Hacker News new | ask | show | jobs
by brutos 3183 days ago
Depends heavily on the projected utilization. If you know your compute node is going to be computing for the next 3 years with at least medium utilization, then the self hosted metal is probably going to be quite a bit cheaper.

Its amazing how much hardware you can pack into a single machine for 10k€. Last year our group bought two additional high-memory (768GB) nodes for around that price each (including support for a couple years from the vendor).

A few years before we bought 40 nodes with 128GB RAM each, for a similar price to last years high-memory nodes (and a fast interconnect and a lot of storage).

If you are at a larger research institution, you probably also have an IT department that can co-locate your hardware for next to nothing (compared to cloud). There you also will save a lot of ingress/egress, storage, backup, etc. costs.

Regarding the per student costs, even with cloud instances I would consider running a traditional HPC job system (grid engine, lsf, torque, ...). The MIT had a nice solution with Starcluster [1] to easily deploy a SGE on AWS. It looks a bit dead now though.

[1] http://star.mit.edu/cluster/

1 comments

> Depends heavily on the projected utilization. If you know your compute node is going to be computing for the next 3 years with at least medium utilization, then the self hosted metal is probably going to be quite a bit cheaper.

Isn't that already the case for 1 month? Bare metal doesn't mean own data centre or colocation. If you go with a hosting provider most offer dedicated hardware on a monthly contract. As long as you need them longer than 1-2 months that should be significantly cheaper than Google/AWS.

That's usually the case for almost everything on AWS/Google. If you're using them for specific features, or for very bursty work (e.g. if you use the instances less than about 6-8 hours a day), they can be cost effective, but the moment you use instances full time and don't leverage/depend on a ton of extra services, you're paying way above the odds.
Kubernetes helps this a bit with bin packing. Much easier to keep 3 32 core servers loaded than 32 4 core servers.
But that's the case whether you're using AWS or self-hosted, so it doesn't really alter that calculation much