Redhat AS 4 on DL385 G1. Poor disk performance.

[Robert Monn]

I installed RH AS4 r4 on a DL386G1 and run it as an oracle instance. 4GB RAM, mirrored 72GB boot disks and four 300GB 15k rpm disks in the default RAID config for my dataset.

I installed this system straight from the media, meaning that I have the generic cciss driver which while perhaps not being ideal has worked without issue for me previously on similar systems.

Here is the issue. Even though the system's vmstat shows no swapping and the oracle instance is performaing ok the filesystem performance for reads and especially writes is absolutely awful. As is it takes 3 minutes to copy a 70mb file. Watching the stats I can see that the systems iowait is at 95-100 about 75% of the time and at 1-5 the rest of the time. When the wait is near 100 I see that I'm only getting about 4200 blocks per second's worth of throughput.... Lastly I'd like to say that when I first built this box up I saw this behavior after I set some oracle specific kernel parameters for a couple minutes (might be a coincidence) and then it cleared up and the disks performed well: 30mb/sec or more throughput and no lagging. And now that problem is back again.

Something is really wrong here, right? I need to develop a sharp plan for fixing this issue and fast as it is on a production system.

Thanks in advance,

robb

[skt_skt]

What was those kernel parameters? I would like to go have a lok up that way..

[VINY]

Robb,

I am having the same issue with redhat es3 using xp1024 open-l's. I have checked into the slow read/write performance and came across the elvtune command (for read/write settings) but no fixes. Let me know what you came across.

[Robert Monn]

thanks for the replies.

here are what should be the standard oracle kernel params set at startup:

echo 'setting kernel param values for oracle...'


echo 100 3200 100 100 > /proc/sys/kernel/sem
echo 2147483648 > /proc/sys/kernel/shmmax
echo 4096 > /proc/sys/kernel/shmmni
echo 2097152 > /proc/sys/kernel/shmall

echo 65536 > /proc/sys/fs/file-max
ulimit -n 65536

echo 1024 65000 > /proc/sys/net/ipv4/ip_local_port_range
ulimit -u 16384


what do you folks think?

[Bill McNAMARA_1]

do you have this:
http://h30094.www3.hp.com/product.asp?sku=2552852&pagemode=ca

To check via software if you do: rpm -i cpqacuxe-7.50-23.linux.rpm

cpqacuxe -c

This generates a file in the working directory called acucapt.ini
grep Write acucapt.ini
what you don't want to see is
WriteCache= 0
which would mean you don't have the BBWC