Re: Promise PDC20267 ATA RAID, poor write performance

From: Søren Schmidt <sos_at_DeepCore.dk>
Date: Mon, 13 Sep 2004 11:42:11 +0200
Frode Nordahl wrote:
> On Sep 11, 2004, at 12:49, Søren Schmidt wrote:
>> pizzabox# dd if=/dev/zero of=fill bs=1m count=2000
>> 2000+0 records in
>> 2000+0 records out
>> 2097152000 bytes transferred in 122.908771 secs (17062672 bytes/sec)
> 
> I have done some more test now, reseated cables, disabling other IDE 
> controllers etc. to no avail. I have tested both drives against the ICH2 
> controller with very good results.
> 
> After some fooling around I found the following:
> ar0: ATA RAID1 subdisks: ad4 ad6 status: READY
> # dd if=/dev/zero of=fill bs=1m count=100
> 100+0 records in
> 100+0 records out
> 104857600 bytes transferred in 29.298259 secs (3578970 bytes/sec)
> # atacontrol detach ata2
> # dd if=/dev/zero of=fill bs=1m count=100
> 100+0 records in
> 100+0 records out
> 104857600 bytes transferred in 3.080246 secs (34041957 bytes/sec)
> 
> (this also works when writing 2GB...)
> 
> That is, whenever I run with a single disk or a degraded RAID transfer 
> speeds are ok. I have tried this both ways, so it is not one of the 
> disks / cables that is broken. As soon as I rebuild the RAID again, 
> write speeds drop to 3MB/s.
> 
> Any ideas as of how to debug this?

Sounds strange indeed, there is some performance degradation to be 
expected from RAID1 (you have to write to both disks) but nothing that 
ruins performance this badly. I cannot reproduce it no matter what I 
try, so I'm pretty sure the reason should be found outside ATA.
It could be because the two channels are competing for the PCI bus and 
the arbitration of that fails so you get alot of PCI retries, that is 
outside ATA's reach to control..

-Søren
Received on Mon Sep 13 2004 - 07:42:58 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:11 UTC