IOMeter v2006.07.27 – IOps
Iometer is an I/O subsystem measurement and characterization tool for single and clustered systems. It was originally developed by the Intel Corporation and announced at the Intel Developers Forum (IDF) on February 17, 1998 – since then it got wide spread within the industry.
Meanwhile Intel has discontinued to work on Iometer and it was given to the Open Source Development Lab (OSDL). In November 2001, a project was registered at SourceForge.net and an initial drop was provided. Since the relaunch in February 2003, the project is driven by an international group of individuals who are continuesly improving, porting and extend the product.
These results are both complex and interesting at the same time. Every SSD solution behaves differently as a result of its consruction and implementation. These charts show that bandwidth is not always the solution. For example, we see the Intel SATA 3Gb/sec SATA SSD pushing past devices of much higher bandwidths. This only happens at lower queue depths, where the very low latency of the controller enables it to push past the others.
The case is similar with the ioDrive. It’s driver essentially speaks directly to the bank of flash memory, significantly reducing per-transaction latency. End result: while the RevoDrive 3 X2 was able to match, or even exceed the ioDrive in the case of the Web Server test, it was only able to do so at higher queue depths. The limit here is primarily due to the SandForce controllers at play. As good as VCA is, it can only show its true potential at those higher queue depths.
This is not necessarily a bad thing. It only means the Revo 3’s are meant to excel at moderate to heavy multi-threaded workloads. Also keep in mind – the RevoDrive 3 X2 is no slouch at the lower queue depths, either.
Great review Allyn. You
Great review Allyn. You mentioned that the RevoDrive would be good for entry-level servers or high performance work stations. Would the Raid 0 nature of the RevoDrive make you a little wary of placing this into a server environment? Maybe you’re implying there would be fail over/redundancy protections in place, but at least with hard drives, I would never consider a R0 server setup for our small biz server.
Would you mind elaborating a bit?
Thanks, Chris
For the type of application
For the type of application where super-high IOPS is desired, you reach a threshold where you just can’t get any higher with redundancy included in the package. For these situations you’d have to add the redundancy factor yourself, be it near-line backups or an identical controller installed into the same server and mirrored using the OS perhaps. The Revo is not unique in this bleeding-edge performance niche – the FusionIO products are not redundant either.
As a side note, many VCA controllers have a mode equivalent to RAID-5. This silicon is fast enough to include parity for one (or two) drive failures with a minimal performance hit. OCZ could probably enable this, perhaps for a more purely business-oriented model, but I don’t see it as that high of a need really. It would be a niche of what is already a niche to begin with!
Consider that adding parity to support failure of one of the SandForce channels. That’s some form of chip failure – be it the SandForce or its bank of flash. There’s plenty of other chips on a RevoDrive that are common to the unit as a whole, and a failure of one of those will still take the whole card down (single point failure). If redundancy was so important for a very high performance application, I would rather skip the parity calculation overhead / capacity reduction and go straight for the ‘2 of everything’ approach. Added bonus – a pair of these mirrored at the OS level would have the write performance of a single unit, but read performance would be doubled, so you would *further* increase performance and add redundancy at the same time. That’s a win-win not possible with overhead-inducing RAID / parity-based solutions.
Also note: You could buy 2 of the 960GB models for *less* than a single ioDrive 160. Ouch.
Al
It’s nice, but for $1600 you
It’s nice, but for $1600 you can get a 3Ware 9750 controller and 4 Vertex 3 120GB drives, get the same capacity, 512MB of buffer, and a wider interface, for $100 cheaper. I think I’d rather go that way.
The interface is irrelevant
The interface is irrelevant if the controller itself is the bottleneck. If the 3Ware can manage to peg the PCIe interface in sequential reads, that’s great, but only marginally better than what the OCZ unit can do (as it’s nearly saturating all 4 SSD’s).
The LSISAS2108 chip (on the 3Ware 9750) is only rated at 1.1GB/sec sequential writes, while the Revo3 hits close to 1.6GB/sec. That’s a 50% reduction in performance.
The LSISAS2108 will very likely come nowhere near the Revo3’s 200k IOPS rating in random access. Every RAID solution I’ve tested tops out at roughly the peak IOPS of *1* good SSD (~50k IOPS). This is due to the latency added by managing the cache. 512MB RAM caches are for HDD’s, not SSD’s. While I haven’t personally tested the LSISAS2108 / 3Ware 9750, I’d be shocked if it could break 100k.
That does raise a question:
That does raise a question: THis and the previous Revodrives are internal RAID 0 devices, so why don’t you test it against RAID setups? Of course it’s going to blow a lone SSD out of the water. So would a quartet of those same SSDs attached to a RAID controller, so that doesn’t, by itself, constitute a reason to get one of these.
I have heard the some comment about why not just get a raid controller and a couple of SSDs instead regarding the original revo too. Some data that supports one of those options is much more useful than including a HDD for us to laugh at.
I have had plenty of
I have had plenty of experience with the Revo and Revo2. I wish to express my words of caution of using the Revo3 as a boot drive. With many weeks of testing several motherboards and adjustings various BIOS’s. The Revo’s could never become stable as a boot drive. Somehow, files would always become corrupt. As a secondary drive, for placing the pagefile, all temp files and general work files. The revo’s are mind blowing fast. No doubt, by looking at the numbers for the Revo3, it will impress anyone.
i would only get this storage
i would only get this storage as a boot drive. I wonder what Allyn has to say about that. i would put in the review that it is a boot-able drive though. i am not sure how you would test it for instability as a boot drive. i dont heard in the news about issues of things being corrupt and such. I will keep this in mind.
i like this storage drive though, hate to hear bad things about it.
I’ve tested all three Revo’s
I’ve tested all three Revo’s in a boot configuration and never noted any corruption issues. Revo / x2 were tested under XP and XP64. I’ve only tested Revo3 under 32-bit (Win 7) as we don’t yet have signed 64 bit drivers, which precludes install without additional hackery.
That is one sexy piece of
That is one sexy piece of silicon lol
Well, i am using a revodrive
Well, i am using a revodrive as boot drive for over a year and ir never fails a boot or damage files…
Are you aware of this
Are you aware of this Allyn?
“However I was wondering about TRIM support, as the last time I checked it needs to be supported by Microsoft. A question that I had to verify with OCZ. Though the Revo3 card supports TRIM, because the architecture is based on SCSI, the Microsoft Windows StorPort architecture currently does not support either TRIM or SCSI UNMAP. As such, these commands are not generated by the OS, which of course prevents VCA from executing them. OCZ is working with Microsoft to have this functionality enabled as soon as possible. So that will take a RAID driver update alright, but that should not effect your data already on the drive.”
Source: http://www.guru3d.com/article/ocz-revodrive-3-x2-review/16
Anandtech and tomshardware say something similar.
Apparently TRIM only works on Linux as of now.
Yes. TRIM won’t pass through
Yes. TRIM won’t pass through StorPort until Microsoft adds support for this functionality to Windows. This will likely come in the form of a hotfix, but the ETA on that is totally up to Microsoft.
As with other SF-controlled devices, keep in mind they are very resilient to the performance hit seen when used without TRIM. The same applies to the Revo3.
What an insightful and
What an insightful and well-written review. Its good to see PC Perspective attracting such great talent! I look forward to more of your articles.