samba - Oct 2001 - Samba 2.2.1a On Dual Xeon 866 (1Gb RAM) RedHat Linux 7.1 Server (800 users 300Gb of data)

Hi,

We currently run a Sun E3000, Solaris 2.6, 2Gb RAM, 3x250MHz CPU, 800 users and
300Gb of data. This runs fine. However, we decided to migrate to a Linux based
system as above and this has failed miserably.

Executables are very slow to run from Win9x clients and can be forced to
complete execution by kill -9 on the process from Unix, but are OK from
WinNT/2K. However, an NFS share of the same Samba share is absolutely OK when
connected via NFS to our Solaris server and then reshared via Samba 2.0.6. It
appears to be related to 16-bit files which do not display an icon in a Windows
Explorer view. However, we were not able to replicate this problem using Samba
2.2.1a on another RedHat 7.1 server we use as a Samba/LPRng print server.

On our E3000 we normally have between 600 and 800 smbd processes running. When
we went live on our Linux server these were up to 2000 (and growing) with a load
average of 171 (and growing) and kernel CPU usage at 90%+ before we bailed out
and went back to our E3000.

The following error was recorded at the height of the problem:

Oct 15 09:06:59 hydra kernel: __alloc_pages: 1-order allocation failed.
Oct 15 09:06:59 hydra kernel: __alloc_pages: 1-order allocation failed.
Oct 15 09:06:59 hydra kernel: __alloc_pages: 1-order allocation failed.
Oct 15 09:07:00 hydra kernel: __alloc_pages: 1-order allocation failed.

It is clear that there may be some tweaks we need to make to the Samba config.
If there is a configuration that you would suggest for our set-up then please
advise. I am happy to provide any additional information or config details.
However, if I should seek assistance in a different forum please point me in the
right direction. It is surprisingly difficult to find information on anyone who
runs a Samba implementation of this size which I wouldn't have thought was
that unusual.

If we are unable to resolve the difficulties then we will be compelled to
replace our Linux/Samba file-server environment with a Win2K solution due to
time constraints for getting a new system in place to replace the Solaris
solution.

Kind Regards,
=======================Daniel Van Wieren
IST Infrastructure Team
Tait Electronics Ltd
Christchurch
Ph 64-3-3570763
-------------- next part --------------
HTML attachment scrubbed and removed

Hi,

this __alloc_pages error is a kernel error, if you use the Highmem
option from the kernel. There were some problems with the Virtual Memory
System. It's a known issue and with kernel 2.4.12 this should not happen
anymore (completely new VM-Code).

Hope that helps

Daniel Van Wieren schrieb:
> Hi, We currently run a Sun E3000, Solaris 2.6, 2Gb RAM, 3x250MHz CPU,
> 800 users and 300Gb of data. This runs fine. However, we decided to
> migrate to a Linux based system as above and this has failed
> miserably. Executables are very slow to run from Win9x clients and can
> be forced to complete execution by kill -9 on the process from Unix,
> but are OK from WinNT/2K. However, an NFS share of the same Samba
> share is absolutely OK when connected via NFS to our Solaris server
> and then reshared via Samba 2.0.6. It appears to be related to 16-bit
> files which do not display an icon in a Windows Explorer view.
> However, we were not able to replicate this problem using Samba 2.2.1a
> on another RedHat 7.1 server we use as a Samba/LPRng print server. On
> our E3000 we normally have between 600 and 800 smbd processes running.
> When we went live on our Linux server these were up to 2000 (and
> growing) with a load average of 171 (and growing) and kernel CPU usage
> at 90%+ before we bailed out and went back to our E3000. The following
> error was recorded at the height of the problem: Oct 15 09:06:59 hydra
> kernel: __alloc_pages: 1-order allocation failed.
> Oct 15 09:06:59 hydra kernel: __alloc_pages: 1-order allocation
> failed.
> Oct 15 09:06:59 hydra kernel: __alloc_pages: 1-order allocation
> failed.
> Oct 15 09:07:00 hydra kernel: __alloc_pages: 1-order allocation
> failed. It is clear that there may be some tweaks we need to make to
> the Samba config. If there is a configuration that you would suggest
> for our set-up then please advise. I am happy to provide any
> additional information or config details. However, if I should seek
> assistance in a different forum please point me in the right
> direction. It is surprisingly difficult to find information on anyone
> who runs a Samba implementation of this size which I wouldn't have
> thought was that unusual. If we are unable to resolve the difficulties
> then we will be compelled to replace our Linux/Samba file-server
> environment with a Win2K solution due to time constraints for getting
> a new system in place to replace the Solaris solution. Kind
> Regards,=======================> Daniel Van Wieren
> IST Infrastructure Team
> Tait Electronics Ltd
> Christchurch
> Ph 64-3-3570763