Strange pauses in 2.6.2-rc1 / AMD64

Strange pauses in 2.6.2-rc1 / AMD64

[Aaron Mulder]

	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

Re: Strange pauses in 2.6.2-rc1 / AMD64

[Aaron Mulder]

	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
> >  
> >
> 

Re: Strange pauses in 2.6.2-rc1 / AMD64

[Zan Lynx]

[-- Attachment #1: Type: text/plain, Size: 1182 bytes --]

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 <zlynx@acm.org>

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Strange pauses in 2.6.2-rc1 / AMD64

[Aaron Mulder]

	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.
> 

Re: Strange pauses in 2.6.2-rc1 / AMD64

[Brandon Ehle]

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
>  
>

Status of Athlon 64 K8V-D support was (Re: Strange pauses in 2.6.2-rc1 / AMD64)

[Brandon Ehle]

Stephan Maciej wrote:
> On Friday 23 January 2004 03:42, Brandon Ehle wrote:
> 
>>This is on a Athlon 64 3000+ on a K8V Deluxe, 1GB RAM, Gentoo for x86_64.
> 
> 
> Hi,
> 
> I am playing with the idea of getting exactly this HW config in the near 
> future. I haven't found much user reports concerning running Linux - possibly 
> Gentoo - on such a system (Athlon64 on an K8V-D mobo). Do your SATA 
> controllers work, esp. when booting an AMD64 kernel? How's the rest of the 
> on-board hardware behaving? Is it all working? What memory brand did you buy?
> 
> Thanks a lot in advance,
> 
> Stephan
> 

First off, I'm only running the 2.6 kernel and haven't even tried the
2.4 kernel on this board.

The board comes with 2 software RAID SATA controllers in addition to the
PATA ones.  I've only gotten one of the two SATA controllers (VIA) to
work, and I'm not using RAID (just a single 10,000RPM drive).  I believe
Gentoo's gentoo-dev-sources kernel has a driver for the other controller
(Promise FastTrak 378), but I haven't tried it yet because I'm already
reaching the peak of the drive with the controller I have working.  One
thing to note is that the harddrive does run about 20% slower in x86_64
kernels currently, but I'm not sure why that is.

I went with 2 matched sticks of Geil DDR433 512MB (PC-3500), but one of
the weird things about the board is that if I use the other 2 memory
slots (4 in total), the RAM will only go DDR333 instead of DDR400 (I
picked DDR433 in case I want to try overclocking someday).  I'm not sure
if that's specific to this board or if all of them have that problem
(due to the onchip memory controller).

The onboard 1000MB 3COM card (sk98lin driver) is pretty typical of all
3COM cards and fails under extremely high load cases (I've never used a
3COM card that didn't fail under my high load conditions), so I'm using
a second PCI 100MB NIC (tulip driver) to talk to my high load device and
the 1GB for my Internet connection.  I don't think it is the sk98lin
driver at fault because the card takes a nosedive under high load
conditions in Win2K too (reaching the theorhetical peak of the card).

I don't use the onboard sound as I have an Audigy 2 (OSS driver, ALSA
doesn't work), so I don't know anything about that.  All 8 USB ports
(uhci-hcd driver) work fine.  I don't have any firewire devices, but it
finds the port ok.  For video I'm running a GeForce FX 5900 Ultra
(nvidia proprietary) and that works good too (using the amd640-agp driver).

Everything hardware seems to work equally well in x86 or x86_64 mode,
but I'm not running in x86_64 mode anymore because of all the userspace
problems.  The only issue I ran into when going to x86_64 was needing to
turn off "Legacy USB Support" in the BIOS so I didn't have to pass
idle=poll on the command line which then stops me from getting the 3-5
second pauses when compiling or running benchmarks.

Gentoo was my choice of OS for x86_64 mode because none of the other
distributions have a decent sized package set for x86_64 yet, but I've
fallen back to debian-unstable (for better stability!) until more
packages support x86_64.

Re: Status of Athlon 64 K8V-D support was (Re: Strange pauses in 2.6.2-rc1 / AMD64)

[Malte Schröder]

[-- Attachment #1: Type: text/plain, Size: 4056 bytes --]

On Saturday 24 January 2004 11:39, you wrote:
> Stephan Maciej wrote:
> > On Friday 23 January 2004 03:42, Brandon Ehle wrote:
> >>This is on a Athlon 64 3000+ on a K8V Deluxe, 1GB RAM, Gentoo for x86_64.
> >
> > Hi,
> >
> > I am playing with the idea of getting exactly this HW config in the near
> > future. I haven't found much user reports concerning running Linux -
> > possibly Gentoo - on such a system (Athlon64 on an K8V-D mobo). Do your
> > SATA controllers work, esp. when booting an AMD64 kernel? How's the rest
> > of the on-board hardware behaving? Is it all working? What memory brand
> > did you buy?
> >
> > Thanks a lot in advance,
> >
> > Stephan
>
> First off, I'm only running the 2.6 kernel and haven't even tried the
> 2.4 kernel on this board.
>
> The board comes with 2 software RAID SATA controllers in addition to the
> PATA ones.  I've only gotten one of the two SATA controllers (VIA) to
> work, and I'm not using RAID (just a single 10,000RPM drive).  I believe
> Gentoo's gentoo-dev-sources kernel has a driver for the other controller
> (Promise FastTrak 378), but I haven't tried it yet because I'm already
I currently use an IBM/Hitachi-SATA-drive attached to the Promise-Controller 
on the K8V-D without problems. I use the Promise-driver from the SATA 
support. The controller is configured to work as a plain IDE-Controller.
Kernels I tried are 2.6.1 and 2.6.2-rc1-mm2.

> reaching the peak of the drive with the controller I have working.  One
> thing to note is that the harddrive does run about 20% slower in x86_64
> kernels currently, but I'm not sure why that is.
Haven't tried 64bit-support yet ...

>
> I went with 2 matched sticks of Geil DDR433 512MB (PC-3500), but one of
> the weird things about the board is that if I use the other 2 memory
> slots (4 in total), the RAM will only go DDR333 instead of DDR400 (I
> picked DDR433 in case I want to try overclocking someday).  I'm not sure
> if that's specific to this board or if all of them have that problem
> (due to the onchip memory controller).
>
> The onboard 1000MB 3COM card (sk98lin driver) is pretty typical of all
> 3COM cards and fails under extremely high load cases (I've never used a
> 3COM card that didn't fail under my high load conditions), so I'm using
> a second PCI 100MB NIC (tulip driver) to talk to my high load device and
> the 1GB for my Internet connection.  I don't think it is the sk98lin
> driver at fault because the card takes a nosedive under high load
> conditions in Win2K too (reaching the theorhetical peak of the card).
>
> I don't use the onboard sound as I have an Audigy 2 (OSS driver, ALSA
> doesn't work), so I don't know anything about that.  All 8 USB ports
> (uhci-hcd driver) work fine.  I don't have any firewire devices, but it
> finds the port ok.  For video I'm running a GeForce FX 5900 Ultra
> (nvidia proprietary) and that works good too (using the amd640-agp driver).
>
> Everything hardware seems to work equally well in x86 or x86_64 mode,
> but I'm not running in x86_64 mode anymore because of all the userspace
> problems.  The only issue I ran into when going to x86_64 was needing to
> turn off "Legacy USB Support" in the BIOS so I didn't have to pass
> idle=poll on the command line which then stops me from getting the 3-5
> second pauses when compiling or running benchmarks.
>
> Gentoo was my choice of OS for x86_64 mode because none of the other
> distributions have a decent sized package set for x86_64 yet, but I've
> fallen back to debian-unstable (for better stability!) until more
> packages support x86_64.
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
---------------------------------------
Malte Schröder
MalteSch@gmx.de
ICQ# 68121508
---------------------------------------


[-- Attachment #2: signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]