Yes, this is not ideal for small files, but the application I have is based on processing images hundreds of megabytes (or larger) so I'm trying to optimize large file performance.
What I am trying to understand is if cluster size impacts large file performance. In this particular case I'm curious if matching the cluster size to the RAID block size helps at all - in such a case each file system operation to the disk will equal one RAID operation (I imagine).
Cheers, Eric
You are waiting for two or more distinct physical actions to occur and register within the disk controller. Increase the latency value to compensate for a longer (combined) wait state.
A long standing adjustment known to address similar issues.
BUT NOT ON MS! Never mind. Unconfirmed but from an engineer I trust - private communication not for publication.
I was just directed to a reference that MS, beginning with XP SP2, only allows a max of 8k clusters on the boot volume. XP SP1 was the last MS OS that will allow this.
M3a79-t raid problems - Guru3D.com Forums
See post #4
The default cluster size for NTFS is 4K - that is all file system operations to the disk controller are performed 4K bytes at a time. With 64K clusters, operations are performed 64K bytes at a time.
The default block size for many RAID systems is 128K (mine included). This means RAID controller operations to disk are performed 128K bytes at a time - or 128K is written to each disk in the RAID until moving on to the next disk.
I don't like using the term stripe size as different people use it differently. To some people stripe-size = block-size * number-of-disks, whereas other people assume stripe-size means block-size.
Looking at various benchmarks I have found that block sizes over 64K usually offer little extra performance increase. This is likely because many I/O controllers have been optimized to perform operations 64K bytes at a time - typically 8 DMA operations of 8 KB at once before having to reprogram the DMA controller for the next operation.
Cheers, Eric
OK, I just tried formatting C: with 8K clusters, and I get the same "A disk read error occurred" when the system reboots.
What did you mean by "XP SP1 was the last MS OS that will allow this?" Do you mean that prior to XP SP2 it was possible to have a boot volume with larger than 8K clusters?
Do you think I should still try tweaking the PCI latency?
Cheers, Eric
It took me some trial and error, but I managed to get the C drive to 64k clusters. You'll need to preformat the drive, and leave enough space for Win7 to create a boot partition of 100mb.
1. That is what my friend told me by phone this morning (I just happened to be speaking with him - I did not call to ask). I do not know otherwise. Throw in an XP <SP2 and try.
2. The setting is there for a reason. It is also a known adjustment to deal with some RAID/controller issues. Do I know if it will affect your situation - no. You have stated that no data is at risk. Estimated time on task: < 5 minutes. Time invested to this point: > 5 minutes.
Quote