Clustering Software on CD

[Albert Yang]

Hello, I joined the list with hopes of finding more stuff out about Netbsd
with respect to clustering.

I have been using OpenMosix on Clusterknoppix; I wanted however a BSD
solution that was similiar.

First, the only thing I found was Cluster-it, and even then, it's webpage
seems to be down, so I only read the cache'd version from google.

Second, there seems to be very little documentation about how to use
netbsd in a clustered environment, and what little documentation is
haphazard at best. I read about netbsd cluster used in some Marathon
picture splicing, but it didn't say anything about what software was used
for clustering.

Finally, I'm hoping there's a OpenMosix like solution for BSD, in which I
can drop in a CD, and get it up and running without having to install nor
the need for HD's. Is there anything like that for BSD??

I'm more than willing to help out in projects, but I've not even found a
project to help out with.

Thanks in advance.
Albert

[GuRU]

Hey Albert
check the following site
http://guinness.cs.stevens-tech.edu/~jschauma/hpcf/

also check this introductory email from Jan Schaumann
http://mail-index.netbsd.org/tech-cluster/2003/10/20/0001.html

./i'khala

Post by Albert Yang
Hello, I joined the list with hopes of finding more stuff out about Netbsd
with respect to clustering.
I have been using OpenMosix on Clusterknoppix; I wanted however a BSD
solution that was similiar.
First, the only thing I found was Cluster-it, and even then, it's webpage
seems to be down, so I only read the cache'd version from google.
Second, there seems to be very little documentation about how to use
netbsd in a clustered environment, and what little documentation is
haphazard at best. I read about netbsd cluster used in some Marathon
picture splicing, but it didn't say anything about what software was used
for clustering.
Finally, I'm hoping there's a OpenMosix like solution for BSD, in which I
can drop in a CD, and get it up and running without having to install nor
the need for HD's. Is there anything like that for BSD??
I'm more than willing to help out in projects, but I've not even found a
project to help out with.
Thanks in advance.
Albert

[Jan Schaumann]

Post by Albert Yang
Second, there seems to be very little documentation about how to use
netbsd in a clustered environment, and what little documentation is
haphazard at best. I read about netbsd cluster used in some Marathon
picture splicing, but it didn't say anything about what software was used
for clustering.

I can't speak for that environment myself (Hubert, can you chip in?),
but in the cluster that I maintain, we don't really use anything fancy
at all. We have a shared filesystem, read-only local filesystem on the
nodes and use MPI and PVM to parallelize.

Another cluster uses the same approach, just with diskless nodes that
netboot and NFS mount everything that's needed.

Other people have reportedly used Sun Grid Engine, but I haven't had the
time to set this up myself.

It's true that documentation is lacking... as usual, this task falls
back to all users. :-) If you have a specific setup, why not document
it and make it available to all of us?

-Jan

[Hubert Feyrer]

Post by Jan Schaumann

Post by Albert Yang
Second, there seems to be very little documentation about how to use
netbsd in a clustered environment, and what little documentation is
haphazard at best. I read about netbsd cluster used in some Marathon
picture splicing, but it didn't say anything about what software was used
for clustering.

I can't speak for that environment myself (Hubert, can you chip in?),

The "Marathon Cluster" used hand-made load balancing for the first phase
of the project (splitting mpeg streams into images), and mpeg_encode,
which we used in the second step, did all the load balancing over a given
number of machines itself. Quite straight forward.

Of course we had a common filesystem (NFS) for all machines, and the
applications were available on all clients, but that's what you have to
provide in any cluster. If you know how to run a cluster of lab machines
for students, you should have all the knowledge you need to run such a
cluster, too.

Dunno what else to tell - all this "clustering" is mostly a hype IMO. :)

As for the documentation, there's a bunch available at
http://www.feyrer.de/marathon-cluster/intro-en.html, documentation (both
german and english language) is linked at the german language cluster page
at the bottom of http://www.feyrer.de/marathon-cluster/intro.html, esp.
http://www.feyrer.de/marathon-cluster/mc-paper.pdf.

If you need more information, I'd be interested in knowing what kind of
information you need.


- Hubert

[Albert Yang]

I really would prefer the BSD's because of stability; BUT, OpenMosix can
do auto-detect, which is a big benefit. That means for drone machines,
you just drop the CD in, and then it auto-detects and runs.

HA is difficult, you need mating of hardware and software... That's why
HA is difficult to do for linux etc.. compared to Sun or HPUX...

Does anyone know if there's an open source solution (for any platform)
that has a HA approach to clustering similar to Sun Cluster (or openvms) ?
Sun Cluster is good stuff.
--
jht

[jarkko teppo]

Post by Hubert Feyrer
Dunno what else to tell - all this "clustering" is mostly a hype IMO. :)

Off-topic:

Does anyone know if there's an open source solution (for any platform)
that has a HA approach to clustering similar to Sun Cluster (or openvms) ?

Sun Cluster is good stuff.

[Hubert Feyrer]

Post by Hubert Feyrer
Dunno what else to tell - all this "clustering" is mostly a hype IMO. :)

Does anyone know if there's an open source solution (for any platform)
that has a HA approach to clustering similar to Sun Cluster (or openvms) ?
Sun Cluster is good stuff.

So what's Sun Cluster's approach to HA? I don't know it - technical
details?


- Hubert

[jarkko teppo]

Post by Hubert Feyrer

Sun Cluster is good stuff.

So what's Sun Cluster's approach to HA? I don't know it - technical
details?

Light on technical details: http://www.sun.com/software/cluster/faq.xml
There's more technical stuff at blueprints.sun.com.

Basically it's a framework which you can use to build a HA and/or
scalable system for different applications from apache and bind to
oracle and other heavyweights.

Lots of nice features like global devices, global filesystem,
network failover, resource groups etc.etc...

Basically in a two node cluster might have:
-shared storage (needed in a two node cluster due to quorum, FC/scsi)
-NIC groups (done with IPMP, in.mpathd)
-Logical network addresses on top of IPMP groups which can then
fail-over from nic-to-nic (or be scalable) or from host-to-host.
-Similar fail-over for storage and volumes.
-application agents (from Sun or DIY)

These resources are then gathered into resource groups and you
can specify different actions to do in case of resource or
resource group failure.

There are many nice details in there, for example split-brain
detection and admirable tendency of always failing (panicing)
(a) node(s) if data integrity is at risk.

It's quite flexible which makes explaining it pretty difficult:)

IPMP should be doable (SMOP), shared storage (not NFS) might be
a tad more difficult. Some people use firewire disks for that esp.
with Linux and Oracle 10g.. makes it possible to build a cheap
development rig.

I'm rambling and incoherent. Sorry,