Computex 2011: OCZ Z-Drive R4 88, Eight SF-2281 Controllers in RAID-0
by Anand Lal Shimpi on May 31, 2011 10:06 AM EST- Posted in
- Trade Shows
- SSDs
- SandForce
- OCZ
- SF-2000
- Z-Drive R4
- Computex 2011
OCZ's consumer accessible PCIe SSD line is the RevoDrive, but if you're an enterprise customer and want a custom configuration OCZ will build you a Z-Drive. To show you how far OCZ is willing to go, it announced a brand new Z-Drive at Computex: the R4 88.
Take eight SF-2281 controllers, put them all in a RAID-0 on a PCIe x8 card with an as-of-yet unknown controller and you've got the Z-Drive R4 88:
OCZ will be offering two types: the C Series and the R Series. The C Series has less over provisioning and lacks any polymer capacitors for power loss protection, while the R Series has more NAND set aside as spare area and comes equipped with some form of protection against sudden power loss.
OCZ had a 3U Colfax server with multiple Z-Drive R4 88s running an Iometer 4KB random read test, the end result was a single server that delivered more than 1 million IOPS. The drive is spec'd for up to 2.9GB/s reads and 2.7GB/s writes. Obviously as an 8-way RAID-0 there are reliability concerns and I'm not exactly sure the type of enterprise customer that would deploy such a thing, but it's a neat drive to look at nonetheless.
13 Comments
View All Comments
kanabalize - Tuesday, May 31, 2011 - link
Over 900000???Seriously, only for enterprise right?
or the uber rich...
davegraham - Tuesday, May 31, 2011 - link
RAID 0 is a bad idea for this type of card. Even if they developed some sort of parity schema (R5 or so) or, best case, RAID 1/0, it'd be more applicable. For most of these devices, it's not about capacity, it's about performance.The other thing I'd mention is that without intelligent software controlling the way in which this device is utilized, it's nothing more than an expensive thrill ride. LSI, for example, developed their CacheCade software to allow for their WarpDrive (a more enterprise version of what you see above) to be used as a caching device or as a pure drive.
cheers,
Dave
Havor - Sunday, June 12, 2011 - link
And how is raid 0 more dangerous then a single drive with the same amount of flash chips??Saying that means you don't trust SF-2281 chips it self, because failure of the NAND chips raid is gone be the same, not more and not less.
And what use to be a drag do in SAP now is a lot more "fun" to do.
Do "I" think the the failure raid of NAND chips will be mouths higher then that of controller and the raid chips.
Sure the extra SF + raid chips are extra links in the chain to failure.
But that's why you have raid 10 for critical data, and companies like Shell, GE and so on they need for there SAP database speed so they can afford easily raid 10.
But I could be wrong ^_^
But my experience with a large database on SSD is great at least, as I work as a foreman offshore in the oil industry, and use SAP daily the system was so mouths more responsive, I upgraded to SSD at home and it made my system feel a lot quicker, but this was like upgrading from a 386 to a i7 2600 when using SAP.
And also the general experience was so great that they wrote a article about it in the company newsletter. (and was a nice detailed read)
After the SAP database was set over to SSD drives, productivity when using SAP went up 9% the first week and levelled out at 6%.
(they measured the average time a order was open)
And they calculated that they would save about 10 to 40M Euro a year on man hour's world wide and 1900 Euro's on power with the new servers.
(think they did it for the power savings, they will brake even in 150y ^_^)
Not bad for 280.000 Euro upgrade, replacing the 12 servers of the old database server, for 2 new SSD ones working mirrored (holding 4 cards each), 2 identical spare ones and 2 HDD based backup servers.
There was a usually average backlog of about 2 days for ordering stuff, that went down to now 2 hours in one month.
(don't think they are hiring in the purchase department :-)
Ware it use to be a pain using SAP, specialy if you did complex searches, taking up sometimes up to 1min (usually used for a coffee break, costing even more time ;-), you now have the info for the same data back in under 3sec, and all the normal data from average from 3~5 to less then 1sec.
(it was like upgrading from a old dial-up to cable)
So yeah enterprise costumers don't really look at the price I think
Etsp - Tuesday, May 31, 2011 - link
Inquiring minds would like to know. Like you stated in the article that while 8 of these in RAID-0 is a neat trick, it's entirely unpractical. Would they be able to provide any details on other RAID configurations that this device supports? RAID 10 Perhaps? Is additional hardware required to support RAID?pvdw - Saturday, June 4, 2011 - link
Remember to think ENTERPRISE. I think this would make for a really good web "read" caching server for an ISP or file download service.Or how about when you already have 10 load-balanced DB or web servers and need more IOPS?
Or how about multi-layered decision support for a teradata DB?
I'm sure others can come up with more scenarios where data loss isn't really that important.
dac7nco - Tuesday, May 31, 2011 - link
That has LSI written all over it.Daimon
neotiger - Tuesday, May 31, 2011 - link
The RevoDrive supports TRIM? Does Z-Drive?Also any pricing information?
JasonInofuentes - Tuesday, May 31, 2011 - link
Pricing probably won't roll out for sometime, and it will be geared at enterprise so this isn't going to come close to affordable for most users.TRIM? Not behind a RAID.
neotiger - Tuesday, May 31, 2011 - link
>TRIM? Not behind a RAID.Not true.
RevoDrive uses RAID and it has TRIM.
Casper42 - Tuesday, May 31, 2011 - link
I don't know the internal design, but this has Fusion IO competition written all over it.And as far as what customers could afford to use it (afford the data loss I mean), just go look at pretty much any Fusion IO Marketing slick and you have your answer.
I have seen them personally be installed for Data Mining (Pull a subset of data from a different DB Server that has the redundancy you need) and then run an Analytics package against it to look for trends. And when it comes to Primary DB duties, there are numerous functions that don't require absolute redundancy in place, like Indexes and TempDB.