The ICH10R peaks at aroun 650-750MB/s sequential reads so the RAID controller is the bottleneck. You might be able to squeeze higher synthenic numbers if you use deeper queue depths and large fliles.
However, this is actually great for an onboard controller. If you want more sequential throughput, you will need to invest in a discrete hardware RAID controller ($150+).
Why do you think you should be seeing over 800 MB/s read speeds? What is that figure based on or how is it calculated?
Probably 250MB/s sequential read for each drive times 4... then minus some for imperfect scaling.
Hi Duckie, Yes, that is one obvious but IMO rather simplistic method of determining what the result "should" be. Things are rarely that simple and the entire system must be taken into account. As you mentioned, there is a limit to what the SATA controller can provide. I've never seen the figure you mentioned in any of the ICH10 documents, I'm not questioning it and might have missed it. I wonder if differences in the RAID driver and its settings would make any difference.
Hello and thank you for the replies, I am using the most recent updated drivers for ICH10 raid controller and only option I have toggled is
Write Back Cache : Enabled in my Intel Rapid Storage Technology application. Is there any other setting I should look in to ?
Here is a screenshot of HD Tune Pro stats.
- Intel X25-M SSDSA2MH080G2K5 2.5" 80GB SATA II MLC Internal Solid State Drive ( 4X )
- Sequential Access - Read up to 250 Mb/s on each drive
- Forgot to mention I am also using latest firmware for each ssd drive. On a side note which raid controller would you recommend ?
Parsec, I am not sure if Intel prints the peak bandwidth of the ICH10R anywhere. However, I have seen that 600-700MB/s bottleneck occur on multiple benchmarks over the years.
Certified, I can really make direct recommendations. However, you would be looking at server hardware RAID controllers with dedicated RAM and CPU. The newer models specifically have support for SSDs. Note that RAID0 provides sequential performance gains.for the most part. Your random performance is probably about the same. Unless you have specific workloads that do large (1GB+) sequential read/writes, the benefit is questionable.