In July/August, Rosemary travelled to India and Taiwan alongside EMC and ARM for an EMC EDA user group event. First up on the itinerary: Bengaluru.
Rosemary and the other speakers all had to brave the traffic to get to the event – Bengaluru is famous for bad traffic. The ARM engineers turned up on time as they just had to cross the road.
Andy Atkins, storage architect, dazzled the audience with his account of how he turned around the storage strategy at ARM and completely rearchitected their EDA cluster configuration around EMC Isilon storage for increased reliability and performance. He also spoke about the ARM ten commandments of storage use that form part of the training all engineers have before they use the cluster.
Although Andy has made a lot of improvements, the scale of compute at ARM and the performance they require mean that they are constantly pushing their storage and compute harder than anyone else. There are still some challenges that arise as part of the evolving and growing infrastructure.
This is where Ellexus comes in. Rosemary was speaking alongside Andy about Mistral, a tool suite that Ellexus developed with the IT department at ARM to address some of the remaining storage problems. Mistral primarily solves the noisy neighbour problem: in a compute cluster with shared storage it is possible for a small number of jobs to overload the network or file system. This can affect the performance of all the jobs on the cluster and even bring it down completely.
A lot of the issues seen by anyone using shared storage are manifestations of this issue. By integrating Mistral into the existing monitoring and load balancing infrastructure that ARM already has in place, Andy is looking to address some of these. Information as to what is going on in the infrastructure is key – it’s worth deploying Mistral for that alone, but when Mistral starts to step in automatically avoid a disaster by locating and preventing a bad I/O pattern, that is where the software will demonstrate its value again.
After a lot of technical chat we all needed to cool off and what better way to do it than with a coke from the Isilon storage fridge! Modelled beautifully by Andy. We can’t wait until this fridge has a part number and we can all get one. It’s what every data centre needs.