67.2 Terabyte" quote is misleading. 67.2 Terabytes takes whole racks of the expansion units. Price starts at $100,000 for a single storage unit. So let's run the numbers.

400 Gig drives, 16 in a 3U chassis. Hmm, that's probably a pair of 8x RAID5 units, so call it 2x7x400 Gig, or 5.6 Terabytes in a 3U chassis. Not bad, pricing for those sorts of things with AMD's 64-bit CPU's is in the $15,000 range right now for stand-alone Linux servers. IBM is using the less PowerPC chip, which means it will be tough to run standard software distributions on it rather than IBM's more proprietary offerings, but you may not care about that on a fileserver.

But if the $100,000 IBM wants is just for the basic unit and no expansions, doesn't include the fiber switching and the racks and nice sets of controllers, it's wildly out of scale with the market. Fiber channel and massive SAN's are very nice for certain applications, but they've fallen off the development map for most companies as drive prices continue to fall massive chunks of cheap disk are proving so useful.

Re:With all that storage...

[agallagh42]

Each 3U unit actually holds up to 16 300GB fibre channel disks. That 4.8TB per unit. You can then chain up to 14 units together (one base unit and 13 expansion units) for a total of 67.2TB.

That's raw capacity, not accounting for redundancy.

"IBM is using the less PowerPC chip, which means it will be tough to run standard software distributions on it rather than IBM's more proprietary offerings, but you may not care about that on a fileserver."

These are not servers, and will not run any "software distributions". They are simply storage boxes, and they connect to your servers via fibre channel. They can connect to "IBM z/OS®, IBM OS/400®, IBM AIX®, as well as Linux(TM), UNIX®, Microsoft® Windows®, HP-UX and SUN Solaris environments."

Re:With all that storage...

[Antique Geekmeister]

OK, your post is accurate, and thank you for listing the actual drive specs. But note that "16 * 300 Gig drives yields 4.8 Terabytes" is marketing horseswill. Unless you're actually not doing RAID, you lose at least some of that space to the RAID parity checking. In a 16 drive array, this is typically 2x(8-1), or typically 2 arrays of 8 drives each giving up one drive to the RAID5 parity, so your actual effective space is 1/8 less. Call it 4.2 Terabytes in a 3U rack space, which is not bad at all but all you can really get into it.

And the concept of "simply storage boxes" would only make sense if the array were not connected to anything else, say as a gigantic Just a Bunch Of Disks (JBOD) of doom. You have to share that data somehow. If the sharing is done over fiber-optic to the rest of a network using the PowerPC chips, well and good. Since all of those listed clients now speak SMB, it could be limited to SMB with the pitiful limitations to user authentication and access control that implies. And ye ghods, what about the capabailities of the underlying file system? Where will user login management happen, or backup?

So now you need a box with all these OS level tools associated with the storage array. Is it good to put it on another box, or just integrate it into the storage array? How is IBM doing it? Are they going to overwhelm with network traffic and file system management that 1 GHz 65-bit PowerPC chip with it, rather than using a 2.x GHz AMD 64-bit chip?

Re:With all that storage...

[agallagh42]

Yes, good points about available storage vs. raw storage. They usually market these things based on raw storage, since the user can decide if they want redundancy, and how much.

However, I think you're confusing NAS (Network Attached Storage) with SAN (Storage Area Network). Nothing gets shared to users directly from these boxes. In fact, the SAN doesn't have to be connected to an IP network at all (although most can be for management purposes only). The fibre channel (not necessarily fibre optic, it works on copper as well) connection acts like SCSI. It actually uses the SCSI command set if I recall correctly.

The storage acts as if it's local to each server that attaches, and then the server shares it out however it likes. It can be used for file shares, database storage, mailboxes, whatever.

For example, if you assign 1TB to a windows server, and connect that server to the SAN, you'll get a virtual disk device in your device manager, and disk manager will show an available disk for partitioning. You can then format the new partition, say as the D: drive, and format it with NTFS. Similarly, if you connect a linux server to it, you can partition and format with ext2 or whatever other filesystem you fancy.

The SAN itself doesn't care what filesystem is placed on it, it only cares which servers get space assigned to them, and how much.

DS?

[Anonymous Coward]

Does that stand for *cough* DeathStar, er *cough* I mean DeskStar hard drives?

Re: DS?

[Anonymous Coward]

I wonder if DS9000 would be any better reference either..

Huh?

[bigberk]

IBM has launched its new product line of storage devices

What's that?? I can't hear you over my screeching Deskstar 75 GXP!!

Re:Huh?

[Anonymous Coward]

Totally different depts. within IBM, and in fact the part of IBM that manufactured the Deskstar is no longer part of IBM... was sold due to its repeated failures.

To inform

[a.different.perspect]

More articles, for the more article inclined.

Re:IBM sucks

[Anonymous Coward]

IBM supports Linux and this is Slashdot, so you can't say IBM sucks.

Alike-thinking minds are, well, alike

[a.different.perspect]

In Soviet Russia, The Hard Diks Mounts YOU!

You and that poll troll have something going on, don't you?

i hope these restore ibm's name

[Anonymous Coward]

In recent years, IBM has made horrible horrible storage products. I hope these aren't more of the same.

Re:i hope these restore ibm's name

[LiquidRaptor]

They had all of one bad line, they fixed it without issue. Shit happens, people make mistakes, at least IBM fessed up and fixed them. I would still trust an ibm drive over anything else today, except maybe a samsung who if I remeber correctly bought IBMs harddrive buisness.

Re:i hope these restore ibm's name

[Anonymous Coward]

hitachi

Re:i hope these restore ibm's name

[PygmySurfer]

Hitachi took over IBM's Desktop hard drive business.

And I believe IBM actually had 2 lines that had issues (The 75 GXP and, to a lesser extent, the 60 GXP).

I had 2 30 GB 75 GXP drives, I think I ended up going through 3 RMAs. Eventually, IBM replaced one with a 60 GB 120GXP (I believe it was the 120 GXP) with an 8mb cache (original drives only had 2mb cache). While the RMAs were a hassle, IBM did a pretty good job of taking care of me.

Re:i hope these restore ibm's name

[blowdart]

Actually I've seen problems with some old IBM NAS devices. We had around .75Tb of streaming video we needed to host and a NAS solution (rather than a SAN) worked quite well, we could put really cheap streaming servers and feed the content off a NAS cluster.

There were 2 problems really (excluding that it was a "Windows Powered" device, that part actually worked well - cue -1 Troll). IBM's clustering software was so primative that you had to mirror hardware exactly. Right down to the memory, processors and even the model number of the NAS. Kind of hard to do when products are no longer sold and you want to add a new NAS into the cluster.

What was worse though was the redundancy. Yes we had redundant drives, power supplies, fans, even processors (if one died it would keep going). We were quite smug. Until the motherboard blew up. On both NASes. Within 12 hours of each other. With no external influences like overheating or power fluctuations. And the engineer says that it was a common problem. And took 48 hours to find replacement parts, then the storage array decides that because the motherboard has changed it's not going to mount. And another 24 hours to get that fixed.

The new motherboard on one machine lasted for 2 weeks before blowing again.

Re:i hope these restore ibm's name

[Russellkhan]

It's been awhile since I worked in storage, but when I did the name to trust in SCSI drives was Seagate.

But I must admit, I haven't exactly been keeping track lately and that's the type of thing that may have changed since I last looked.

Re:i hope these restore ibm's name

[Gene Ray]

40 GXP also.

Re:i hope these restore ibm's name

[Anonymous Coward]

The ugly part of the 75GXP fiasco wasn't that IBM had manufacturing and quality assurance problems, but the well-reported (at the time, no link at hand) incident that IBM received a batch of 75GXPs *rejected* by a major OEM and consciously pushed them into the retail channel regardless.

I still generally love IBM (especially the hardware) but there was worse business involved than just mechanical trouble on some batches. Some of those managers should have been fired, dunno if they were. Nobody outside knows how common those kind of shady incidents actually were (in that division).

Brouhaha ensued promptly, no wonder they sold off the biz to Hitachi.

Re:i hope these restore ibm's name

[Antique Geekmeister]

The 40 Gig drives were also known as "Deathstar" instead of "Deskstar. If you were one of the poor bastards who populated a RAID set with those and never backed it up because the "RAID protected you from drive failures", you were dead when the drives started popping like marathon runners at the 20-mile mark and your RAID set lost 2 at a time, or another one failed while you were rebuilding form the previous drive failure. This was very, very nasty to deal with.

Actually it's 4.8TB for a single rack

[Anonymous Coward]

It's 67.2 TB if you have 14 racks (224 disks)...a single rack only allows 300Gb x 16drives = 4.8 TB...quite still a lot though.

Re:insert witty pr0n comment here

[Anonymous Coward]

Alrighty well from now on we'll make fun of your name instead Erik Sodomystorm.

Writeup is wrong

[amorsen]

The DS6000 supports up to to 67.2TB, but not in one enclosure. The DS6000 only fits 16 disks per enclosure, and with 400GB disks that is 6.4TB. 400GB disks seem to only be available as SATA and PATA, the largest SCSI disks I could find are 300GB. That means 4.8TB per enclosure. 16 DS6000's per 48U rack, that's 76.8TB. Remove every 8th disk for RAID-5, that's 67.2TB.

Re:Writeup is wrong

[OrangeTide]

8 disk RAID-5? You have a lot more guts than I do!

Maybe raid5+1 or maybe four 4-disk raid5s stuck together in an append or raid0. Or maybe raid6, if anyone ever releases a product that makes it easier to manage.

Re:Writeup is wrong

[Russellkhan]

RAID 5+1? You have a lot more time and money than I do!

RAID 10 I can see if you really need high redundancy/availability, but 5+1 is just way too slow and too disk-hungry for any practical use. (what company or person wants to buy 16 disks for every 7 disks worth of storage they get to actually use?) For most uses, RAID 5 or RAID 3 do the job very nicely, providing decent redundancy without trading off too much space and performance. And yes, I mean 8-disk RAID 5.

RAID 6 is like RAID 5+1, but not as bad - poor performance (compared to RAID 5 - twice the parity to calculate) decent as far as disk usage - two disks out of every 8 go to parity. But it has nearly no support in the industry - RAID 6 solutions tend to be expensive enough that most who would consider it just go all the way and get a RAID 10 setup - which is also much easier to manage.

What exactly do you do with your arrays that makes you require 5+1?

Re:Writeup is wrong

[Anonymous Coward]

Our racks are only 42U. Where are you getting yours? (PS - I'm glad someone else did the math - it was pretty obvious 16 drives in (3U!) wouldn't make 67.2 TB without some seriously monstrous drives that haven't been invented yet.)

Re:Writeup is wrong

[teg]

RAID 10 I can see if you really need high redundancy/availability, but 5+1 is just way too slow and too disk-hungry for any practical use. (what company or person wants to buy 16 disks for every 7 disks worth of storage they get to actually use?)

Lots of different terminilogy here.. when talking about 5+1 in a RAID5 setting, it's often a short way of saying "5 data disks +1 parity disk" for each set. 5+1 is a common configuration for write intensive tasks, since parity will be a limiting a factor. For more read intensive tasks, 8+1 isn't uncommon

But to add to the confusion, you do have raid 51 too... for business critical systems for large companies, using more disks isn't that much of an issue either.

Re:Writeup is wrong

[Russellkhan]

I have never seen the term RAID 5+1 used the way you describe, and in the context of the post I was responding to, it doesn't seem to make sense. I'll quote:

"8 disk RAID-5? You have a lot more guts than I do!

Maybe raid5+1 or maybe four 4-disk raid5s stuck together in an append or raid0."

Note that the poster uses the term "8 disk RAID-5" to refer to a RAID 5 setup where one out of every 8 disks worth of space is dedicated to parity; then he uses "4-disk raid5s" to refer to 4disk arrays with one drive's worth of parity - right between the two he mentions "raid5+1" as an alternative to the two. To me, it's pretty clear that the meaning you're inferring is not the one the poster intended.

I am familiar with "raid 51", it's the RAID 5+1 that I and the poster I replied to above were referring to. Go back and read PC Guide again, the author points out that it is also known as RAID 5+1. He also points out that it's too slow and expensive, and recommends RAID 10 or other, non-RAID solutions to those who may be considering it. (PC Guide is a great site, BTW - it's the one I used to recommend to new (non-technical) employees as a primer on RAID when I worked for a storage company.

Re:Writeup is wrong

[Russellkhan]

OK just re-read your comment. I knew it had to make more sense than I originally thought. And it did.

"when talking about 5+1 in a RAID5 setting, it's often a short way of saying "5 data disks +1 parity disk" for each set."

I have seen this usage, and it does make sense, but not in the context - and it would be very sloppy to say it as RAID5+1, which is what the poster I replied to said, and which is how I misread your post earlier.

Sorry about the misunderstanding.

Re:Writeup is wrong

[City Jim 3000]

Doesn't RAID5+1 normally mean RAID5 with one hot spare?

Re:Writeup is wrong

[keesh]

IBM's standard is 6+P+S (six normal, one parity, one spare). Since the monitoring setup is damned good, and the CEs are really fast in replacing drives, it seems to work. The only reason raid 6 exists at all is because EMC accidentally shipped a bunch of duff drives once.

Re:Writeup is wrong

[Russellkhan]

It can mean that, but I don't think it does in this context.

To my mind, 'normally' it means RAID 5, mirrored.

Hopefully the poster I replied to will reply at some point and clarify his meaning for us.

Re:Writeup is wrong

[drsmithy]

RAID 6 is like RAID 5+1, but not as bad - poor performance (compared to RAID 5 - twice the parity to calculate) [...]

It's not the parity calculations that are the bottleneck for RAID5, it's all the additional I/O required.

Re:Writeup is wrong

[amorsen]

Our racks are only 42U. Where are you getting yours?

You are right, IBM probably considered 42U racks (although 48U racks are available, as another poster mentioned). This means that the capacity is 67.2TB without RAID or with RAID-0. Now how many people dare make a RAID-0 of 224 disks?

Re:Writeup is wrong

[amorsen]

They arent SCSI drives, theyre FC drives.

That makes sense. So basically, this story is a total non-story. An expensive enclosure full of expensive disks is hardly news.

Re:Writeup is wrong

[minion]

8 disk RAID-5? You have a lot more guts than I do!

Maybe raid5+1 or maybe four 4-disk raid5s stuck together in an append or raid0. Or maybe raid6, if anyone ever releases a product that makes it easier to manage.

We have (counting on fingers) 8 storage array cabinets very similar to IBM's - they're Infortrend devices. All of these units are either 12 disk or 16 disk devices. And we break them in half. So, we have one global spare for each cabinet, and an array of:

6 RAID-5 and 5 RAID-5 drives (1 spare)

8 RAID-5 and 7 RAID-5 with one spare.

No problems.These things have proven to be VERY reliable.

Re:Writeup is wrong

[OrangeTide]

two RAID-5s mirrored with a hotspare is a very common configuration. be it a 9, 13, or 17 disk. This is almost the rule for "enterprise" IDE and SATA raids. And extremely popular for SCSI setups too.

Re:Writeup is wrong

[Russellkhan]

"two RAID-5s mirrored with a hotspare is a very common configuration. be it a 9, 13, or 17 disk"

OK, so that's what you meant by RAID5+1. I have never heard the term used that way. I read your RAID5+1 as meaning something like this, which was the the most common meaning back when I was more aware of current terms and trends in the storage industry.

My argument was based on the belief that that was your meaning, so pretty much none of it applies to what you said.

hyperbole.slashdot.com

[wombatmobile]

"These are the most significant storage announcements we have made in more than a decade. IBM is focused on being the storage innovator and clear technology leader," said Dan Colby, General Manager, IBM Storage Systems. "Today, we are delivering new economics and choice by leveraging common components, breakthrough technologies from mainframes and supercomputers, and unmatched virtualization and management capabilities."

Most significant in a decade? New economics? Wow, this is too important for Slashdot. Somebody should call Time magazine. Or Newsweek.

Re:hyperbole.slashdot.com

[Anonymous Coward]

Wow, this is too important for Slashdot. Somebody should call Time magazine. Or Newsweek.

Where's Steve Jobs when you need him?!!

Only 125 pounds?

[mrjb]

At that price I'll have one.

Re:Only 125 pounds?

[Kyont]

Sure it only weighs 125 pounds now, but what about when you load it up with all those terabytes of data?

(This would be funny, except I stole that joke from a customer of mine who came up with that innocent question when we were speccing out the floor strength of his server room).

not 67Tb in 3U

[swmike]

It'll grow by the modular 3U unit.

The single 3U unit won't hold 67.2Tb, that's a bunch of them linked together.

Finally enough space!

[indiancowboy]

Phewww! finally enough space to keep all my porn in one place!

Longhorn System Reqs.

[jaephu]

uh oh... Microsoft Windows Longhorn Minimum System Requirements: ... Hard Drive: 30TB Memory: 2 GB

Re:Longhorn System Reqs.

[rock_climbing_guy]

No, you need 2 GB memory just for all the spyware and worms you'll be running. It's in addition to the everything else.

Re:Longhorn System Reqs.

[Entropius]

Wow.

The KDE guys are going to have to get busy if they're going to be competitive with those specs.

Re:Longhorn System Reqs.

[rts008]

Not hardly! If they keep dropping Features to get it out on time like they have been, it will only take 25TB, but because of all the "rush it out the door quick" bugs (and all the crap it will contract over a network) you will need 4GB mem

That's not 67TB in a single enclosure..

[jcr]

You misread the spec, I believe. There's 16 drive bays, and the biggest single drive I'm aware of at the moment is 400 gig.

What they said was: "Using modular, 3U, 16 disk drive, rack-mountable enclosures, the DS6000 series can grow along with your storage needs up to 67.2TB physical storage".

According to the datasheet, they offer drives up to 300gig in each bay, which works out to around 4.8 Tb per enclosure.

-jcr

Product pricing and availability

[just someone]

Product pricing and availability
IBM's new storage offerings with enterprise class functions reset the bar with minimum configurations starting at half a terabyte and list prices starting as low as $97,000. The DS6000 series and the DS8000 series come standard with a four-year warranty on hardware and software, which is unique in the industry.


What are they smoking? 9.7 k a terrabyte, maybe. 97k. Even EMC is not that high any more.

Re:Product pricing and availability

[Anonymous Coward]

Yes, these prices are ridiculous.
You can get 5.6 TB for $10k in true 3U using VTrak 15100 from Promise.
That's $4k for VTrak plus 15 x $400 for 400GB drives.

Re:Product pricing and availability

[Pemdas]

I think the target here is ultra-dense high capacity storage users. It has to be, though I can't imagine there are that many customers in this space.

Certainly for lower capacities (e.g. 10 TB), there are much, much cheaper turnkey solutions in the form of the XServe RAID from Apple, not to mention the stuff Promise sells.

Re:Product pricing and availability

[MemRaven]

What you're paying for in the $97k base configuration is the chassis with a few disks. That chassis may seem expensive, but considering the kind of redundancy built in (dunno if $97k is the fully redundant system with dual caches, dual FC switches, dual PowerPC processors, dual power supplies, you get the picture), it's a pretty low price.

Remember, when looking at stuff like this, you're really looking at the price of the chassis as the base of a much larger system, not a "I need half a terabyte, what should I use?" If you just needed half a terabyte in total you almost certainly wouldn't be using one of these.

Re:Product pricing and availability

[bleh-of-the-huns]

There are plenty of customers in this space, government, financial and ISPs to name a few, look at products like nixon, (whatever NFR's flight recorder is called today) and other products that store every single piece of data that goes in and out of a network. I work for a civilain gov agency, we generate 2TB of data a day, store that for 10 years, something like this becomes usefull, although I suspect in our case, a larger SAN would be much more efficiant. Banks, same way, they need to store huge amounts of data for long periods of time. The fact that this is modular is great, you buy what you need now, and add continuously. Credit reporting companies, I despise them as much as the rest of the folk, but they still need to house the data somewhere, and I suspect they probably have the largest DB of any customer/gov/whatever in existance.... I could be wrong there but I suspect I am not.

Re:Product pricing and availability

[Anonymous Coward]

What are you smoking? When you are buying storage like this, you're not simply paying for a half a terabyte of disk. You're paying for massive scalability, integrated data services at the storage level such as instant snapshot copies. You're paying for dense fibre channel connectivity, very large disk caches. High End Enterprise storage systems can have on the order of 128 GIGS (See Sun/HDS 9990 for an example of a very high end storage system) of disk cache, to speed up the access to all the porn the admins hide on their SANs. That's a hell of a lot more than the 8 megs of cache my desktop HDDs have. So it may seem very expensive per terabyte, but it does a hell of a lot more than just a bunch of SATA drives configured for RAID5 or RAID1+0.

Re:Product pricing and availability

[sirwired]

Folks, the 97k is a base unit, with dual, bulletproof controllers, and the minimum amount of storage. Doubling the storage does not double the price.

SirWired

I have two responses

[a.different.perspect]

"Gods witnessed glaciers forming faster than my downloads...now I can finally download the whole internet"

Yet presumably only after the entire planet is covered in glaciers.

"Hot Damn... Gods witnessed glaciers forming faster than my downloads"

If you already have a hot dam, logic dictates that all you need is some way to empty it onto the glaciers to speed up your Internet connection.

67 Tb ?

[kinsoa]

the system has 16 discs, for now we don't have 4Tb harddrive, so you can forget your 67 Tb.

And I have build a such system myself, it's really not new.

only 67Tb in 13 units?

[OrangeTide]

I dunno. 67Tb in thirteen 3U 16 drive units doesn't sound all that impressive. Maybe if you could fit 100Tb in 50U of space I would be impressed. If this could even scale that high you could only fit 80Tb in that amount of space. 3U for 4.8Tb of raw storage is not a big deal. You can build your own low quality system with that kind of capacity yourself out of cheap disks. Obviously not with the same performance though.

Although I will admit that this is a very fast product with decent redundancy. Although I generally believe dealing with redundancy at a higher level with software is much more flexible than controller level redundancy. And cheaper.

Fibrechannel drives sound neat and all, but if someone can fit 3x as many "lower end" drives in the same amount of space that's lower cost, higher redundancy, higher capacity and higher performance. I'm sure they are good for something though, else IBM wouldn't have such a sales drive behind them. *snicker*

Re:only 67Tb in 13 units?

[miyako]

It's really about the performance. Your correct in that anyone can build a system with the equivilent storage space cheaper.
I have a few TB of storage on my own network, and it's great for archving stuff, but it would be crap for trying to use this for storage on a high load server, that's the situation that these will be useful in, a good amount of storage with good performance, from a well known vendor. Especially for cases where a business already owns an IBM server, and want to ensure compatibility and keep a single vendor.

Re:only 67Tb in 13 units?

[Anonymous Coward]

67 TB within the space thirteen 3U enclosures is 67 / (13*3) ~= 1.7 TB per U. So in 50U, one can fit 86 TB. Putting it another way, to fit 100TB, the system requires 58U.

So basically what you're saying is "This would be impressive if they could fit 100TB into 8 fewer U spaces."

...someone can fit 3x as many "lower end" drives in the same amount of space that's lower cost, higher redundancy, higher capacity and higher performance.

From TFA: The DS6000 series is equipped with four paths to each HDD and uses storage expansion enclosures with redundant Fibre Channel switches providing switch access to disks.

I guarantee that anything you make from consumer grade parts will not touch the performance or redundancy of this system.

Re:only 67Tb in 13 units?

[Anonymous Coward]

To revise my above comment, the 3U cabinet is 5.25" tall by 18.8" wide by 24" deep. That's 2368.8 cubic inches. 4.8TB / the volume of the cabinet ~= 2GB per cubic inch. I'd say it's pretty damn impressive.

Re:only 67Tb in 13 units?

[Anonymous Coward]

well.. a Maxtor DiamondMax10 is 4x5.8x1'' and it holds up to 300 GB, which makes... yep, 12GB per cubic inch.

Re:only 67Tb in 13 units?

[Anonymous Coward]

I'm sure that harddisk can do very impressive things with no power hooked up to it. Include an ATX power supply (6x3.5x5.5 inches) for it, and it's suddenly 2.16GB/cu inch. This does not account for any controller interface for the drive.

Re:only 67Tb in 13 units?

[Anonymous Coward]

Maybe if you could fit 100Tb in 50U of space

I can already tell you don't have modern enterprise experience, even with your low uid, and therefore reduced the rest of your comment to "-1, Speaking out of his ass". Sure enough, as I glance through the rest of your post, more immature posturing by an inexperienced hobbyist. Go back to your xbox, now.

If you did have the experience, you'd know that standard racks are 42U and 48U (depending on whether or not you want to lose a few U to the UPS in the bottom and the KVM). Don't bother replying, I'm not secretly tracking this anywhere. I just wanted to let you know what us in the working world see when we read this stuff.

Re:only 67Tb in 13 units?

[Anonymous Coward]

sure. although I don't suppose they mount one ATX PSU per disk - it's more like "some disks+PSU+control unit". still, all that stuff put together is pretty impressive, no doubt :)

Re:only 67Tb in 13 units?

[OrangeTide]

four SATA drives on four different controllers is more redudant than one fibrechannel drive to two switches to two controllers. It's as simple as that.

An open question.

[Temporal]

If only there were some sort of visual stimuli -- say, something which appeals to our most basic primal instincts -- which could be stored on such a device, and subsequently accessed whenever one is bored and no one is watching. Alas, I am unable to imagine anything suitable. Perhaps one of my fellow Slashdotters has an idea?

Re:An open question.

[mvdw]

I can see it now: 4 Ask Slashdots in the next week: "I have 67.5TB of online storage, what do you guys use to back up all that data?"

Re:An open question.

[zoeblade]

If only there were some sort of visual stimuli -- say, something which appeals to our most basic primal instincts -- which could be stored on such a device, and subsequently accessed whenever one is bored and no one is watching. Alas, I am unable to imagine anything suitable. Perhaps one of my fellow Slashdotters has an idea?

Pictures of yummy doughnuts?

Re:An open question.

[Anonymous Coward]

Screenshots of /. reloading?

Re:An open question.

[R2.0]

Please add a constraint such that said data must be fully accessible using only one hand for input devices while the other hand is otherwise occupied.

67.2TB

[TheRealStaunch]

67.2TB should be enough for anyone!

Re:67.2TB

[Anonymous Coward]

67.2TB should be enough for everyone!

Re:67.2TB

[Anonymous Coward]

Now that we have reached Terabytes, it's time to move to a whole different nomenclature for size. I am currently using SLOB for all my computer storage size describing needs. SLOB stands for Shit Load Of Bits and can be used for any size at all. Enjoy. Yes, thank you. You are too kind. No, no, please sit down. Okay, I'll accept the roses, and the check, and the trophy. Your wife? For the evening? Well, if you insist.

No quantum leap, but ...

[foobsr]

... are designed to deliver a generation-skipping leap ...

They even know how to make use of proper wording. No "quantum" here (presumably because IBM has some background on the real thing).

CC.

Expensive logo?

[Doc Ruby]

Actually, the DS8000 is marketed as expandable up to 192TB. Since it's marketed as starting at 580GB, and priced starting at $97K, that's about $167:GB. Considering that a single 160GB drive, without redundancy, integrated POWER uC and other server hardware, IBM support or management software costs about $0.50:GB, and probably less in quantities of 1200 (==192TB), are those extras worth it compared to rolling your own RAID?

Re:Expensive logo?

[Doc Ruby]

OTOH, the DS6000 takes 300GB SCSI drives. 192TB is 640 300GB drives, which cost at least $197 in EIDE, for a total of $126K. While SCSI drives cost well over $500ea at 300GB, though about $1:GB at 147GB. If a loaded DS6000 costs anywhere near $325K, IBM really has great prices at the high end. That's about 768K FLAC'ed CDs, which would cost $15.4M to buy at $20ea.

Re:Expensive logo?

[roror]

yes, especially because, "No one gets fired for choosing IBM" while if you build your own RAID, you might get.

Re:Expensive logo?

[zbuffered]

That's 88 years' worth of music. Finally, something that can store enough music so that I don't have to change disks every 50 years!

Re:Expensive logo?

[Doc Ruby]

That's a lifetime of CD quality earsdump. Finally something to carve my interactive tombstone on!

Re:Expensive logo?

[guacamole]

When I hear someone suggest to roll your own anything, I want to scream and run as they probably haven't worked a day in a real production environment. I'd like to see you roll your own, manage, and support a multiterabyte storage system and then decide by yourself whether it was worth it or not (assuming you're lucky and get a chance to do so, after not being fired because something has gone wrong and ate your data or caused downtime)

As for this particular case, this system was obviously designed to efficiently manage vast amounts of storage. It is not worth buying if you only need a 580GB of storage. Besides, no one pays the list price in the enterprise storage market. No one also buys IBM's enterprise hardware just because they think they need the hardware alone.

Re:Expensive logo?

[F'Nok]

make install, not war

Nice sig, but did you consider this change?

make install --not-war

Wouldn't want the command to be rejected after all. ;)

Re:Expensive logo?

[swordgeek]

Yes. Yes, it is.

Now go back to your parents' basement, and pretend you have a clue about enterprise environments. In the meantime, read these hints.

SO YOU WANT TO MAKE YOUR OWN ENTERPRISE RAID SYSTEM
(from 'clues for the terminally clueless')

First of all, get your hardware right. SATA doesn't cut it in the real world. Modern SCSI is acceptable for small to medium systems, but large-scale is FCAL all the way, and that costs money. Count on $1800/146GB drive, or $12.50/GB. Add cabinets ($10,000/14-disks or about $5/GB) and you're already up to 12% of IBM's LIST price, or about 20% of what you'll actually pay for a small system. Now consider that IBM's price will scale up quite nicely. $97k/580GB is the (listed) entry level, but additional storage on an existing system will probably run about half of that figure.

Now add RAID controllers, computers, networking gear, redundant power supplies, and all of that good stuff. You're going to end up spending about $25/GB. From IBM, you'll probably spend about $45 for a decently large system (figuring 40% discount on the price listed). For that (still significant) cost savings, are you willing to bet your entire company's future, including guaranteed hardware and software support for the next three years?

Enterprise gear is expensive. Enterprises need that gear. It's not difficult once you've actually worked in a large shop, but kiddies playing at home still think they can do better for a fraction of the cost. Give it up.

Re:Expensive logo?

[Twylite]

When you talk about SATA, SCSI and FCAL, are you meaning the interface or the drives themselves?

There are numerous cheaper products out there that stuff PATA or SATA disks into a rack chassis and provide a SCSI and/or Fiber interface to the collection as a single logical drive.

On the other hand I have seen a fair amount of information on speed and reliability of SCSI drives versus IDE drives, and it all points to SCSI being a better choice.

I'd imaging that a "cheaper" solution that uses PATA/IDE disks will end up costing more than expected because (1) performance won't match the IBM system, (2) you'll be replacing individual drivers more often, and (3) the reliability of the "cheaper" chassis & electronics won't be up to IBM's standard so ultimately you'll take more downtime.

Is that it, or is there something more I need to know?

Re:Expensive logo?

[Antique Geekmeister]

The trade off has been shifting. I haven't seen any FC-AL or SCSI hard drives over 80 Gig actually in any machines yet, but SATA drives of 400 Gig are commodity items you can buy at your local computer store. At that price, it's a lot cheaper to have a lot of redundancy in cheap hardware than run the higher speed SCSI/FC-AL. And if you've ever tried to wire SCSI inside a tightly laid out rackmount box, you'll see where both SATA and FC-AL are huge advantages in wiring and cooling. But FC-AL is grossly more expensive than SATA, so SATA comes out looking pretty good. With reasonable amounts of cache, you're really not worried about the larger 8ms seek times of the 7200 RPM SATA vs. the 5 ms seek time of the SCSI. You're potentially concerned about streaming data, which matters if you've got a set of big local Gigahertz Ethernet connections pulling huge amounts of data at top speed, but few of us do that, so few of us are willing to pay the price for SCSI moving forward.

Re:Expensive logo?

[Doc Ruby]

When you've got a $100K+ budget for a rackmounted box for storage, in addition to the salaries of its administrators and people to fill it, you can "roll your own" pretty well. Especially when you're in whatever cutting edge biz generates 192TB. You're probably building lots of your own custom apps, rolling your own TCP/IP (or other) network. In a few months we might hear about IBM's customer list for this new product. Those customer's competitors are probably rolling their own. Several months after that we might get to hear which switched in favor of these products. And maybe get a chance to buy the dumped tech on eBay :).

Re:Expensive logo?

[Doc Ruby]

Thanks - good insight. Poetry that parses in both natural and artificial intelligence is worthy. Though I'll have to consider whether that revision will be as popular among the machines as the familiar version has been among the humans :).

Re:Expensive logo?

[Anonymous Coward]

I haven't seen any FC-AL or SCSI hard drives over 80 Gig actually in any machines yet,

My SAN uses 146GB FC-AL disks, and Sun's newer servers like the v890 can come with these too. A few months ago our SAN vendor said there were issues with the > 146GB FC-AL disks, but didn't elaborate.

which matters if you've got a set of big local Gigahertz Ethernet connections pulling huge amounts of data at top speed, but few of us do that,

Since we're originally talking about a SAN here, and some people don't understand the underlying concepts, it is important to point out that a SAN may be moving data at at least 2Gbps to an HBAC in a server. If you have a nice fabric set up, then you might be pushing 4 (or more) times 2Gbps.

For new people:
A SAN has fiber connections, usually 2 or more. These fiber connections run either directly to an HBAC (fiber controller card) in a server or to some fiber switches (these switches are a little different than your network switches), and then from those switches to the HBAC's. The setup with switches involved is called the fabric, the other is direct-connect.

If you have the money for a large fabric, and your SAN can support it, you could potentially have 100's of servers accessing your SAN, and to each server its space on the SAN will look like a local disk or two or more.

Since you potentially have many servers accessing many logical disks on many physical disks via many switches and many HBAC's, you need strong management software to handle the layout, as well as very fast disks and switches/HBAC's to handle the performance, as well as dually-redundant everything, like switches, HBAC's etc, and if you really want to be redundant, then a second SAN in a second location, which creates more management.

So, it is important to have very very good software that you can understand and use easily and quickly, and that anyone else coming in can use easily and quickly, and that is very, very well supported. This is a major difference between roll your own and a SAN that you buy.

Next week i hope someone brings up a new virtualization device on Slashdot.

Re:Expensive logo?

[Anonymous Coward]

"..the DS8000 is about $167/GB, while normal hard drives are around $0.50/GB - are those extras worth it compared to rolling your own RAID?"

B..but it's from IBM!

Re:Expensive logo?

[swordgeek]

I was talking about the drives. We use SATA drives with an FCAL back-end connection for first-tier backups, but never for user-facing data. ALL user data comes off of SCSI or FCAL drives running in a fibre-attached disk system. Of course, we're a moderately large shop--roughly 35TB of front-end data.

Comparing SCSI vs. IDE, the former will provide better reliability and FAR better performance for multi-access scenarios (which is essentially everything except one-machine one-disk one-user setups).

For certain situations (large amounts of low-access data, or nearline storage) ATA works. For nearly all enterprise applications though, it's the wrong solution from a performance, availability, and reliability perspective.

Re:Poll Troll Toll

[Bi()hazard]

To reach solid decision's, youl'l need more infermation then the slashdot writeup supplies. Like this article featured on linuxtoday.com, which are surely slightly more independent than IBMs' press release's; (click complete story under the summary) From it:

The DS8000 is unique in the industry because it features two logical partitions too run management or utility applications such as the companies SAN Volume Controller and Tivoli Storage Manager for backup and data management.

That sounds like a pretty interesting feature. Anybody's in the industry care to comment on the portential for these new development?

This article on lightreading.com elaborates a little more.

IBM's DS8000 handles virtualization different then the competition. While HDS does virtualization in the controller and EMC plans virtualization on intelligent switches IBMs' new system does virtualization at the chip level (see EMC on Virtualization: Wait for Us ). Using the Power5's IBM Virtual Engine, the DS800 can divide servers into logical partitions (LPARs). Each LPAR can run different storage systems that run separate code. ... "You can run different operating systems, even different releases of operating systems on isolated LPAR's. Rock!"

Thats a truly impressive level of flexibility their. And of course, its great for Linux, the ability to run multiple OSe's in hardware on one box play's to Linuxes strength's and deal's a blow to Microsoft's monopoly lockin strategy. What Im really shocked about is that there slashdot writeup included only some bland "durr big numbers" product placement, while IBM is effecting an interesting Linux-related change's in the marketplace's if you look a little deeper.

--sig: why a duck?

Re:Poll Troll Toll

[a.different.perspect]

Why did you reply to the poll troll? Your comment has nothing to do with the poll troll.

Re:Let me guess

[Anonymous Coward]

Well, duh.

Linux rules: 3u dual AMD 64, 4GB, 6x250GB for $5k

[Anonymous Coward]

I just spec'd out a 3u high enclosure that has 12 drive bays, 4GB memory, dual AMD 64 processor, dual processor motherboard, 2 mirrored system disks, raid controller, populated with 6 x 250GB drives. All this for around $5k. I like populating with half the drives so I can easily pop in the next gen drives. Oh and don't forget Linux. And don't forget, its 64 bits. yeah!

I think the price for an ideal configuration is 16GB, 12x400GB for $12k.

WhatMeWorry!

How many partitions?

[Anonymous Coward]

Ok, call me crazy, but there's one thing I want to know: how many RAID sets, and how many partitions, can this thing handle? Right now, where I work, we have what IBM used to call a Fast T storage array (we're busy rolling it out at the moment). One thing which may impact in the future is the limitation on the number of partitions that the controllers can handle (aka LUNs); IIRC, it's 64 (might be 128, I can't remember for sure.)

I've also had experience with FC setups which have a limit on the number of RAID sets you can have hooked off a single controller -- typically around 16. Now, if you don't have a lot of disk drives, that's not a problem, but if you want good redundancy, and you have (say) 200 hard drives in the set, it is: you don't want to have the whole damn thing in one massive RAID 5 array and suffer a huge performance hit when one of the disks dies (let alone what happens if two disks die at nearly the same time -- don't laugh, it happens more often than manufacturers would like to admit.)

Yes, performance and capacity are important. But so too are things like this that you don't think about asking about until you bump into the limitations. Most people will happily roll out huge chunks of disks for their databases and so forth, in which case this isn't likely to be an issue. But -- depending upon the circumstances -- you need to know what can bite you down the track so you can plan for it.

Don't get me wrong -- there are several ideas in this stuff that look extremely interesting (not least of which is the prospect of being able to do backups without involving the servers using the disks at all) -- but you do need to lift the carpet and see what's been swept underneath before you buy.

Re:How many partitions?

[Harassed]

Just a quick one - Partitions != LUNs.

A LUN (logical unit number) is specific to the host and is effectively the physical disk number. The number of LUNs supported is very much dependant on the OS (Windows/Solaris support 256 and Linux supports 128 due to RDAC limitations currently).

Storage partitioning is configured at the storage device level and is a logical grouping of logical drives, host groups and hosts to control access and improve performance. The number of partitions supported depends on how much you want to pay IBM :) A standard FAStT600 supports 1 storage partition, a FAStT600 Turbo will support up to 4 and a FAStT900 will support up to 8 iirc.

The IBM FAStT range have just been renamed to DS4x00 (the 600 is the DS4300, 700 is the DS4400 and 900 is the DS4500). Saying that, they are all rebadged LSI Logic (Engenio) devices anyway. The DS6000 and DS8000 products are basically a refresh/replacement of the old ESS product set.

You mention the ability to do server-free backups - this can also be done on the FAStT (DS4x00) products if you pay the extra for the Premium Features such as Flashcopy, Logical Drive Copy and Remote Mirroring which are available on the 600 Turbo, 700 and 900 models.

Alex (currently actually on an IBM DS4x00 training course!)

Re:How many partitions?

[Anonymous Coward]

Oh-boy. Are you in for a fun time. We have a 200TB SAN of Fast T and it is nothing but hell. A single drive fails but the controllers report multiple drive failures, rebuilds take and entire day for one stripe (admittidly the stripes are 800GB in size), etc..

Sure, it's cheap. But you get what you pay for...

Re:How many partitions?

[Anonymous Coward]

Not a bad explanation, but the numbers are a little off. A FAStT600 comes with 1 partition standard, and is upgradeable to 16 maximum. A FAStT600 Turbo comes with 8 partitions standard and is upgradeable to 64 maximum, a FAStT900 comes with 16 partitions standard and is upgradeable to 64 maximum. There are a maximum number of LUNs that are supported per each controller type as well: a FAStT600 has a maximum of 512 LUNs, a FAStT600 Turbo has a maximum of 1024 LUNs, and a FAStT900 can handle a maximum of 2048 LUNs.

Re:How many partitions?

[Pinback]

The FastT subsystem is remarketed LSI aka Symbios aka NCR. (Yes, as in the old SCSI card maker NCR.) FastT has a really cheap heritage, and only supports active/passive failover like other low end products like the EMC Clariion line.

For a better feel for the DS line, you have to look at the feature set of the ESS (shark) line.

The sharks have two pSeries boxes in them that act as an intermediary between the FC (fabric) host-adapters on the front end, and the IBM SSA disk loops and trays on the back end. These servers also handle caching and RAID functions for the SSA disks.

(The shark line was backended on SSA disk. But they did go to SCSI disks with an SSA-shim around the time they sold their disk business to Hitachi.)

The DS8000 appears to be based on FC-AL on the back end, much like the EMC DMX line.

In addition to the software based features of the ESS, some models in the DS8000 lineup will have enough spare CPU to host an LPAR. Presumably this LPAR will be used to run the SVC virtualization software. (And who knows what else? Maybe TSM some day?)

If the pricing is good, it could be a compelling product to VAR.

Re:How many partitions?

[spell]

Well, I've been ploughing through the DS8000 planning guide as it is very likely that I'll be putting at least one in our datacentre by the end of the year. The DS8000 can handle 67.5k LUNs and over 500 host log-ins per FC port. This should be enough for most people; the DS8000 architecture is designed to scale to multiple Petabytes...it'll be interesting to see EMC's response.

Re:How many partitions?

[Harassed]

From past experience, if they can come up with something it will perform much worse and then they will just refuse to release meaningful benchmarks and ban their resellers (i.e. the likes of Dell) from doing so either. If you look at the Storage Performance Council website, you will notice a lack of EMC benchmarks you will notice one glaring omission - EMC! This is because their performance wasn't up to spec and they threw their toys out of the pram...

Re:Poll Troll Toll

[Anonymous Coward]

Finding a bead that support the mare while you mount it may be difficult!

Re:insert witty pr0n comment here

[pchan-]

i'm going to say this slowly so that you can follow along:

porn ... never ... gets ... old.

LAN party

[Wookie_StarFire]

I wonder how people would feel if I'd take this baby to a LAN Party and download all the stuff from each and every one of those two thousand PC's. Would people still like me?

Backup 4TB?

[soundman32]

And how the eckle feckin do we back that baby up?

Re:Backup 4TB?

[cakefool]

punch cards. lots and lots of punch cards...

Re:Backup 4TB?

[Wookie_StarFire]

Or get another baby to backup the baby...????

Re:Backup 4TB?

[HoneyBunchesOfGoats]

I'm sure IBM would be happy to sell you more than one.

Re:Backup 4TB?

[keesh]

LTO2 tapes are 200GBytes each... Remember that these boxes can flashcopy (instantly do a complete copy of your data, kinda like LVM snapshot support but actually working and a hell of a lot more powerful, oh, and done in hardware), so you don't need to stop your database whilst you're waiting to write it to tape.

Re:Backup 4TB?

[Anonymous Coward]

At that size even ultra-320 scsi or fiber channel looks like a very small pipe.

Also, snapshoting the lvm volumes still requires the databases or filesystems to be checkpointed into a consistent state. Not only is that non-trivial but it all has to be done at once, including the apps that use that data. Restoring files and a database to the point in the middle of a massive application update of the same isn't feasible.

We're way beyond the point where conventional backup solutions work anymore.

Re:Backup 4TB?

[Anonymous Coward]

Depends on the type of backup that you want to do. You could do a backup to tape using either LTO or 395x drives. You could do a FlashCopy if you just want to have a spare copy on the floor for any number of reasons. If you need to constantly have another copy off site you can use PPRC (in any of its forms) or XRC to another DS8000.

Re:Backup 4TB?

[keesh]

FibreChannel Switched Fabric can do 10GBit full duplex speeds *per link*. The ESSM800T could take sixteen of these links. I've not had a chance to work with any of this newer kit yet, but I'd imagine it's even better off connectivity-wise.

Re:Backup 4TB?

[rhaig]

backing up 4TB is almost trivial, I handle almost 5 now. The more interesting case would be the full 67.2 TB

ok, so if you want 12 weeks of retention, and do nightly incrementals and weekly fulls, 67.2 TB would require about 952TB of tape capacity.

(figure 3% daily change * 6 incrementals/week + 1 full/week * 12 weeks, so 14.16*disk=tape)

so if we round up to 1PB of tape, and let's assume LTO2, with 320GB/tape (about the numbers I'm seeing for binary data), that's roughly 3200 tape slots + 260/week for offsite copies. Call it 3500 slots.

Since I know the storagetek line of tape backup products, I'll talk about them...

you could start with 3 L1400 libraries with 10 drives each, and it would take you less than 24hrs/week to back up all that. And duplicating the tapes for the offsites another 24-48.

the libraries and drives would cost you near $500000, the tapes (assuming an 8 week offsite rotation, and buying an extra 10% for the partially used tapes that will always be present) would cost you about $340000.

now I know Veritas netbackup is ungodly expensive, but I have the prices for it in my head, so I'll quote that too...

for 30 drives handling this much data, I'd probably put in 6 servers. ($10K each) Veritas licenses drives ($3k each) for a software total of $150K.

Throw in some fibre switches to mesh it all together, the cost for the servers, (IBM 345s with 2 fibre cards and a couple of 146G scsi drives each: about $50K) and the total ends up being about a million bucks. ($1.04M)

Go with something like Tivoli Storage Manager and that 150K software cost drops to more like $15K (they don't license on a per drive basis)

want to build an enterprise level backup solution... there you go.

Re:Backup 4TB?

[Pinback]

The native capacity of 3592J1A (jaguar) drives is 300GB. If you backup in parallel to 20 jag drives at approximately 200GB per hour, you can backup 4TB in just about an hour.

Re:Backup 4TB?

[spell]

Well, as we back-up that sort of volume of data on a more or less nightly basis; it is not as hard as you might think. Put the database into hot-backup mode; start the flashcopy and then bring the database out of hot-backup. We do this to multiple targets as well, lets us build a reporting database and do lanless backups etc. We've been doing this for sometime now. Okay, so you need to spend some money; at some point next year, I'll probably do a tech refresh on our tapes and go to 3592; this gives me 300 gig per cart native, with compression I can get a lot more. It also gives me a road-map which takes me to a terabyte per cart.

The biggest problem is potentially logical corruption in the database; so to get round that, we run a trailing database using dataguard. This database currently lags the main production database by an hour; there is alot of debate whether it actually needs to be that far out but an hour seems to be about right at the moment. I also offsite a copies using PPRC to remote mirror to our DR site; this introduces a small amount of latency due to having to wait for the write to be committed and acked from the remote site but with PPRC over FC, we have found that the latency introduced is only slightly in excess of the latency introduced by the speed of light in fibre; we have let IBM off breaking the laws of nature, that might have been unreasonable.

Is this a conventional solution? In my experience, these days, yep. It is a fairly standard solution.

Re:Backup 4TB?

[Antique Geekmeister]

You back it up with a 2-stage system. First, you dump to a Quantum "disk tape library", where they configure a big old RAID set to pretend it's a tape library, just to get your data offline ASAP. Then you clone your tapeset from that to a real tape library or set of them, such as a Quantum 300 tape automated library system stuffed to the gills with the new 600 Gigabyte compressed SDLT style tape drives. I priced a set of these recently to explain the costs of a proposed backup policy. They're not cheap, but since each such tape can hold 300 Gig uncompressed and you can stuff in 300 of them into such a library, you can dump 90 Terabytes or so without swapping tapes. And since each drive can handle Ultra320 SCSI, figure packing it to the gills with 12 tape drives each on their own controller yields roughly 12 * 320 = 3.840 Gigabits/second, or something on the order of 300 Megabytes/second. For the 90 Terabytes, that's roughly 300,000 seconds, or at 86,400 seconds/day, roughly 3 days to dump the whole thing assuming everything is dumping flat out. You did ask.

Re:Backup 4TB?

[Anonymous Coward]

Periodic FlashCopies in the remote site are always advisable as one method of recovering from a corrupt database. Alternatively you can always use PPRC-XD to take your PIT copies in the remote site periodically instead of having the norm be taking a synchronous copy.

Not the first vendor to offer this...

[jtharpla]

We've been getting disk arrays like the DS6000 for months now... for example:

RocketSTOR R2221
or
Silicon Mechanics SM-316RX

Re:insert witty pr0n comment here

[Anonymous Coward]

sorry, but the number of people who are into old people porn is much lower than the number who like to look at young cheerleaders...

Re:Poll Troll Toll

[Anonymous Coward]

indeed, that was my first thought...
It takes one to know one?

Re:insert witty pr0n comment here

[Tezkah]

I dunno man, after listening to William Shatner's new album, I dont feel the need for porn.

after reading the spec sheet

[halaloszto]

The 3U box can only hold 16pcs of 300G disks max, that is about 5T. The 67T can only be achieved with additional boxes of course. v

Over here in the old world

[Anonymous Coward]

125 pounds is 57 kg...

Re:Poll Troll Toll

[Kethinov]

To get his above the other threads.

LOC sighting

[quintessent]

The enterprise storage system, which is available in either dual two processor or dual four processor configurations, includes an architecture that can address over 96 petabytes of data - or more than 4500 times the amount of information found in the Library of Congress.

DS6000 -Cool technology, but price is tooooo high!

[zorang]

Good: - Robust technology - Modular - IBM support Bad: - Expensive - Only 2 GB of cache (mirrored) - Slow, check out http://www.storageperformance.org/results

Remember....

[Anonymous Coward]

Joke smart, and rotate those punchlines every ten thousand posts. A failure to act responsibly may put both your freak list and your karma at risk.

Rack Mathematics

[WillerZ]

One DS6000 is 3U high and stores 4.8TBytes.

One rack is 42U high and can therefore contain 14 DS6000 units.

So the statement of 67.2 TBytes/rack is correct.

Phil

One word

[boots@work]

Google.

Re:One word

[Tony-A]

Google is a completely different animal.

Google itself is ultra reliable so long as most everything is working kinda sorta well. Something breaks and Google just researches the web, which it was going to do anyway. Google can function perfectly well with lots of its components broken. Almost nobody else can.

Re:One word

[TTL0]

Google can function perfectly well with lots of its components broken. Almost nobody else can.

right. That is why Google should stop being the poster boy for "look what you can do w/ Linux and lots of cheap hardware"

Re:One word

[happypork]

Yes, but what about Gmail? If they lose my messages I'm fucked and royally pissed because Google is the only one storing my messages. They can't just recrawl the Internet to bring 'em back.

In other words, whatever system Google's using for Gmail has to be rock solid -- otherwise their e-mail system is shit. If they're using their commodity-computer storage array for Gmail, that tells me it's actually pretty damn good at retaining information.

Re:One word

[Tony-A]

That is why Google should stop being the poster boy for "look what you can do w/ Linux and lots of cheap hardware"

Why? It is precisely the openness of Linux that allows them to tweak/mutilate/whatever the system so that their Linux serves Google's purposes rather than Google serves Linux's purposes.

Re:One word

[Tony-A]

that tells me it's actually pretty damn good at retaining information.

Assuming Google actually knows what they're doing, and that seems highly likely, it will take more than one simultaneous disaster for them to lose information. At all times, all the information exists in more than one form in more than one place. A bit of heads-up and redundancy and it is impossible for a single failure to wipe out everything. Whereas a bone-headed operator and a sophisticated backup system and a single indication of failure does give fairly good odds for wiping out everything.

Re:One word

[boots@work]

Sure, they're a different animal. Every business is different: banks from airlines from oil from mum's embroidery shop. Google are far from the only organization to be using farms of PCs, but just the most prominent, and perhaps the most ambitious.

As with everything else today, Google is designed to withstand component failures. You cannot buy hard drives/memory/processors/data centers that won't eventually fail, so you design around it. People seem to have the superstitious idea that there is something dodgy about counting on redundancy to be reliable, but it is often the only feasible design.

If their whole system went offline, Google would be as unhappy as any other company. By various estimates it takes them days or weeks to crawl the whole web, and probably some additional time to build the index. If their store catastrophically failed, they'd be gone for weeks, which would be very damaging.

Re:One word

[Anonymous Coward]

I heard they use NetApps. At least, a friend who has space in a colo room Google recently moved into said they moved in with a shitload of NetApps.

mod parent down, -1 misleading

[WillerZ]

I did.

That page has no results from the DS6k or DS8k.

Phil

Explanation

[boots@work]

Google have a hand-rolled filesystem and hardware, supporting several petabytes of data. (Exactly how many is anyones guess, but reckon on tens of thousands of servers × hundreds of gigabytes each.)

Apparently they're pretty happy with it.

Re:My question...

[boots@work]

It probably has Linux inside, seeing as it just has PowerPC processors. Linux would be the obvious platform for the "host" part of the system, running the web/telnet interface or whatever. The hard work is probably done by some non-Linux embedded code on separate processors.

My advice is to wait five or ten years and get one at a fire sale, then pull it apart.

Re:insert witty pr0n comment here

[Anonymous Coward]

Yes, it does get old. I saw a vid yesterday of two 80-year olds gettin it on.

Pr0n gets old. And shriveled too. Viagra has brought on a lot of new possibilities - many of which I don't want to even think about for 50 more years.

f#ck google

[Raindeer]

No matter how much google stores, it is not the one to look at when you're talking corporate data storage. Corporate datastorage is about storing all the data of all your oil fields, in a way you're sure you don't loose it. It is about storing every single product that you make in a database, complete with tracking of location and which customer bought it. It is about all those things Google doesn't do and doesn't care about. I am willing to bet that for its financial system Google is using similar to the one shown here. Why? because it is reliable.

Google is using of the shelf hardware, because it doesn't matter to them if they loose data because of disk failure. As long as it doesn't happen too often and from the perspective of the customer doesn't matter, it is not a problem.

Now think of google having to have an accurate and 100% corrrect archive of the internet, which has to be searched and correlated 7 years back and then see what they would come up with.

Re:f#ck google

[Antique Geekmeister]

You forgot DNA sequencing. The big data fast data storage market right now is in genetic work, which trivially blows away any typical corporate industrial database application.

Re:f#ck google

[Raindeer]

ha, I think I can top that when it comes to research. Look up a project called LOFAR. It is a distributed sensor array for astronomy, basically creatin a huge radio observatory of a couple of hundred kilometers in The Netherlands. The amounts of data those guys talk about is measured in tens of gigabits per second, continuously for years and years and years. Big Blue is building the supercomputer for that.

But your right DNA sequencing, Biotech, but also Medical Imgaging demands huge huge amounts of storage

Re:f#ck google

[paitre]

And we have no problem spending tens and hundreds of thousands on storage arrays that we know Will. Not. Fail.
COTS doesn't cut it when you have data worth billions, guys.

Re:f#ck google

[legirons]

"Corporate datastorage is about storing all the data of all your oil fields, in a way you're sure you don't loose it. It is about storing every single product that you make in a database, complete with tracking of location and which customer bought it."

Actually, we use Word documents for that. Lots and lots of word documents. On Windows servers with blank administrator passwords.

Don't you just love the corporate IT myth of everything being neat and organised, and running suitable good-quality software?

make love to google

[boots@work]

If you actually read the link, you'd see there is at least as much redundancy designed into the system than in most NAS systems, and it has been very reliable to date. You are familiar with the idea of RAID, aren't you -- Redundant Arrays of Inexpensive Disks? This is the same approach as in the IBM hardware, but at a much higher level.

For example they maintain integrity checks of every block, to catch silent corruption. This is not done by many competing systems -- it is a major selling point of Sun ZFS that they do.

The primary reason why Google don't use this for their financial systems is likely that it is custom designed for their search applications, not for whatever financial systems they use. Secondarily the volume is so small that an off-the-shelf system probably works fine.

Do I expect everyone to build their own system? No, but for some users it works well.

(Why are people happy to use the thought "fuck", but not the letters? Bizarre. And what is this loose data you speak of?)

Re:make love to google

[Raindeer]

If I'm not mistaken there is a filter on certain words here... So I bleep stuff out. :-)

Loose data is what you get when English is ones third language, typing is quick and thinking is slow. It is called a spelling error.

Re:make love to google

[ckaminski]

you are mistaken... fuck fuck fuck fuck fuck.

It's because the OP who F#CK'd google wasn't ready to go the final mile and write FUCK google, for whatever reasons he has in mind, which is what the GP was pondering about.

Re:make love to google

[Anonymous Coward]

Like fuck there is..

Naming...

[imsabbel]

Most often Raid5+1 means a Raid5 Array +1 hot spare (to minimize critical array time in case of a disc failure)

48U racks

[Robbat2]

http://www.server-rack-online.com/48u-90--server-r ack.html

Good when you get given a crazy retro-fitted server room with stupidly high ceilings.

Re:insert witty pr0n comment here

[Guillermito]

porn ... never ... gets ... old.

Oh! Is it that so?

Prices

[Yartrebo]

Sure it might be cool to jam 16 HDs into such a small space, but for about $100K, it sure seems a bit pricey. Couldn't you get maybe 2-4 of the most reliable PC cases you can get and stuff them with HDs.

Or is there some super-important feature that I'm missing.

Re: Prices

[Gothmolly]

Or is there some super-important feature that I'm missing.

Yes, there is.

Re:Prices

[Anonymous Coward]

As previous posters have mentioned you don't pay the list prices, it's cheaper.

What do you plan to do with the 4 PCs that you need to maintain for each unit? And then you have a rack full of units, so to replace that, lets see, thats another 48 PCs or so. I don't know any medium sized company that wants to put in another 50 computers in their server halls, do you?

The other stuff you're missing is support. IBM won't stand behind your homemade Linux distribution where you use the volume manager of this month to bundle disks together.

Products are best when they just work. Fiddling around with stuff might be amusing but your employer pays for your time too. I bet there are more important things to do than roll your own linux distribution to provide storage for your databases.

why moderates me as "troll" ??

[kinsoa]

the article said something wrong, I said it, and I'm a troll ?

and later, all comments are repeating what I've said and they received a "+5" karma :(

Re:why moderates me as "troll" ??

[rts008]

Somebody had to plow new ground? I'm with you...can't figure out the pattern myself.

Death Station 9000

[BillGodfrey]

Next in the series, the DS9000!

Bill, comp.lang.c

192tb

[Anonymous Coward]

Finally!

I can convert my 274,285 CD's to a harddrive!

(yes, that is 192tb)

The obligatory shot from the OS X administrator...

[profmathers]

Yeah, this new IBM stuff looks pretty revolutionary. It does have several features that I wish were incorporated in my xServe RAIDs that we've been running for almost two years...on both Apple xServes and Dell/Linux solutions... I hope that this doesn't come as a surprise to TOO many people. Gratuitous link here

Finally, Some REAL Storage

[jaklein]

I've been a mainframe programmer for close to 30 years now. I can remember a time (seems like yesterday, damn it was yesterday) where it was close to impossible to get more than 50 Mbytes of storage for files. I used to sit at my desk musing about attaching a 30 GigaByte PC drive to the mainframe so I could easily store large files. Its about time IBM came up with the technology to take cheap (?) PC type storage and use it on their zSeries computers.

Re:Finally, Some REAL Storage

[Pinback]

There is at least one third party vendor that does just this. (Sorry, don't remember the name at the moment.)

The ESS and DS series gear does have CKD sized blocks, and ESCON/FICON type connections. So they can be used with the 390.

Re:My question...

[Anonymous Coward]

Thats a nice conclusion. "I see a processor name I recognize, therefore it must run an OS I've heard of".

Shame

[oojah]

As a Brit, I'm somewhat disappointed that the writeup meant the other "pound".

The ONLY 125 pound storage device that will hold up to 67.2 TB!

I don't really need 67.2TB of storage, but at £125 I would certainly have considered it. £1.86 per TB is not a bad price (US$3.33)

Cheers,

Roger

Fast developers there at IBM

[Anonymous Coward]

Saw advert for the lead firmware developer two weeks ago, and it's done already, wow! Glad I didn't apply. They must have extracted the guys brains, ground 'em up, run 'em through a lyofolizer, then dialysis, then stuffed the extract into the chips.

Hey wait a minute, they are still advertising the position. Must have an order already!

Re:Linux rules: 3u dual AMD 64, 4GB, 6x250GB for $

[chathamhouse]

Ha ha, funny troll.

You seem to be confusing the market that this thing is targetted as. This is Storage Area Networking, normally applied to systems for who's downtime cost far more to the company than this disk.

First, it's fast disk. Fibre Channel drives, using 15 000 rpm (up to 146GB now?), or 10 000rpm (300GB) disks.

Second, it's expandable. Just add extra drive chassies on the expansion loops.

Third, and the reason people buy these, is that it makes managing storage for 10s to 1000s (DS8000) of machines simpler. Only allocate the amount you need, but grow it easily without the hassle of dealing with normally under utilized scsi system disks.

This equipment is for "big time", highly reliable, yet highly redundant computing. That costs money. Your suggestion is for cheap cheap disk - and you should be looking at someone like www.infortrend.com if you had a $5k budget and wanted SATA-to-SCSI. The dual AMD and 4GB ram is a waste in your config.

Re:My question...

[Anonymous Coward]

Power 5 != PowerPC. PPC is the consumer subset of the Power architecture.

Ma Bell

[Doc Ruby]

Your comments about enterprise gear are worth making. And that's a nice compliment to mistake the Manhattan skyscrapers of the giant, global banks and publishing companies where my network architectures run "my parents' basement". Would you ask Mom to turn down those Wall Street squawkboxes, an maybe fry up another $200M fixed-income coupon issue? There's a game on the compliance workflow archive server, and Rover has gotten down into the raised floor again.

Re:Ma Bell

[swordgeek]

OK, I just don't get it. On the one hand, you say you design network architectures for global banks; and on the other hand, you suggest replacing a highest-end storage system with crappy home-brew SATA arrays. The two most obvious conclusions are that you either don't know what you're talking about, or are trolling.

Do you really think that there's a chance of replacing ULTRA-high-end storage with SATA and making it work? Do you think that any money will have been saved by the end of three years? I can tell you now that the answer is no.

Re:Ma Bell

[Doc Ruby]

RAID stood for "Redundant Array of Inexpensive Disks", though the marketers turned that to "Independent". The idea is inevitable failure, planned as MTBF, with redundancy instead of pretended invulnerability. I merely suggested that the Dx000 economics are comparable to DIY, rather than a vacuum. Of course all that extra IBM tech, including their support (typically hotswap onsite replacements), has a lot of value. If you're confused because I'm saying that DIY is better than these IBM arrays, maybe it will help tht I'm not. I'm merely offering the comparison as we all discuss this interesting new tech in these threads. That's how we do it in the critical-path architecture biz. We don't jump to conclusions, and we consider many options as new solutions are introduced, especially by giant marketers amidst claims of "reinventing the market" or "revolutionary new economics".

Re:Ma Bell

[Antique Geekmeister]

There certainly is such a chance. I'm doing it regularly with Terabyte size arrays these days. Taking IBM's system as an example for comparison, a 3U box of 16 disks in SATA running Fedora Core 2 with built-in copper GigE and good quality controllers and drives is about $10,000 off the shelf. Sticking with 300 Gig drives, that's 4.2 Terabytes of usable storage. That is way, way more than enough for most smaller companies these days, and can also be their UNIX login manager and Windows Primary Domain Controller with a bit of installation work. Voila. No extra boxes, no racks of equipment sucking huge power, and you can use the spare money to plan the next upgrade replacement with instead of blowing all your money on a system that's obsolete in a year anyway.

Even worse than you think.

[RadRafe]

Look at those figures again. It's $97 000 for half a terabyte. That's insane! When the 1TB Xserve RAID goes for $6000, I don't care how much more redundancy this DS6000 unit has. I am not blowing more than sixteen times the dough for half the disk!

Insane!

Re:Linux rules: 3u dual AMD 64, 4GB, 6x250GB for $

[agallagh42]

"I just spec'd out a 3u high enclosure that has 12 drive bays, 4GB memory, dual AMD 64 processor, dual processor motherboard, 2 mirrored system disks, raid controller, populated with 6 x 250GB drives. All this for around $5k. I like populating with half the drives so I can easily pop in the next gen drives. Oh and don't forget Linux. And don't forget, its 64 bits. yeah!

I think the price for an ideal configuration is 16GB, 12x400GB for $12k."

Great, you just spec'd out a pretty nice server. Contgratulations.

That has absolutely nothing to do with this conversation though. This IBM system is not a server. It is for providing gobs of local storage to MANY servers. It's highly available storage, period.

What you could do is take all the storage out of the box you just specced, other than the system disks, and just pop in a fibre channel card (or two for redundancy). This IBM system now provides your data storage volumes, as well as volumes for the tens or hundreds of other servers on your SAN.

If one server is running out of space, assign it a little more from the free pool. If another server is not using all of it's space, take some back.

When you have all your storage in one place, it makes your whole server farm much easier to manage. All your backups are centralized. If you have 10 servers that need a pile more storage, you only have to upgrade this one unit, instead of visiting all 10 machines. Since it's designed to scale, that upgrade is very easy.

This is what you're paying $100K for.

Product vs Solution

[Roadkills-R-Us]

Everyone wants to sell me a solution. Solutions cost more, produce more profit, and tie me to the vendor. In most cases, I don't need or want their solution; I just want and need their product.

But there are times I want solutions, and solutions cost more. They come with uptime, top notch support, etc. When there's a problem, they often know it before we do, and notify us how and when it wil be solved.

For our compute farm and desktops we buy products. For our networked mass storage, we buy solutions. To date that's been NetApp. but it could easily be something like the DS6000.

At home I'll roll my own networked mass storage. At work, no way. I want uptime and throughput. I want redundancy and integration. I want someone showing up immediately, if not sooner, to resolve problem, which better be few and far between.

That costs more. Sometimes that's the right answer.

FAsT and Clariion

[Genady]

With a quick look I wonder if the FAsT is destined for EOL now. This looks to be taking aim directly at EMC's Clariion line (I've admined both).

If you really can maintain seperate install images inside LPARed storage processors that would be too freekin' cool. No more closing your eyes and trusting the vendor's QA when it's time for an upgrade.

Re:My question...

[reedk]

I'd bet dollars to doughnuts that it's running AIX 5.3.

how does this compare to suns 6960

[Anonymous Coward]

I know it can go to 65TB and has been available for a while: http://www.sun.com/storage/midrange/6000/6920/

Imagine ...

[Anonymous Coward]

... a Beowulf cluster ...

seems like a car!

[morcheeba]

The picture of the ds8000 looked small, like something I could stick under my desk. Then I looked at the specs - it's 6 feet tall. It also has
- 8 processors (power5)
- 32-256 GB RAM
- up to 640 disk drives
- 4 port 2gig FC (anyone know where I can get a USB-to-FC adapter?)
- weight: 2880 pounds (each expansion is 2400 additional)
- 30,000 BTU/hr. Converted, that is 8800 watts, or 12 horsepower!

Re:seems like a car!

[rts008]

Makes a dandy "smart" riding mower!

Wrong numbers!

[gweihir]

Using a single of these devices, you can get up to 6TB with 400GB disks. The 67TB figure is for multiple devices!

The weight figure is a bit on the low side, yet not that far off, but again for multiple of these boxes and it essentialy reflects the weight of the disks plus some overhead.

Re:Poll Troll Toll

[Frizzle Fry]

Why is this insightful rather than offtopic? It has nothing to do with the post to which he is responding. (Before modding me offtopic, stope to note that I actually am posting something relevant to the parent... then mod me offtopic).

Re:The obligatory shot from the OS X administrator

[Anarchitect_in_oz]

I know the standard Aluminium Xservers look pretty good, but those black ones are just too sexy. Servers shouldn't look that good they should all be butt ugly just like these new IBM ones.

Re:insert witty pr0n comment here

[chawly]

Why did you listen to Mr. Shatner's new album ? Or am I missing something ?

Wording is deceiving

[sf_basilix]

While IBM has produced a large array, it still is not "the most powerful storage system to date" as quoted from your initial posting. Hitachi's new line of storage still outperforms and outsizes IBM's new array by a large magnitude. The new TagmaStore is capable of 332TB. Therefore, Hitachi has "the most powerful storage system to date"

Re:insert witty pr0n comment here

[Anonymous Coward]

its actually really good.

Re:insert witty pr0n comment here

[chawly]

To each his own, brother, to each his own .....