Re: High Utilization kswapd and kupdated in Large memory system

[Marcelo Tosatti]

On Thu, 23 Oct 2003, vwake wrote:

> Hi,
> On large memory machine ( 32 GB RAM) any continuous disk activity puts kswapd and kupdated to 99% utilization and eventually ( in about 10 mins) brings down the whole machine.
> 
> The machine works well when the memory was set at 4GB using the 'mem="
> entry on grub.conf. The same disk operations succeed when kept in this
> config.
> 
> 
> The kernel is version 2.4.22 stable downloaded from kernel.org, and compiled with 64GB option enabled. The machine is installed with Redhat Linux 9.0.
> 
> No errors gets logged anywhere and also there are no console error
> messages. The machine eventually gets locked up, and only a hard reset
> will bring it back into shape. I do not suspect the hardware because it
> is reproduceable in a different machine with similar hardware config.
>
> 
> The hardware config as below:
> 
> Dell PE 6650 / 4* Xeon 2GHz / 32 GB RAM / 500 GB Raid 5 on PERC 3DC ( AMI megaraid driver) 6 DISCS / 2* Broadcom 100/1000 NIC ( bcm5700 driver) 
> 
> The system has / and /boot in ext3 and rest in ReiserFS. The 500 GB data partition is in ReiserFS.
> 
> The problem is reproducible by just copying some large data ( 3-5 GB ) to any of the filesystems.
> One observation ( may not be useful) is that the symptoms starts after the cached memory in /proc/meminfo goes beyond 16GB !
> 
> 
> I had tried changing the "elvtune" parameters to " -r 4096 -w 8192" as adviced in some of the archived maillist mails. But this didnt help !
> 
> Please let me know if any further information is needed from the machine.

Hi, 

Can you please try 2.4.23-pre?

There have been significant highmem balancing changes.



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/