[parisc-linux] clustering ??

[parisc-linux] clustering ??

[John Hagreis]

Hi there,

I've got two B132L+ systems running with the 0.6 release. I wondered if
there was someone who has experience with clustering these systems .....

Anyone .....

Tia,

John Hagreis.

Re: [parisc-linux] clustering ??

[Carlos O'Donell Jr.]

On Mon, Jun 11, 2001 at 09:50:54PM +0200, John Hagreis wrote:
> Hi there,
> 
> I've got two B132L+ systems running with the 0.6 release. I wondered if
> there was someone who has experience with clustering these systems .....
> Anyone .....
> Tia,
> 
> John Hagreis.

John,

Clustering -> HA Systems -> Clustered Services -> Linux Virtual Server ?
Clustering -> Computational Clustering -> Beowulf / COW / NOW ?

I'll assume the later.

We have a cluster of ~48 715/50 Apollo systems.

Our current bottleneck (will be working on this tommorow evening) is
not having a functional rsh for MPI to use (MPICH 1.2.1).

Our setup:

- 3x 10/100 24 Port Switches, Baystack E450's
- NIS / NFS / Gateway Machines (x86 boxes)
- 2x 8' Racks of 48 712/50 PA-RISC Boxes (can you say warm :)

Visit: http://www.baldric.uwo.ca/index.php3
For our latest exploits (recipies coming soon).

What are you looking to do with your systems?

You might want to check out http://www.beowulf.org
(Or even join the mailing list! Excellent discussion for those interested).

Clustering using PA-RISC hardware is the same as clustering using
any other hardware, with the expection that:

- OS is under development (stability may vary)
- Need to understand low level OS functionality.

If anyone else on the list is working on clustering using PA-RISC
boxes, please share the stories.

As far as we know, we are the only group doing something as crazy
as using unsupported PA-RISC hardware to run a beowulf.

War stories? Success stories?


Thanks to Byron Sonne for donating his 712/60 box for testing!

Cheers,
Carlos O'Donell
-------------------------
Baldric Project
http://www.baldric.uwo.ca
-------------------------

Re: [parisc-linux] clustering ??

[Aaron Macks]

> If anyone else on the list is working on clustering using PA-RISC
> boxes, please share the stories.
> 
> As far as we know, we are the only group doing something as crazy
> as using unsupported PA-RISC hardware to run a beowulf.
> 
> War stories? Success stories?
Well I have a cluster of misc machines and 8 HPPA boxes, the problems
is that they are all 715/33's, so adding them to the mix is not yet
> possible.  The good news is that every other flavor of Linux I
> use(PPC, i386,i586,i686,m68k,SPARC) and Irix all play nice with each
> other and I can't figure out why HPPA linux would be much different.

As far as RSH goes, have you tried to use SSH as a transparent
standin(sometimes you need to replace rsh with 'ssh -x' for everything
to be happy.  I use LAM instead of MPICH, but replacing rsh for me is
as simple as a configure-time selection.
Aaron
http://wiglaf.cs-i.brandeis.edu

Re: [parisc-linux] clustering ??

[Carlos O'Donell Jr.]

[snip]
> is that they are all 715/33's, so adding them to the mix is not yet
> possible.  The good news is that every other flavor of Linux I
> 
> As far as RSH goes, have you tried to use SSH as a transparent
> standin(sometimes you need to replace rsh with 'ssh -x' for everything
> to be happy.  I use LAM instead of MPICH, but replacing rsh for me is
> as simple as a configure-time selection.
> Aaron
> http://wiglaf.cs-i.brandeis.edu
>

Aaron,

Yes, ssh does work as a direct replacement for rsh.
However, I believed that ssh was in haitus (due to OpenSSL not working). 

With a little work the 715/33's should work.
We have various 715/33's that hobble on a leg!
If you have any serious problems just email us, or spam the list ;)

Cheers,
Carlos

Re: [parisc-linux] clustering ??

[Carlos O'Donell Jr.]

On Tue, Jun 12, 2001 at 10:07:39PM +1000, Pete wrote:
> > We have a cluster of ~48 715/50 Apollo systems.
> 
> Great!! :)  I have one in the shed that I need a serial cable for... :)
> 
> > Visit: http://www.baldric.uwo.ca/index.php3
> 
> Did that.. nice site... didn't quite find out what youa re planning on 
> with the cluster... then again I'm not the most observant...
> Pete.

Pete,

We used our first PPC cluster to enumerate Prime Alternating Knots
to 18 crossings. 
http://baldric.uwo.ca/article.php3?section=baldric&article=knots

After that, it's been in use, on and off by other students.

Our second cluster, the PA-RISC system, will be used for validating some
quantum gravity research (interesting stuff).

Along with some other pet projects.

We are currently still in the setup phase.

- Putting together root tarball to use as node install
- Get rsh to work

And yes, SSH will work as a direct replacement for RSH in MPICH.

However, I believed that SSH was currently broken on PA-RISC due to OpenSSL.

Haven't tried compiling non-open SSH :)
I think we might give it a shot tonight.


Cheers,
Carlos O'Donell
-------------------------
Baldric Project
http://www.baldric.uwo.ca
-------------------------

Re: [parisc-linux] clustering ??

[Tom]

On Tue, 12 Jun 2001, Carlos O'Donell Jr. wrote:
[snip]
> And yes, SSH will work as a direct replacement for RSH in MPICH.
>
> However, I believed that SSH was currently broken on PA-RISC due to OpenSSL.
>
> Haven't tried compiling non-open SSH :)
> I think we might give it a shot tonight.

Not sure what "broken" may mean, but my 0.9 install has a working SSH
server running on it. Mine *was* installed with contrib and non-free
however, so it may be that a "pure" install doesn't have it.

If I remember right I was asked during the .deb install if I wanted to run
it as a server, answered yes and it's worked ever since (Same as a x86
Debian install). The 0.9 install sure does work better than the 0.6 one
did - no seg faults or errors...

Re: [parisc-linux] clustering ??

[Matthew Wilcox]

On Tue, Jun 12, 2001 at 08:32:39AM -0800, Tom wrote:
> On Tue, 12 Jun 2001, Carlos O'Donell Jr. wrote:
> [snip]
> > And yes, SSH will work as a direct replacement for RSH in MPICH.
> >
> > However, I believed that SSH was currently broken on PA-RISC due to OpenSSL.
> >
> > Haven't tried compiling non-open SSH :)
> > I think we might give it a shot tonight.
> 
> Not sure what "broken" may mean, but my 0.9 install has a working SSH
> server running on it. Mine *was* installed with contrib and non-free
> however, so it may be that a "pure" install doesn't have it.

the current openssh in the archives has a known bug (which i haven't seen
discussed on the list), which is that X forwarding does't work right.
at least not on my machine :-)

A bug which you'll see ssh trigger (but is actually in libc) is that it'll
set your rlimit to 2GB instead of infinity.  this should be fixed RSN.

-- 
Revolutions do not require corporate support.