Mascot: The trusted reference standard for protein identification by mass spectrometry for 25 years

Posted by John Cottrell (September 18, 2018)

Mascot Server cluster mode

Most modern Intel processors have at least 4 cores and some models have 12 cores or more. Mascot Server is licenced by the CPU, where each CPU corresponds to 4 physical cores, so a single PC is perfectly sufficient for licences of 1 or 2 CPU.

If you have a larger licence, there comes a point where it is not practical to cram all the processing into a single PC, because processors with very large numbers of cores can be expensive and may require exotic system board hardware to avoid bottlenecks with memory and storage. The most efficient and cost-effective solution becomes a cluster of ‘commodity’ PC’s, each of which has a relatively modest specification. Cluster mode is a standard feature of Mascot Server; you don’t need a special version. In fact, you may have noticed it offered as an option during installation.

For licences of 3 or 4 CPU, you could choose a single machine, possibly with dual processors, to keep things simple, although you might achieve the same performance at lower total cost with a couple of single processor PCs. For licences of 5 CPU or more, a cluster is usually the most practical option. Note that all machines in a cluster should have processors of the same speed. Otherwise, the PC with the slower processor(s) will become a bottleneck.

These numbers are not hard and fast, of course. If you already have hardware sitting idle, first preference will be to make use of it. Or, there may be a particularly good deal on a system which makes it attractive to configure a 6 cpu licence on a single machine. Besides hardware cost, the other factor is speed. Since a Mascot licence is for a fixed number of cores, it is the single thread benchmark that matters. For example, an Intel Xeon E5-2660 v4 has 14 cores while an Intel Core i7-8700 has only 6, but the single thread benchmark for the Xeon is 1509 and for the Core i7 is 2631. Whatever the price to performance ratio for the hardware might be, the price to performance ratio for the Mascot licence favours the Core i7.

Only the processors used for searching require a Mascot licence. For a large cluster, it may be best not to run searches on the master (head) node of the cluster, but leave it free to run the web server, handle database updates, and generate reports. This ensures the server will be responsive even when there are several searches running, using all the processor time on the search nodes. Having a non-searching master node also gives you a spare node, in case one of the search nodes has a hardware failure.

The main down-side to cluster mode is that there is more to configure and more to go wrong. The master node in the cluster needs to be able to communicate freely with the other nodes, including copying files and starting and stopping processes – the very things that default security settings try to prevent. This means that someone who tries to configure Mascot Server on a cluster without referring to the manual is less likely to succeed than if they were ‘muddling through’ the installation for a single PC.

Currently, virtualisation is strongly in vogue. If your Mascot Server is specified and configured by your IT group, make sure they understand that Mascot is ‘processor bound’. We have encountered cases where a cluster turned out to be multiple virtual machines, all running on the same host PC and sharing the same set of physical cores. The users wondered why the performance was so poor!

Keywords: , , ,

Leave a Reply

Your email address will not be published. Required fields are marked *


HTML tags are not allowed.