Okay.

We had a lot of changes in September and in the next slides you will get a quick wrap

up.

So, first of all, let's start with Maggi, which is still our main tier 3 parallel computer.

It has been upgraded from CentOS 7S operating system to Alma Linux 8, which again is a Red

Head clone, but yeah.

Storage system and queue configuration is the same.

Maggi is open again since Sunday, but FastTemp is not available again yet because of hardware

issues which we reported somewhere in August.

We tried to get a replacement part, but nobody, even the manufacturer itself, cannot provide

one because disk arrays are end of life since several years.

So we still hope to get FastTemp back online, but there's no guarantee and there's no estimated

time frame for getting news because our workshop has to try to fix the broken part itself.

That may work, but it also may fail.

Recompilation of applications is very likely as old libraries and so on are no longer available

and in principle, you could have started recompilations already mid of August.

Configuration is still fine-tuned.

If you have any requests, open a ticket.

For those who use legacy job monitoring for jobs on Maggi, that's gone.

Maggi is now integrated in Cluster Cockpit.

And the available software is mostly in line with Fritz and the new generation of Woody.

And Maggi is still for parallel workload, so either single or multi-node jobs.

Next big change is Woody, which remains a throughput resource.

The hardware has changed.

Woody now only has nodes with 8 gigabytes per core and 7.75 gigabytes are available for

jobs.

We now have two different node types in Woody.

SYN nodes with fast quad-core desktop-like CPUs.

You know them already from the past.

There's W1, 2xx, 3xx, 4xx, and 5xx nodes.

And there are new big nodes with current server CPUs, 2x16 cores per node.

And about 60% of these new nodes have been financed by ACUP.

Keep in mind that these new big nodes, those with node names starting with W2 and three

further digits and also the new login nodes, support AVX512 instructions, while the SYN

nodes don't.

Currently, the new login nodes are still called woody-ng.nhr.fau.de.

That will change to woody.nhr.fau.de after September 18th.

The new login nodes of Woody cannot be used to submit jobs to tiny GPU or tiny FAT.

And if software is compiled with AVX512 instructions, the binary may not run on the SYN nodes.

On the other hand, if you do not compile for AVX512 and run on the big nodes, you may lose

a factor of 1.7 in performance.

So it may be required to compile two versions of your binaries and select at runtime depending

on the node you are.

Otherwise, it may not run or you may lose significant performance.

The software environment of the new generation of Woody also changed significantly.

Instead of Ubuntu, we are now also running Alma Linux 8.

The batch system is SLURN and recompilation of applications is almost always required.

Like on Magi, configuration is still fine-tuned.

Open a ticket if needed.

Next data of jobs on the next generation of Woody are also available in Cluster Cockpit