2004-01-22 02:17:37

by Aaron Mulder

[permalink] [raw]
Subject: Strange pauses in 2.6.2-rc1 / AMD64

I tried out 2.6.2-rc1 today, and saw some very strange behavior
with Java. I run a 30s build process (Java/Ant), and it would basically
stop in the middle. If I run a "top" next to it (this was all in X),
during the pauses, the CPUs would go down to nearly 100% idle, and all the
Java processes disappeared from the top of the list. If I moved my mouse
at all, the Java build would wake up for a bit, then stop again 10 or 15
seconds later. The build would easily take several minutes if I didn't
move my mouse much (or hang indefinitely if I didn't move my mouse at
all). If I consistently jiggled my mouse through the entire build, the
performance was approximately what I would expect (35-40s).

I tried the same thing under my normal kernel (SuSE
2.4.21-178-smp) and the build ran continuously with no pauses (35s).

The pauses on 2.6.2-rc1 occured with both 32-bit (Sun) and 64-bit
(Blackdown) Java implementations.

I can't explain it, but if there's any pertinent info I can
provide, I'll be happy to.

Thanks,
Aaron

2x Opteron 248, Tyan Thunder K8W, 4GB RAM, SuSE 9.0 for AMD64


2004-01-22 19:36:58

by Aaron Mulder

[permalink] [raw]
Subject: Re: Strange pauses in 2.6.2-rc1 / AMD64

Thanks for the tip. Unfortunately, disabling Legacy USB Support
didn't stop the pauses for me. It's weird -- if I lay off the mouse, the
CPU goes to 99.7-100% idle, the load goes to 0, and top shows only 1
running task (presumably top itself). Every little while, Java will wake
up and do some more work, then go to sleep again. It seems to happen
during the "jarsigner" phase of the Java build (there are about 20
invocations of that), and it may be launching a new process to sign each
JAR, I'm not sure. This last time, I noticed that Java woke up briefly
when Mozilla hit the top list. It just seems to need some kind of
external stimulus. Keyboard doesn't do it. FYI, I have no USB devices
connected at the moment.

Thanks,
Aaron

On Thu, 22 Jan 2004, Brandon Ehle wrote:
> Aaron Mulder wrote:
>
> > I tried out 2.6.2-rc1 today, and saw some very strange behavior
> >with Java. I run a 30s build process (Java/Ant), and it would basically
> >stop in the middle. If I run a "top" next to it (this was all in X),
> >during the pauses, the CPUs would go down to nearly 100% idle, and all the
> >Java processes disappeared from the top of the list. If I moved my mouse
> >at all, the Java build would wake up for a bit, then stop again 10 or 15
> >seconds later. The build would easily take several minutes if I didn't
> >move my mouse much (or hang indefinitely if I didn't move my mouse at
> >all). If I consistently jiggled my mouse through the entire build, the
> >performance was approximately what I would expect (35-40s).
> >
> >
> >
>
> I was running into the same thing last night. In order to get the
> kernel to boot on my machine, I had also been passing idle=poll on the
> commandline. According to an earlier message, I turned off "Legacy USB
> Support" in the BIOS and the kernel booted fine without passing the
> idle=poll parameter and the pauses no longer happen for me.
>
> This is on a Athlon 64 3000+ on a K8V Deluxe, 1GB RAM, Gentoo for x86_64.
>
> > I tried the same thing under my normal kernel (SuSE
> >2.4.21-178-smp) and the build ran continuously with no pauses (35s).
> >
> > The pauses on 2.6.2-rc1 occured with both 32-bit (Sun) and 64-bit
> >(Blackdown) Java implementations.
> >
> > I can't explain it, but if there's any pertinent info I can
> >provide, I'll be happy to.
> >
> >Thanks,
> > Aaron
> >
> >2x Opteron 248, Tyan Thunder K8W, 4GB RAM, SuSE 9.0 for AMD64
> >
> >
>

2004-01-22 20:03:58

by Zan Lynx

[permalink] [raw]
Subject: Re: Strange pauses in 2.6.2-rc1 / AMD64

On Thu, 2004-01-22 at 12:36, Aaron Mulder wrote:
> Thanks for the tip. Unfortunately, disabling Legacy USB Support
> didn't stop the pauses for me. It's weird -- if I lay off the mouse, the
> CPU goes to 99.7-100% idle, the load goes to 0, and top shows only 1
> running task (presumably top itself). Every little while, Java will wake
> up and do some more work, then go to sleep again. It seems to happen
> during the "jarsigner" phase of the Java build (there are about 20
> invocations of that), and it may be launching a new process to sign each
> JAR, I'm not sure. This last time, I noticed that Java woke up briefly
> when Mozilla hit the top list. It just seems to need some kind of
> external stimulus. Keyboard doesn't do it. FYI, I have no USB devices
> connected at the moment.
>
> Thanks,
> Aaron

jarsigner might be waiting on /dev/random for some cryptographically
random bytes. One source of randomness is mouse interrupts.

If that's the case though, I'm surprised that the keyboard doesn't work.

Does that motherboard have support for a random generator chip? If so,
try loading that module in.
--
Zan Lynx <[email protected]>


Attachments:
signature.asc (189.00 B)
This is a digitally signed message part

2004-01-22 20:35:45

by Aaron Mulder

[permalink] [raw]
Subject: Re: Strange pauses in 2.6.2-rc1 / AMD64

This appears to be the issue. If I load hw_random and run rngd,
then the pauses go away. I may have been confused about the keyboard
-- I probably only typed a couple characters, which I guess wasn't
enough input.

Any idea why I wouldn't see this under 2.4? Perhaps something I
missed moving from /etc/modules.conf to /etc/modprobe.conf? I didn't see
anything about random in there, and /dev/hwrandom didn't even exist (I
had to mknod it).

Thanks,
Aaron

On Thu, 22 Jan 2004, Zan Lynx wrote:
> jarsigner might be waiting on /dev/random for some cryptographically
> random bytes. One source of randomness is mouse interrupts.
>
> If that's the case though, I'm surprised that the keyboard doesn't work.
>
> Does that motherboard have support for a random generator chip? If so,
> try loading that module in.
>

2004-01-22 18:42:53

by Brandon Ehle

[permalink] [raw]
Subject: Re: Strange pauses in 2.6.2-rc1 / AMD64

Aaron Mulder wrote:

> I tried out 2.6.2-rc1 today, and saw some very strange behavior
>with Java. I run a 30s build process (Java/Ant), and it would basically
>stop in the middle. If I run a "top" next to it (this was all in X),
>during the pauses, the CPUs would go down to nearly 100% idle, and all the
>Java processes disappeared from the top of the list. If I moved my mouse
>at all, the Java build would wake up for a bit, then stop again 10 or 15
>seconds later. The build would easily take several minutes if I didn't
>move my mouse much (or hang indefinitely if I didn't move my mouse at
>all). If I consistently jiggled my mouse through the entire build, the
>performance was approximately what I would expect (35-40s).
>
>
>

I was running into the same thing last night. In order to get the
kernel to boot on my machine, I had also been passing idle=poll on the
commandline. According to an earlier message, I turned off "Legacy USB
Support" in the BIOS and the kernel booted fine without passing the
idle=poll parameter and the pauses no longer happen for me.

This is on a Athlon 64 3000+ on a K8V Deluxe, 1GB RAM, Gentoo for x86_64.

> I tried the same thing under my normal kernel (SuSE
>2.4.21-178-smp) and the build ran continuously with no pauses (35s).
>
> The pauses on 2.6.2-rc1 occured with both 32-bit (Sun) and 64-bit
>(Blackdown) Java implementations.
>
> I can't explain it, but if there's any pertinent info I can
>provide, I'll be happy to.
>
>Thanks,
> Aaron
>
>2x Opteron 248, Tyan Thunder K8W, 4GB RAM, SuSE 9.0 for AMD64
>
>