I have a Lion with 4 666mhz B3 stepping cpus and 4GB ram running Debian
unstable with kernel 2.4.18 and the 020410 ia64 patch (I have the same
problem with 2.4.9-itanium-smp from the archive).
I have a program that reads large files in increments of 81920 blocks.
After about 9600 read calls I get about a dozen reads that take about 3
seconds each. Does anyone have any ideas as to a cause/solution?
(I have 4 other threads working/possibly writing output at the same
time, although in this case only 1 of them would be active at
the same time). I am also running a program that callocs almost all my
ram to make sure none of the file is cached.
I can provide more info if necessary.
Here is a partial log I made (read#/time in seconds):
9597 2.823812008
9598 0.009531000
9599 0.000220000
9600 0.010351000
9601 0.014768000
9602 0.000213000
9603 2.910599947
9604 0.009694000
9605 0.000193000
9606 0.014762000
9607 0.000193000
9608 0.009892000
9609 2.869304895
9610 0.000717000
9611 0.014134000
9612 0.010046000
9613 0.000200000
9614 0.009780000
9615 0.000200000
9616 2.974085093
Thanks for your help.
--
Rick ([email protected])
http://dxr3.sourceforge.net
http://rsub.sourceforge.net
I think the slogan of the fansubbers puts
it best: "Cheaper than crack, and lots more fun."
>>>>> On Thu, 18 Apr 2002 10:06:22 -0400, Rick Haines <[email protected]> said:
Rick> I have a Lion with 4 666mhz B3 stepping cpus and 4GB ram
Rick> running Debian unstable with kernel 2.4.18 and the 020410 ia64
Rick> patch (I have the same problem with 2.4.9-itanium-smp from the
Rick> archive).
Rick> I have a program that reads large files in increments of 81920
Rick> blocks. After about 9600 read calls I get about a dozen reads
Rick> that take about 3 seconds each. Does anyone have any ideas as
Rick> to a cause/solution? (I have 4 other threads working/possibly
Rick> writing output at the same time, although in this case only 1
Rick> of them would be active at the same time). I am also running
Rick> a program that callocs almost all my ram to make sure none of
Rick> the file is cached.
I don't think anyone will be able to help you without a test case.
Do you have a minimal test case that reproduces the problem?
--david
On Fri, Apr 19, 2002 at 02:13:05PM -0700, David Mosberger wrote:
> >>>>> On Thu, 18 Apr 2002 10:06:22 -0400, Rick Haines <[email protected]> said:
>
> Rick> I have a Lion with 4 666mhz B3 stepping cpus and 4GB ram
> Rick> running Debian unstable with kernel 2.4.18 and the 020410 ia64
> Rick> patch (I have the same problem with 2.4.9-itanium-smp from the
> Rick> archive).
>
> Rick> I have a program that reads large files in increments of 81920
> Rick> blocks. After about 9600 read calls I get about a dozen reads
> Rick> that take about 3 seconds each. Does anyone have any ideas as
> Rick> to a cause/solution? (I have 4 other threads working/possibly
> Rick> writing output at the same time, although in this case only 1
> Rick> of them would be active at the same time). I am also running
> Rick> a program that callocs almost all my ram to make sure none of
> Rick> the file is cached.
>
> I don't think anyone will be able to help you without a test case.
> Do you have a minimal test case that reproduces the problem?
I got a response from Andrew Morton that sounds promising. He says
it's probably that "writeback has kicked in, and your reads are stalling
behind the writes". I'll send another email after I try his patch.
--
Rick ([email protected])
http://dxr3.sourceforge.net
I think the slogan of the fansubbers puts
it best: "Cheaper than crack, and lots more fun."