I installed an SSD on my main machine. I'm using this machine for way too
many hours per week. I measured the performance of the SSD with SysBench
when it was new and after nearly three months, measured the performance
again. For a control, I also measured a new SATA 500 GB hard drive to show
how that had aged.
Here are the raw results.
January 30, 2011
Disk I/O disk 0-1: 38162 MB - Mushkin MKNSSDCL40GB-DX
Avg. data access time : --.--- milliseconds
Cache/Bus xfer rate : 214.346 Megabytes/second
Track 0 xfer rate fwd : 170.150 Megabytes/second
Middle trk rate fwds. : 208.443 Megabytes/second
Last track rate bwds. : 229.395 Megabytes/second
Average Transfer rate : 202.663 Megabytes/second
Disk use CPU load : 4.530 percent
-----------------------------------------------------------------------
Total : --.--- Disk I/O-marks
Disk I/O disk 0-2: 476938 MB - Hitachi HDP725050GLA360
Avg. data access time : 18.300 milliseconds
Cache/Bus xfer rate : 217.855 Megabytes/second
Track 0 xfer rate fwd : 87.965 Megabytes/second
Middle trk rate fwds. : 79.320 Megabytes/second
Last track rate bwds. : 47.720 Megabytes/second
Average Transfer rate : 71.668 Megabytes/second
Disk use CPU load : 1.820 percent
-----------------------------------------------------------------------
Total : 445.426 Disk I/O-marks
April 21, 2011
Disk I/O disk 0-1: 38162 MB - Mushkin MKNSSDCL40GB-DX
Avg. data access time : --.--- milliseconds
Cache/Bus xfer rate : 214.309 Megabytes/second
Track 0 xfer rate fwd : 169.736 Megabytes/second
Middle trk rate fwds. : 205.306 Megabytes/second
Last track rate bwds. : 229.634 Megabytes/second
Average Transfer rate : 201.559 Megabytes/second
Disk use CPU load : 4.560 percent
-----------------------------------------------------------------------
Total : --.--- Disk I/O-marks
Disk I/O disk 0-2: 476938 MB - Hitachi HDP725050GLA360
Avg. data access time : 18.200 milliseconds
Cache/Bus xfer rate : 217.789 Megabytes/second
Track 0 xfer rate fwd : 87.885 Megabytes/second
Middle trk rate fwds. : 79.311 Megabytes/second
Last track rate bwds. : 47.762 Megabytes/second
Average Transfer rate : 71.652 Megabytes/second
Disk use CPU load : 1.840 percent
-----------------------------------------------------------------------
Total : 444.863 Disk I/O-marks
The amount of aging is barely larger than the expected measurement error
for the benchmark program. Based on this test, I'd expect the SSD to
remain pretty fast for years.
This is very good news. I have an Agility 2 that I can put into operation as long as I can partition it properly on the right boundaries to make the IO efficient.
Is there any reason the speed should degrade over time? After all it is using electronic memory.
The only degrading should be of capacity as memory cells die and even that will depend on how the firmware is written.
Most SSD have now a system like "Garbage Collection" which runs during idle time and optimize nand usage for constant performance. Under Windows 7, you'll have TRIM function
Sandforce have integrated similar system and Kingston build special SSDs V+xxx with a higher efficient "garbage collection" (as good or better as trim) process usable under any OS.
May be you could see same kind of hung for very short time when going out of idle state after optimization !
Without this, your SSD could have its perfs going down more an more an may be lower than a normal HD.
Quote from: ivan on 2011.04.22, 14:02:48
Is there any reason the speed should degrade over time? After all it is using electronic memory.
Flash memory need to be erased before it can be rewritten. Since a virgin SSD is completely
erased, most write operations can start immediately. When it fills up, chances that a block
that needs to be erased is hit will grow. This is causing the slowdown. There are several ways
to avoid this: A dedicated "TRIM" tool, that is run manually by the user (file system specific,
mostly Windows-only). A file system, the emits TRIM SCSI commands to tell the drive which
blocks are no longer in use so that it can erase these blocks when it is not doing any other
useful stuff (I think Windows 7 does this). Hardware based "garbage collection" solutions, that
combine block erasing with remapping. The latter is needed anyway in order to avoid premature
wearing out of flash cells in case the same block is written over and over again.
In other words rudi, you are saying speed is governed by firmware housekeeping which is invisible to the user, therefore the answer to my question is NO.
ivan
Replying to Ivan, I think I agree. The Mushkin SSD in my Pubopeep computer has the Sandforce controller. I installed eCS to the SSD in exactly the same way as I would to a hard drive. An SSD is just a more expensive hard drive as far as eCS is concerned.
The speed-up is less impressive in a desktop than it is in my Mobilopeep laptop. I think that's because the desktop was faster to begin with.
Has anyone checked the speed of HPFS vs JFS for an SSD? The larger cache in JFS means it's twice as fast for disk intensive operations like building PMMail/2 from source code. I still need to try that on SSD.
It's important to get an SLC SSD if at all possible. I put a 16GB IDE one in an old notebook and it absolutely flies with ecs. The MLC drives in my experience do have a tendency to degrade over time, it's due to the fact that they store data using a double state bit that can hold 2 bits in one cell. It gets messy if allignment is not set right.
Mind you, SLC drives are even more small and expensive.
http://translate.google.fr/translate?u=http%3A%2F%2Fwww.presence-pc.com%2Factualite%2Ftrim-ssd-36838%2F&sl=fr&tl=en&hl=&ie=UTF-8
Thanks for good article about the TRIM command. I notice that it makes no claim that TRIM helps at all. I'll bet it doesn't help much for clients, seeing as Windows 7 client does not have it.