Hi everyone,

I bougth a new mainboard (with Intel chipset i865PE, ICH5) + new ram and installed the new kernel v2.6.12rc4-mm2 with reiserfs4 harddrives. Then I checked up hdparm and changed some options for every HDD (which are all ATA100):

hdparm -c3 -X69 -u1 -d1 /dev/hdX

according to INTEL these settings are correct.

After testing with hdparm -tT -dev-hdX I got speeds of ~50MB/second for every hard drive.

While transfering some data from one drive to another the file transfer suddenly STOPs for 1 or 2 seconds and then continues with a much lower transfer rate of 5 - 10 MB.

What the heck is wrong ?
Is it reiserfs4?
Do I need to disable some general chipset code in kernel configuration?

No problem in win32 though.
I appreciate any help. Thanks!

The disk utility hdparm just gives you raw benchmarks. Copying from a filesystem is different issue. Depending on the latency of the filesystem, what mount options used, and what formatting options used you will not get the same results as hdparm. You can include atime and notail when mounting them to increase the speed. If you are using LILO, include notail because if you do not it will corrupt Reiser partitions. If you are using grub it does not have this flaw.

I use XFS because it has a much faster throughput ranging from 30 MB per second up to 100 MB per second.

For the pause for a second or two, probably your system is going through dirty buffers. Use sysctl to optimize the kernel for disk performance.

I leeched Gentoo RR LiveDVD with Reiser4 support. No problem there with udma ide transfers from HDDs I mentioned. I checked up hdparm settings of this livedvd for the hddrives. I haven't changed my own bootup defaults to these settings yet but I am quite sure that this is the problem.

IMHO lilo cannot compete with GRUB - but otherwise I must say I never used lilo for my LFS system - it's been grub in the beginning.

I know about the hdparm tests - raw access - no real transfer from disk to disk - eventually of no use for me here.

Thanks for the tip with dirty buffers and mount options with atime and notail. I will check this for next reboots and testing phase.

Do you think my ICH5 has something to do with it?

My previous mainboard had ICH2.
ICH5 seems to be a lot different because there is no Application Accelerator Software for it.

(ICH = I/O Controller Hub; equal to southbridge; Intel used expression ICH since i810 chipset)

I get the same with any chipset such as INTEL and VIA. Also I see it more with USB hard drives or USB to IDE converters. If someone can add a patch so Linux is able to switch to BIOS for disk operations. The BIOS knows more about the system than Linux will ever will. Unfortunately, older BIOS do not detect large hard drives.

I use the option sync as well as noatime and notail to increase the performance during video recordings on an AMD Athlon 700 Mhz 'Classic' system. If I did not, frames will be dropped non-stop. The filesystem that I used for video recording is ReiserFS because the AMD Athlon 700 Mhz 'Classic' system used kernel version 2.4.19 which does not have real-time XFS.

You can try to use the third kernel scheduler which will balance the load of all programs. I think this scheduler closely resembles Windows scheduler.

running top in another virtual console gave me 40-75% CPU usage for process pdflush.

Really..after reading a little bit about it I sould check if it ococurs with other filesystems or if it even has to do with Intel HT SMP

running smoother with Kernel 2.6.12-rc5-mm2

I'll have a look at http://ftp.namesys.com/pub/reiser4-for-2.6/

maybe an infinite loop which is fixed here:
http://ftp.namesys.com/pub/reiser4-f...er4-only.patch