backtop


Print 23 comment(s) - last by Wwhat.. on May 1 at 10:37 AM

2760 drives, 500GB each, one huge storage array

Storage has become monumental these days, especially with various companies broadcasting companies going online and making TV shows and movies available for IP viewing. Such trends in the industry push storage companies to create and invent new ideas and products that not only drive prices down, but give consumers plentiful choices in storage technologies.

Fujitsu recently launched what it claims to be the world's largest storage array, the ETERNUS 8000 and ETERNUS 4000 storage arrays. Weighing in at 1.36 petabytes, or 1.36 million gigabytes, the ETERNUS file storage arrays push the envelope for enterprise data storage systems. Fujitsu uses 2,760 nearline fibre-channel 500GB disk drives in its flagship ETERNUS server (model 2100) and can be configured with up to 256GB of cache.
  • World's largest storage capacity (over 1 petabyte)
  • World's highest storage performance from up to sixteen 3.6GHz dual processors per system
  • RAID6 (double parity) support. Enhanced reliability even if simultaneous HDD failures occur
  • Disk to Disk backup using Nearline FC disk drives
  • Multi-functional backup options with OPC, QuickOPC & SnapOPC
  • Cost-effective disaster recovery solutions using iSCSI
  • Data encryption for online and offline information security
Fujitsu packs in preventative maintenance features such as tape backup, data encryption, and data automation features into the ETERNUS arrays just in case disaster should strike. Performance wise, Fujitsu says that its servers can be equipped with up to 32 dual-core processors, giving the arrays unmatched throughput. Fujitsu hopes to ship 14,000 units over the next two years.

While Fujitsu claims the largest array available, EMC Corporation launched the first petabyte server that was commercially available back in January of this year.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

i wish...
By covertbit on 4/24/2006 12:41:37 PM , Rating: 2
I wonder how long it will take to do a full format of that hard drive? I am sure Google and other search engines will be the biggest buyers of these systems. I wish I had one =[




RE: i wish...
By peternelson on 4/24/2006 5:44:46 PM , Rating: 4

Yeah, a service like Gmail could use this kind of space.

As for formatting.... well, it could format multiple drives at the same time, so need not take an eternity.

Also, if it is clever like EMC similar product, it uses its own virtual object based filesystem. The storage system tells the computer it has loads of space even if the disks are not physically present. You can then buy the disks later, as and when your computers have filled up most of the real space. The system can extend the actual volume size dynamically as you hotswap more disks. Often the array is shared between multiple hosts (so called "storage consolidation"). eg you could tell your mainframe that it has 75% of the capacity and your unix box that it has 75% of the capacity. Either can merrily just save stuff, and the storage system will decide who gets the underlying actual disks. Assuming only one system actually needs what you said it could have then it saves you buying more disks than needed.

Also note in the description "nearline" that means the disks not being used right now are spun down, and only accessed when required. This is part of something called hierarchical storage management. The most regularly accessed sectors live in the cache, the next most accessed are in powered up disks, then the less accessed stuff can be on the spun down disks to save power. An extension of this is the stuff you want to keep but never gets accessed can be dumped off to tape and recovered if it is ever needed.

By doing all this, it can format the new hard drives as you add them with its own object filesystem format (if it wants to). The storage system could tell the connected computers that the space is already there and formatted even before you install the disk. The management software will prompt you to install more disks to grow the array in response to the usage approaching some threshold.

There is lots of clever stuff to manage all this, and that is why this system is likely to cost well over $1m, and certainly much more than the drive cost alone.


RE: i wish...
By pinzboy on 4/26/2006 9:49:35 AM , Rating: 2
Gmail has no use for this box. Gmail runs on the Google filesystem. Poke Google for URLs. They run commodity boxes in large clusters. No large disk arrays. That's the whole point of Google.

And "nearline" does NOT mean they spin down disks. In this case "nearline" is referring to the FATA disks they use. They have dual-port fibre channel interfaces, but SATA canisters. They cost much more than commodity desktop drives, and provide port failover if one of the interfaces fails, but they're still just 7200RPM ATA quality drives inside.

Go poke NetApp's website for their R100 "nearline" appliance. It doesn't spin down disks either.

Spinning down disks is dangerous, and often leads to data loss. Companies like Copan have spent years working out the issues with spinning down disks, duty cycles, and other issues. This is not a MAID box.

And there's no heirarchical storage management going on here. There's only one tier of storage. Even MAID boxes don't qualify as HSM. The simple fact that they're fronting the array with cache doesn't make it HSM.



RE: i wish...
By peternelson on 4/29/2006 6:23:51 AM , Rating: 2

RAM > CACHE > PRIMARY STORAGE > NEARLINE > TAPE

By definition using nearline is hierarchical.

It is not as expensive or fast as primary but is great as an intermediate place to make tape backups from.

Primary storage may be something like 15K scsi or solid state disk array.

Many people set their disks to spin down. All you need to be aware of is a per-drive limit of (say 40,000) startups.

So its no use spinning down your primary storage.

But you COULD spin down SOME of your nearline storage if what you put on it is infrequently accessed.


RE: i wish...
By peternelson on 4/29/2006 6:32:04 AM , Rating: 2
Using the "MAID" (idle) definition ie standby mode, "MAID" would sit between NEARLINE and TAPE in my > diagram.

However there could be a hybrid nearline box that supports MAID operation on SELECTED disks, the remainder being kept up. Therefore it may spin down a particular or not depending on what is on it and access patterns.

The article does not state whether THIS device offers such a function but I accept nearline does not necessarily imply the device has it.


RE: i wish...
By Wwhat on 5/1/2006 10:37:16 AM , Rating: 2
Yuo wisehd you had one?
What would you do with it really? no single person needs that kind of space (and hassle)


Cost?
By TomZ on 4/24/2006 12:01:32 PM , Rating: 2
Just curious, how much would such a server cost?

Or is it the case of, if you have to ask the cost, then you can't afford it?




RE: Cost?
By theubergeek on 4/24/2006 12:40:05 PM , Rating: 2
Well, if you figure $250 for each of the 2760 drives, you'll be spending $690,000 for drives alone, then add the cost of the other stuff, you'll be pushing better than a million.


RE: Cost?
By segagenesis on 4/24/2006 1:09:58 PM , Rating: 2
The interesting note here is this is still cheaper than the original of them all!

The 350 Disk File consisted of a stack of fifty 24" discs that can be seen to the left of the operator in the above picture. The capacity of the entire disk file was 5 million 7-bit characters, which works out to about 4.4 MB in modern parlance. This is about the same capacity as the first personal computer hard drives that appeared in the early 1980's, but was an enormous capacity for 1956. IBM leased the 350 Disk File for a $35,000 annual fee.

I believe the 350 RAMAC was close to one million in 1956 dollars to buy outright (!)


RE: Cost?
By Phynaz on 4/24/2006 1:17:36 PM , Rating: 2
Those drives are probably 5 times that much.


RE: Cost?
By DarthPierce on 4/24/2006 1:23:51 PM , Rating: 2
500Gig FibreChannel drives are nowhere near $250 more like $2000 a pop..... so you're looking at ~$6,000,000 for drives alone.... I'd say it's probably over 10 mil for a fully decked out one. (processors, power supplies, tapes, cache, etc)


RE: Cost?
By OrSin on 4/27/2006 9:41:19 AM , Rating: 1
Your an idiot Fiber channel is not a type of drive if means of connecting the array tot ht oether boxs and external data.

The driver are SATA. NOw i'm sure the mark up for the drive make them cost about 500-600 but $2000 is just stupid. Fiber channel drive is like discribing a plan by where it lands. They have nothing to do with each other.


RE: Cost?
By Wwhat on 5/1/2006 10:35:23 AM , Rating: 2
On top of it all you need an expensive cooling system and how many PSU's does it all take? I think just counting up the cost of harddrives will get you less than 30% of the cost, and for maintenance and power and cooling you can expect a hefty monthly bill too.


1 PB != 10^6 GB
By Trisped on 4/24/2006 2:10:23 PM , Rating: 3
quote:
1.36 petabytes, or 1.36 million gigabytes

Well, which is it? 1.36 PB or 1.36*10^6 GB?
Since in computers everything is done in powers of two every stage is counted in 1024 in stead of 1000. So 1 KB = 1024 bytes, 1MB = 1024 KB, 1GB = 1024GB, 1 TB = 1024 GB, and 1 PD = 1024 TB. So 1.36 PB = 1392.64 TB = 1426063.36 GB which rounds to 1.43*10^6 GB, not 1.36

From the article:
quote:
The top of the ETERNUS8000 line is the Model 2100, which provides up to 1.36 PB of internal storage capacity, the world's largest





RE: 1 PB != 10^6 GB
By TomZ on 4/24/2006 4:23:45 PM , Rating: 2
It's pretty common for HDD storage to be quoted in MB=1000KB, GB=1000MB, etc.


RE: 1 PB != 10^6 GB
By Steve Guilliot on 4/24/2006 5:31:29 PM , Rating: 2
Wow, you're a fricken super genius.


LAN
By Orpheus333 on 4/24/2006 12:45:56 PM , Rating: 2
Sweet, I've been in the market for a file server for our neighbor hood LAN parties...




RE: LAN
By TheLiberalTruth on 4/24/2006 2:05:18 PM , Rating: 4
Yeah, if your neighbourhood is the hamptons, you might be on to something. :D


Porn
By latino666 on 4/24/2006 11:45:42 AM , Rating: 2
Thats a lot of porn!




RE: Porn
By goku on 4/26/2006 11:03:45 PM , Rating: 2
Thats a lot of HD porn..


Tape backup?
By Homerboy on 4/24/2006 5:36:34 PM , Rating: 2
"Fujitsu packs in preventative maintenance features such as tape backup..."

Uhhh how long does that take?





RE: Tape backup?
By peternelson on 4/24/2006 5:49:50 PM , Rating: 2
That depends how much of the data is going onto tape.

It also depends on the tape standards, modern ones are pretty fast.

It also depends on the number of tape drives you have all storing simultaneously.

Tape may lack true random access but it still has a lower cost per unit storage than upmarket hard drives.


By Johnmcl7 on 4/24/2006 11:28:40 AM , Rating: 2
Just under 3,000 drives - woah!




"It's okay. The scenarios aren't that clear. But it's good looking. [Steve Jobs] does good design, and [the iPad] is absolutely a good example of that." -- Bill Gates on the Apple iPad











botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki