Review: OCZ SSD – Solid State Disk

Uncategorized — By Chris Evans on February 10, 2009 at 3:04 PM

We’re all aware of how solid state disks are becoming mainstream storage devices. Vendors are adding SSDs to arrays and to servers in order to increase levels of performance. I’ve been curious to understand how an SSD performs relative to traditional disk devices so I’ve purchased a typical SSD from OCZ. Specifically I’m testing a 128GB model generation 1 of the Core Series SSD (pictured).

Physical

The device itself is pretty neat. It has a SATA-II interface and only weighs about 77g in a standard 2.5″ drive form factor. This is really light; compare this to a Seagate 15K Savvio drive, which at 210g is nearly 3 times as heavy.

Power

Unfortunately OCZ don’t provide any specific details on power consumption and I don’t have any technology to measure the power draw of a single drive, however they do claim that the Core SSD operates at approximately 50% of the power of an equivalent 15K 2.5″ hard drive. This is a substantial reduction (and not really surprising due to the lack of moving parts) however the saving is only likely to be relevant where many SSDs are installed as the reduction represents only a few watts. What is clear is that compared to my 15K Savvio drives on test, the OCZ SSD dissipates almost no heat, whereas in operation, the Savvios are too hot to touch.

Cost

Typical UK street price for the OCZ 1C128G model (128GB) is about £270 or around £2.11/GB. A Seagate Savvio 15K 72GB drive retails at about £327, or £4.54/GB. It may seem a little odd that the SSD is more expensive however the OCZ Core SSD is not the top of the range and the Savvio 15K is not a typical consumer device. In fact a 10K Savvio drive can be purchased for £1.35/GB, pricing the SSD midway between the two standard drives. What this highlights is that SSD currently offers a competitive price point compared to traditional hard disks.

Performance

You would expect that performance is the area where SSDs excel – and they do. This isn’t really surprising as SSDs have no moving parts and therefore don’t suffer from latency (rotational delay) and seek time. A high level of performance can be seen on the first of the graphs displayed. This shows sequential read and write performance (100% read, 50/50% read/write mix and 0% read, or 100% write) at varying queue depth.

A respectable 5000+ IOPS (~85MB/s) was achieved for reads and around 4000+ (66MB/s) IOPS for writes. However a mixed workload of 50% reads and writes was less than successful, performing worse than writes alone. SSDs suffer on write performance due to the architecture which requires data to be erased to achieve writes. A block of data must first be read, modified then written back into place, even for a single byte change. This means multiple operations are needed for each write and if they are intermixed with reads, then poor performance is the result. This is evident in the random performance test, which shows less than 10 IOPS for totally random workload.

The third graph shows how the OCZ SSD performs for a range of block sizes. There’s clearly a breakpoint at 4KB where performance improves. This could be for two reasons (a) this is an NTFS file system and all writes will be changes to blocks of 4K, or the device itself copes better with greater than 4K-blocks. As I collect more data from other devices, then this should become apparent.

The last two graphs compare the OCZ SSD to other unused HDDs I had. Sequential Write performance is not much better than the Western Digital 250GB drive I have; it certainly is the worst performer in the Random Write tests, for the reasons mentioned earlier.

It is perhaps unreasonable to look at the results of the random test and see a problem. Most operating systems buffer data and lazy write out to disk, so here the OCZ SSD shouldn’t suffer too much. I’ve already tried the device in a VMware ESX server and anecdotally, the installation of Windows seemed faster. I will repeat the tests again and compare two installation times.

For reference, tests were performed using IOMETER on a standard AMD Althon 64 X2 Dual Core 4400 CPU with 4GB of RAM and running Windows XP. The SSD was connected directly to a SATA interface on the motherboard, disabling write caching.

Tags: Uncategorized

(No Ratings Yet)

Loading ...

1 Comment

Barry Whyte says:

10 February, 2009 at 8:06 PM

Chris,

The random write performance is the achillies heel of all “laptop” SSD devices. Unlike the enterprise equivalents there is always a MASSIVE disconnect between random read and random write.

In general you will find that today’s “laptop” SSDs are not worth the money, unless you are doing some seriously read intensive work.

Think about having your windows pagefile on one of these devices, and you will see worse end user performance than from a standard HDD.

For now, the only option is to spend many thousands of $ on an enterprise device if you want to see end user performance gains – even at home.

Barry