Lustre discuss - Nov 2006 - PPC64 Lustre

Dear all,

 

I have seen people talking about PPC64 Lustre but I have not had much
success.  I (naively) tried to use the intel kernel sources

But they do not compile out of the box.  Is this supported and
operational with either 1.4.7.x latest versions or 1.6b5 ?

 

We have RHEL4.3 PPC64 blades and we would like to test them as Lustre
OSS and Clients.

 

John Belshaw

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20061122/abd38736/attachment-0001.html

On Nov 22, 2006  15:47 -0000, belshaw, john wrote:
> I have seen people talking about PPC64 Lustre but I have not had much
> success.  I (naively) tried to use the intel kernel sources
> 
> But they do not compile out of the box.  Is this supported and
> operational with either 1.4.7.x latest versions or 1.6b5 ?
We do have a customer using PPC64 clients with 1.4.7 in production.
Could you elaborate on your problems?

There is one known issue with PPC64 related to do_div() (patch available
in bug 11028), and another with the use of big-endian find_next_bit()
(patch available in bug 10995, fixed in 1.4.8) for PPC64 OSSes.
> We have RHEL4.3 PPC64 blades and we would like to test them as Lustre
> OSS and Clients.

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

On Wed, Nov 22, 2006 at 01:13:07PM -0700, Andreas Dilger
wrote:
> We do have a customer using PPC64 clients with 1.4.7 in production.
For the record, we successfully ran lustre 1.4.6 on PPC64 with a SUSE kernel.
The testbed was quite small (1 PPC64 OSS, 1 PPC64 MDS & 2 PPC64 clients),
but we hope to run some tests on a bigger cluster soon.
> Could you elaborate on your problems?
> 
> There is one known issue with PPC64 related to do_div() (patch available
> in bug 11028),and another with the use of big-endian find_next_bit()
> (patch available in bug 10995, fixed in 1.4.8) for PPC64 OSSes.
I guess you meant bug 10634 aka "endianness issue in
ext3_mb_generate_buddy()".

AFAIK, the patch from bug 10995 has not landed yet. Besides, this issue should
only occur with kernels older than 2.6.9.

Johann

I have not even got a Lustre kernel to compile yet.

 We are using RHEL4.3 on PPC

Using the kernel source  :
kernel-source-2.6.9-42.0.2.EL_lustre.1.4.7.3.x86_64.rpm

I did rpm -i
Make mrproper
Make menuconfig
cp .config /from /boot

then
I get the following when I try and compile :


.... arch/ppc64/xmon/xmon.c: In function `xmon_core'':
arch/ppc64/xmon/xmon.c:393: warning: passing arg 1 of `__cpus_weight''
discards qualifiers from pointer target type
  CC      arch/ppc64/xmon/ppc-dis.o
  CC      arch/ppc64/xmon/ppc-opc.o
  CC      arch/ppc64/xmon/subr_prf.o
  CC      arch/ppc64/xmon/setjmp.o
  LD      arch/ppc64/xmon/built-in.o
  CC      kernel/sched.o
kernel/sched.c:3104: warning: `interruptible_sleep_on'' is deprecated
(declared at kernel/sched.c:3092)
kernel/sched.c:3138: warning: `sleep_on_timeout'' is deprecated
(declared
at kernel/sched.c:3124)
  CC      kernel/fork.o
kernel/fork.c: In function `do_fork'':
kernel/fork.c:1304: error: `phase'' undeclared (first use in this
function)
kernel/fork.c:1304: error: (Each undeclared identifier is reported only
once
kernel/fork.c:1304: error: for each function it appears in.)
kernel/fork.c:1304: error: `child'' undeclared (first use in this
function)
kernel/fork.c:1304: warning: left-hand operand of comma expression has
no effect
make[1]: *** [kernel/fork.o] Error 1
make: *** [kernel] Error 2



J Belshaw

+44 1737 857502

-----Original Message-----
From: johann lombardi [mailto:johann.lombardi@bull.net] 
Sent: 23 November 2006 10:23
To: belshaw, john; adilger@clusterfs.com
Cc: lustre-discuss@clusterfs.com
Subject: Re: [Lustre-discuss] PPC64 Lustre

On Wed, Nov 22, 2006 at 01:13:07PM -0700, Andreas Dilger
wrote:
> We do have a customer using PPC64 clients with 1.4.7 in production.
For the record, we successfully ran lustre 1.4.6 on PPC64 with a SUSE
kernel.
The testbed was quite small (1 PPC64 OSS, 1 PPC64 MDS & 2 PPC64
clients),
but we hope to run some tests on a bigger cluster soon.
> Could you elaborate on your problems?
> 
> There is one known issue with PPC64 related to do_div() (patch
available
> in bug 11028),and another with the use of big-endian find_next_bit()
> (patch available in bug 10995, fixed in 1.4.8) for PPC64 OSSes.
I guess you meant bug 10634 aka "endianness issue in
ext3_mb_generate_buddy()".

AFAIK, the patch from bug 10995 has not landed yet. Besides, this issue
should
only occur with kernels older than 2.6.9.

Johann