I have two systems with a serious I/O subsystem based on Areca PCIe
cards, but the results I am getting from simple write benchmarks are
extremely slow.
The details are two 64-bit Opteron systems, one with an Areca PCIe
1210 and the other a PCIe 1220 and both with Seagate SATA II 750 GB
drives. The motherboard is a Tyan S2891 (Thunder K8SRE). The 120
system has 1 GB of PC2700 RAM and the 1220 system has 2GB of PC3200
RAM. The Areca BIOS on both cards is version 1.17a and the firmware
is 1.41.
I have tried two different kernels, 2.6.17 from Ubuntu, which had the
Areca driver added by Ubuntu and 2.6.18 from Gentoo with the Areca
driver added manually from 2.6.19-rc3.
In the initial tests, both computers had a RAID 5/6 configuration,
but to confirm the result, I setup a single Seagate as a pass-through
drive and had the same results. The drive was set to SATA II with NCQ
and the cache was enabled to write-back.
dd if=/dev/zero of=output oflag=sync bs=100M count=1 gives an
excellent result, around 188 MB/sec.
dd if=/dev/zero of=output oflag=sync bs=200M count=1 gives an
excellent result, around 167 MB/sec.
dd if=/dev/zero of=output oflag=sync bs=300M count=1 gives an OK
result, around 117 MB/sec.
dd if=/dev/zero of=output oflag=sync bs=400M count=1 gives a very
poor result, around 35 MB/sec.
These very low numbers around 30 MB/sec persist as I increase the bs number.
As I continue to run the tests, the bs that gives a poor results goes
down to about 200 MB. The results are from the system with the 1220
card. The system with 1210 gives slightly lower numbers overall.
I have also confirmed these low numbers using a benchmark called dm
from the network RAID package, drbd.
Reading from the drives (based on hdparm -tT testing) gives excellent results.
When I use the drive directly connected to the SATA on the
motherboard, all the write tests hover around 56 MB/second regardless
of bs value.
Since both systems are affected, my guess is there is bug in the
Areca driver or with the cards themselves.
--
Maurice Volaski, [email protected]
Computing Support, Rose F. Kennedy Center
Albert Einstein College of Medicine of Yeshiva University
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]