Another report: I now have pre7-6 + classzone25. dual PII-450, 148MB
RAM, U2W 10k RPM SCSI.
There is a *huge* improvement in the behaviour of kswapd.
BUT a program that does a lot of disk i/o is still incapable of
getting much more than 1/10 of the performance it did under
2.3.51.
This particular program is an audio hard disk recorder; 2.3.51 allowed
it to record 24 channels without pushing the CPU load much above 40%;
2.3.99-*ANYTHING* makes it impossible to record more than 2 or 3
channels without the kernel causing time delays that lead to
dropouts. the delays are almost certainly caused by kswapd activity.
note that the program itself is fully mlocked.
What on earth was so valuable in the VM changes between .51 and .99
that they were worth killing the performance of any program that wants
to a lot of streaming disk i/o ? is this really just a buglet waiting
to be solved, or was there something in the new VM code that is
fundamentally inimical to such applications ? or does nobody even know
this yet ?
and does anyone understand why 2.2.15 is as bad (or worse) than the
2.3.99 series in terms of this kind of application behaviour ?
--p
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 21:00:18 EST