Unsolved

This post is more than 5 years old

6 Posts

27315

July 15th, 2008 12:00

Configuring a multi chassis HPC cluster with 2950 head node

I think it would be beneficial to provide a video documenting how to configure a basic HPC cluster with a 2950 head node and 2 or more m1000e chassis.

We recently purchased a similar configuration, installer got everything bolted into the rack and powered on, but nothing past that. Being new to the blade arena, I found it a bit confusing how all the parts connected to form the HPC cluster.

For example, the 2950 normally would connect eth1 to the public network and eth0 to the private network, and the compute nodes would also connect to the private network. With the blades, each chassis has a 20 port GigE switch, 16 internal and 4 external, how are these supposed to be connected and configured along with eth0 on the 2950 to make up the private network.

We sysadmins can certainly figure it out, but it would make for a nice video, especially for those who are moving to blades in the future for their HPC needs.

8 Posts

July 16th, 2008 15:00

flakrat,

Can you be a little more specific about your situation? Did the installer just "rack-n-stack" and not install and software? Did you also not buy any core switches for the cluster?

I'd like to better understand the baseline of the situation, so I can see if we have anything to help you. Are you looking for a video to describe the hardware installation portion or the software side or both?

Thanks!

Jeff

6 Posts

July 16th, 2008 16:00

On a related note, I spent a lot of time trying to get CMC daisy chaining to work between the two m1000e chassis. I finally talked to someone in support who says that it doesn't work at the moment and support will be added in a future firmware update.

The solution, skip the daisy chain and wire both CMC's directly to your LAN.

6 Posts

July 16th, 2008 16:00

Howdy, The confusion was created by the "rack-n-stack" guy forgetting to install the stacking modules in the m6220 switches installed in the two m1000e chassis. I found the modules underneath a stack of manuals and paperwork left over from the install O_o

Once I found the modules, the whole configuration made a lot more sense. Without the modules, staring at the back of the rack we knew something wasn't right, but just couldn't put our finger on it.

8 Posts

July 16th, 2008 17:00

Got it. So did you buy the software with the cluster or did you install your own? (Out of curiosity f you did install your own, what did ouy use?).

Are you looking for some kind of instruction video on how to connect everything together? (I don't know if we have something like that, but it sounds like something that other people could use).

Thanks!

Jeff

6 Posts

July 17th, 2008 08:00

I'm installing Rocks 5 on the cluster, no software was included in the purchase. We have several Rocks and Platform OCS clusters, so that's what the admins and users are familiar with (as well as a Blue Gene, which is a completely different animal :-)

Yes, a video that shows how to connect up at least two m1000e into a single HPC cluster might be helpful for those brand new to blade technology. Now that I've been messing with the hardware for a couple days, it seems trivial, but reading or viewing something like that ahead of time would have been a nice introduction.

I'm really enjoying the delltechcenter.com site, keep up the good work.

Mike

8 Posts

July 17th, 2008 10:00

Mike,

Thanks for the reply and the clarification. That really helps me. We may have some kind of video about blade deployment, but they are probably not for HPC (which makes things a bit different). I'll check into this. If you can send me an email with your observations/comments/etc about your experience, that would help me. You can email me at jeffrey_layton_at_delll.com.

BTW - I'm trying to revive the HPCC site within delltechcenter. I write a blog on the HPCC page delltechcenter.com and I'm working on adding more relevant content. I would encourage you to read the site and let me know what you think. If you would like to start threads describing your experiences that would really be helpful (you don't necessarily have to ask questions). Even if you want to start a thread about "best practices" for things such as installing ROCKS, installing OCS, installing configuring ganglia, Torque (or whatever job scheduler you use), or any other tools/tricks you have. I would love to have customers do this.

BTW - I'm what Dell calls an "Enterprise Technologist" for HPC at Dell. While I'm not fond of the phrase, I'm the subject matter expert for HPC :) So, HPC stuff tends to flow through me :)

Real quick (I've probably gotten too long) - I've written some blogs on technical subjects (looking at IO patterns, etc.). I've uploaded some code for these blogs. I'd like to continue to do this type of thing, so let me know what you think.

Thanks!

Jeff
No Events found!

Top