2.2.6-pre6 ext3

2.2.6-pre6 with ext3-2.4-0.0.8-246p5
System is a dual PIII/1Ghz 2G memory
Qlogic 2100 Fibre Channel

This is on a raid5 -- since both linux version and ext3 were changes not sure which is the cause yet. I'm waiting for resync to finish to try it on ext2.
tiobench.pl --size 4000
Size is MB, BlkSz is Bytes, Read, Write, and Seeks are MB/secd . -T

         File   Block  Num  Seq Read    Rand Read   Seq Write  Rand Write
  Dir    Size   Size   Thr Rate (CPU%) Rate (CPU%) Rate (CPU%) Rate (CPU%)
------- ------ ------- --- ----------- ----------- ----------- -----------
   .     4000   4096    1  69.09 73.2% 0.909 1.74% 25.75 94.8% 1.247 0.63%
   .     4000   4096    2  23.53 24.4% 1.001 1.63% 25.83 103.% 1.241 0.71%
   .     4000   4096    4  20.07 21.1% 1.157 2.04% 27.32 98.7% 1.233 1.00%

Run #1: /usr/local/bin/tiotest -t 8 -f 500 -r 500 -b 4096 -d . -T Message from syslogd@yeti at Thu Jun 28 08:11:08 2001 ... yeti kernel: JBD: out of memory for journal_head

System locked up hard.

This same test was run multiple times on 2.2.5 without ext3 with much better results too:
Done 6/7/01
Linux yeti 2.4.5 #2 SMP Sat May 26 07:13:52 EDT 2001 i686 unknown

root@yeti:/usr5# tiobench.pl --size 4000
Size is MB, BlkSz is Bytes, Read, Write, and Seeks are MB/secd . -T

         File   Block  Num  Seq Read    Rand Read   Seq Write  Rand Write
  Dir    Size   Size   Thr Rate (CPU%) Rate (CPU%) Rate (CPU%) Rate (CPU%)
------- ------ ------- --- ----------- ----------- ----------- -----------
   .     4000   4096    1  81.48 65.7% 0.499 0.54% 34.33 24.0% 1.481 1.51%
   .     4000   4096    2  77.10 77.2% 0.643 0.96% 33.04 23.6% 1.493 1.81%
   .     4000   4096    4  67.18 72.4% 0.797 1.39% 28.03 20.8% 1.494 1.94%
   .     4000   4096    8  58.29 66.2% 0.954 1.95% 24.89 19.8% 1.498 2.15%

________________________________________
Michael D. Black Principal Engineer
[email protected] 321-676-2923,x203
http://www.csihq.com Computer Science Innovations
http://www.csihq.com/~mike My home page FAX 321-676-2355

Could well be ext3. We're going through the out-of-memory corner cases at present.

I'd be surprised if read performance was significantly different. It shouldn't be.

Thanks - I'll go try tiobench.

-

Trying to recover from ext3 journal failure....swtiched drive back to ext2

Now my fiber channel driver is complaining:
qlogicfc0: no handle slots, this should not happen. hostdata->queue is 2a, inptr: 74
And a bunch more (77,79,7b,7d,7e) after which it locks up completely.

This is while the raid5 is resyncing and it's trying to do an e2fsck at the same time.

I'm now going to try letting the resync complete before doing the e2fsck.

Looks like I'm just running into cascading problems here...sigh...

________________________________________
Michael D. Black Principal Engineer
[email protected] 321-676-2923,x203
http://www.csihq.com Computer Science Innovations
http://www.csihq.com/~mike My home page FAX 321-676-2355

Could we please have more details on the "journal failure"?

I'd suggest you fix one problem at a time: get the driver and raid working solidly with ext2, then try ext3.

-