« Disks Ain't Dead Yet: GraphChi - a disk-based large-scale graph computation | Main | Cinchcast Architecture - Producing 1,500 Hours of Audio Every Day »

Strategy: Kill Off Multi-tenant Instances with High CPU Stolen Time

Are all instances created equal? Perhaps because under multi-tenancy multiple virtual machines run on the same physical host, not all applications will run equally well on every instance. In that case it makes sense to measure and move to a better performing instance. 

That's the interesting idea from @botchagalupe:

Imagine something like a "performance monkey" where an infrastructure is so bound that it can kill lower performing instances automatically.

@adrianco says Netflix has throught of doing the same: 

We've looked at killing off multi-tenant instances that have high CPU stolen time...

Related Articles

Reader Comments (2)

We do this as well. We ensure to take 2% (configurable now, self determining in the future) peaks out in determination and also make sure long standing transactions are left in a recoverable state. Rolling back midway of a partially complete transaction which is part of a long transaction helps as long as other instances are available to take over.

July 19, 2012 | Unregistered CommenterDickey Singh

You missed out one of my follow-up tweets. We don't actually see enough of a problem with stolen time for this to be worth worrying about.

July 20, 2012 | Unregistered CommenterAdrian Cockcroft

PostPost a New Comment

Enter your information below to add a new comment.
Author Email (optional):
Author URL (optional):
Some HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>