Tweet
Here is something that other readers might find interesting. We were trying to install a new RAID disk system on our server. We wanted RAID 1 (Disk mirroring) for security. We were thinking of writing a short review on RAID-1 benchmarks for this post, but it turned into more a RAID trouble shooting session.
We decided to use a Promise FastTrak S150 TX4 SATA RAID controller and a couple of new Seagate drives. Seagate Barracuda ES 250GB ST3250620NS SATA-II Hard Drive with 16MB Cache & NCQ. These were all put into a HP xw8200 workstation running Windows server 2003 32bit. The FastTrak S150 was put into a 64bit PCI slot but Promise claim this is a valid config.
This was not the boot drive. The boot drive was a separate 36GB 10K SCSI drive.
The RAID install went fine. But when we tried to use the drive, the problems started.
We were getting very very slow write speeds. With RAID 1, it is normal that we loose some write speed. But we were seeing speeds of between 1 and 3 MB/sec. Which is about twenty times slower than expected.
Initially we blamed PerformanceTest for the low result. As surely the hardware couldn't be this bad for a sequential write. (In hindsight, we should have trusted our own software).
We also did a lot of testing with the PerformanceTest advanced disk test, which reported the same poor write results (between 1 and 5MB/sec) as soon as the cache was off, or the cache's capacity was exceeded.
So to confirm this result we used Windows Explorer to copy a few files onto the new mirror. The first few (small) files we copied, wrote very quickly (> 200MB/sec). But this was just the Windows cache in action.
When we copied a large 4GB file we saw the awful truth.
It was going to take almost 4 hours to write a 4GB file!!!
Clearly there was something very wrong. We checked the controller BIOS version (1.00.37), check the driver revisions, checked the DMA settings (UDMA mode 6), etc... We couldn't find anything wrong.
We have reported the problem to Promise and if we get an solution I'll post it here.
In the meantime we'll be pulling out the Promise controller and try using the software RAID facility provided within Windows, which I have heard good things about.
Update - Dec 06:
We have been using software RAID (dynamic disks) on Window 2003 for 3 months now. It has been working well and in fact has some significant advantages over a hardware solution. You should consider this option, especially for RAID-1 on machines with dual CPUs (as software RAID takes a bit of extra CPU time).
We decided to use a Promise FastTrak S150 TX4 SATA RAID controller and a couple of new Seagate drives. Seagate Barracuda ES 250GB ST3250620NS SATA-II Hard Drive with 16MB Cache & NCQ. These were all put into a HP xw8200 workstation running Windows server 2003 32bit. The FastTrak S150 was put into a 64bit PCI slot but Promise claim this is a valid config.
This was not the boot drive. The boot drive was a separate 36GB 10K SCSI drive.
The RAID install went fine. But when we tried to use the drive, the problems started.
We were getting very very slow write speeds. With RAID 1, it is normal that we loose some write speed. But we were seeing speeds of between 1 and 3 MB/sec. Which is about twenty times slower than expected.
Initially we blamed PerformanceTest for the low result. As surely the hardware couldn't be this bad for a sequential write. (In hindsight, we should have trusted our own software).
We also did a lot of testing with the PerformanceTest advanced disk test, which reported the same poor write results (between 1 and 5MB/sec) as soon as the cache was off, or the cache's capacity was exceeded.
So to confirm this result we used Windows Explorer to copy a few files onto the new mirror. The first few (small) files we copied, wrote very quickly (> 200MB/sec). But this was just the Windows cache in action.
When we copied a large 4GB file we saw the awful truth.
It was going to take almost 4 hours to write a 4GB file!!!
Clearly there was something very wrong. We checked the controller BIOS version (1.00.37), check the driver revisions, checked the DMA settings (UDMA mode 6), etc... We couldn't find anything wrong.
We have reported the problem to Promise and if we get an solution I'll post it here.
In the meantime we'll be pulling out the Promise controller and try using the software RAID facility provided within Windows, which I have heard good things about.
Update - Dec 06:
We have been using software RAID (dynamic disks) on Window 2003 for 3 months now. It has been working well and in fact has some significant advantages over a hardware solution. You should consider this option, especially for RAID-1 on machines with dual CPUs (as software RAID takes a bit of extra CPU time).
Comment