Re: PAID-SUPPORT required: boot problem with openssi/1.2.2/fc2 moaning about SYSFS

----- Original Message ----- 
From: <[email protected]>
To: <[email protected]>
Sent: Wednesday, June 28, 2006 3:54 PM
Subject: Re: [SSI-users] PAID-SUPPORT required: boot problem with 
openssi/1.2.2/fc2 moaning about SYSFS


>
>> On 6/28/06, Oyku Gencay <[email protected]> wrote:
>>> I've had the same difficulty but it's only Debian sarge that made the
>>> installation flawless. I'd highly recommend too.
>>>
>> Rather than, or in addition to, individual users paying "gurus" to
>> give them individual hand-holding, I'd like to see small incremental
>> improvements to Debian OpenSSI.  For one thing we've had some
>> difficulties in re-compiling the OpenSSI kernel from the source
>> package.   2.4.22 is very old and the hardware RAID controller we are
>> using was recently discontinued.   (It's still sold by resellers who
>> have some stock: "the 3Ware Escalade 8506 4 Port Serial ATA RAID
>> Card".)   I'd like to use more modern devices and take some baby steps
>> in improving OpenSSI 1.2.2.
>
> As a prospective user of OpenSSI (and a business person interested in a
> product which delivers), I'd say "in addition to".  There are so many
> great open source projects let down by:
>
> - poor documentation
> - infrequent releases/upgrades
> - no or limited support (this forum is the extent of support for openssi)

> Baby steps I agree with - but for god sakes, release -often- (Pg/Mysql are
> excellent examples of how to do things right).  The release schedule of
> openssi is not exactly impressive (stated with understanding of the
> constraints of open source developers).  Requiring users to checkout
> software of cvs (which often doesn't allow you anony access - like today)
> to somehow obtain updates is mind numbing.
>> How many of us what like to see that happen?  How many of us are
>> interested in commercial support?  Or providing commercial support?
>> (Again, all my experience is with Debian OpenSSI 1.2.2.)   I could
>> imagine that supporting other distros wouldn't be so hard but I'm
>> talking "baby-steps" here.
>
> Consider this:  I've spent most of the day googling/surfing/evaluating
> alternatives to openssi (after struggling for -days- with openssi just to
> get the thing to boot) simply because I need a product which not only
> works, but has a modicum of backup/support - paying for it is the least of
> our concerns.
>
> OpenSSI has stupendous potential - I recognise that even though I've yet
> to install/boot successfully (stop your laughing).  *Having* a product is
> one thing, but providing a *package* which includes support avenues for
> people like myself is priceless and will help grow the product (we
> frequently pay open source developers to add customisations which we need
> - but we move on when the developers don't bother responding).
> This is one of the serious failings of open source - lack of professional
> support (certain exceptions: red hat, et al).

Correct, did you consider building a cluster without openssi and using the 
'relative simple' to install pdsh?

MPI (of the manufacturer of the network cards) runs pretty well then.

The big problem of OpenSSI is not so much the difficulty of install, but the 
lack of
performance between the nodes. The latency between the nodes, even if you 
have
a highend network is very poor.

Let's not discuss about things like shared memory migration, as there is 
solutions for
that.

For example i've got a bunch of quadrics QM400 cards here and as soon as i 
cluster them (assuming
i get them to work somehow as i'm not so sure they will function at 
woodcrest nodes / A64 dual core nodes),
so i'm gonna experiment within a few months with stuff like openmosix and 
openssi.

Real important for me is getting that 5 us one way pingpong latency that the 
cards can deliver, though it
might soon go up to 9 us when i manage to find a switch that works for it.

Yet the main point about things like OpenSSI and OpenMosix is and remains 
that i like to install and use it,
yes i'm even prepared hacking OpenSSI so that it can get close to the ping 
pong latencies, as i simply need regurarly
to get a byte or 4 from 1 remote node and need to get that 4 bytes as soon 
as possible. Very regurarly i just want 64 bytes
from a remote node and need to get that as soon as possible; in short the 
PERFORMANCE of the shared memory
is the issue. In fact the entire speedup that my chessprogram gets out of 
the cluster depends upon the latency that's needed
to get 64 bytes from a remote node (usually each time a different remote 
node). So that's a blocking read.

Porting itself the software which works changing 3 lines of code again at a 
512 processor shared memory SGI machine,
to OpenSSI which would get good latencies after hacking, is pretty easy.

Even losing 50% in latency is not a real problem. If i need 20 us to get 4 
random bytes from a remote node
instead of 12 us, i'd sign blindfolded for it.

Changing my code such that it would work on such an OpenMosix/OpenSSI shared 
memory supercomputer is quite peanuts work.

Porting the code to MPI is quite hard on the other hand.

However there are some issues.

OpenBSD is not so ideal.
For parallel machines Linux is superior.
Those drivers from quadrics i bet work easier on a vanilla linux kernel than 
they work on a vanilla BSD.

We'll see how the experiment turns out when i found some money to buy some 
nodes to experiment with.
Just got a dual opteron dual core at this moment  :)

My experience is that usually such stuff takes months before you got it to 
work. I'll have something that's 100x more
powerful than deep blue in that case.

And *that* IMHO is the big problem of open source stuff if you want to use 
it for cutting edge technology.

> As with Platform Rocks/Myql/Pg/redhat/suse/Sendmail.com and others, why
> don't a group of OpenSSI developers formalise some kind of support
> framework (even paid email-only support would be useful)?  A link from
> www.openssi.org (use www.openssi.com!) for "Commercial Support" would give
> it that essential ingredient which is severely lacking.
>
> ... phew, there, I feel better now ;-)
>
>
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job 
> easier
> Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
> _______________________________________________
> Ssic-linux-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/ssic-linux-users
> 


Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642