30 April 2016

30th of April

Its all all about the Pi's again this week.

I have ordered three more Pi3's which will be used to replace the remaining Pi2 and the two Parallella's that are still crunching. That will bring the Pi part of the cluster up to 7 and should make maintenance a little easier with only 1 flavour of Linux to deal with.

I have joined a team called Raspberry Pi over at the Einstein project under the handle PorkyPies.


Beowulf Cluster
I used the now redundant Pi2's for medical, I mean computer, experiments. I was trying to follow a tutorial about setting up a Pi cluster. A true Beowulf cluster. Not what I am running at the moment which is called a CoW (Cluster of Workstations). In a Beowulf cluster you use off the shelf hardware and have a head or master node and it issues tasks to the compute nodes to run and typically uses MPI (Message Passing Interface) to communicate.

The tutorial I was using was this one: Build-a-compact-4-node-raspberry-pi-cluster

There are other tutorials but most seem dated and were for other flavours of Linux so I felt this one was closer to what I wanted to try. There are a few things in there that I didn't want to do like using a USB hub to power it (I have plenty of power adapters) so I skipped that bit and they also added a Blinkstick which I wasn't going to do either. I didn't need a fancy case either, just two Pi2's should be enough to try it out.

Following the tutorial using the March 2016 release of Jessie lite was a failure. It seems the rpcbind service fails to start up. Having manually worked around that I got the ssh keys setup and mpich installed and while I could run a task on the head node I couldn't get it to use the compute node.

I had a bit more success using Stretch which at least seems to have fixed rpcbind not starting up, but still can't get it to run tasks on the compute node. At this point I am stuck.

1 comment:

Mark G James said...

Update 2nd of May
I managed to get the cluster to work, so now to see if its possible to have some of the apps run using MPI.