Skip to content

Comment on Full-scale file system acceleration on GPU [pdf]parent

Comments

In systems performance I would advise to never think of any workload as unidimensional (ie: Any file system optimization can either improve IO latency or be useless)

Issuing individual truncates of 1B files can be just as much of a CPU problem then an IO one for example.

But why wouldn't using one of many CPU cores be sufficient?

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.