Whose SSD are they using in Project Quicksilver to get that kind of performance?
|
Does Verizon's Voyager stack up to the iPhone? |
|
|
5 IT skills that won't boost your salary 
[1,407]
Women 4 times more likely than men to cough up personal info [589]
Japan's 10 funniest tech-related commercials [Videos] [407]
Throwing away a promo CD is "unauthorized distribution"? [1,265]
Adults too quick to dismiss educational video games [682]
Attack of the iPhone clones [Slideshow] [578]
10 things IT needs to know about AJAX [1,258]
This Year's 25 Geekiest 25th Anniversaries [Slideshow] [409]
|
|
The IBM article says Fusion-io, http://www...
The IBM article says Fusion-io, http://www.ibm.com/developerworks/blogs/page/InsideSystemStorage?entry=ibm_breaks_1_million_iops
Their own.
Since IBM Research is involved, this is unlikely to be a case of integrating off-the-shelf parts to build a solution. IBM does a great deal of basic scientific research and in-house invention, has its own chip fabs, etc etc etc.
benchmarks
**edit: they claim to be using off the shelf SSDs from Fusion. Also, their SVC engine which is based on off the shelf intel servers running a version of Linux.
They said they did an SPC-1 benchmark which is modeled for about 70% random reads and 30% writes. Most SSDs get about 30,000 IO/s with random reads and about 5000 IO/s with writes. Also, they say that they put them behind an SVC controller.
Parallel
My guess is that they're getting the speeds via parallel access. It would be a lot like RAID striping.
Technical Details
We are indeed using FusionIO cards to build a storage controller unit that presents itself onto a fibre-channel network, in this case storage is presented behind a large SVC cluster.
The workload benchmarked was 70/30 4K all miss which is somewhat representative of a typical database workload.
If people have actually played with SSD devices, they will know that the claimed X0,000 reads and Y,000 writes are all very well when doing 100% read or write at say 512 byte transfers, but bump up the transfer size, and throw in a mixed workload and you will prune the men from the boys... i.e. IOPs will be no where near those claimed by manufactures...
Anyone interested in more details of what we have been doing see my recent blog post :
http://www.ibm.com/developerworks/blogs/page/storagevirtualization?entry=1m_iops_from_flash_actions
it would obviously run raid. duh
it would obviously run raid. probably raid 10
A little perspective, please
For a bit more balancing perspective, including additional admissions from the guys who ran the tests, check out my blog and the comments on this post: http://thestorageanarchist.typepad.com/weblog/2008/09/1023-its-just-a.html
raw speed
The raw speed is there - 600 MB/s writes and 700 MB/s reads ( 4k blocks sustained ). I played with an IOdrive on a linux box. Read a 4.7 GB ISO image in under 7 secs., copied the file in 17 secs.
how to create a project using flash memory
help us! can anyone tell us where we can find the
Post new comment