I've seen the hypervisor steal back ~70% CPU time from m1.large for many seconds at a time, according to top.
If using EC2 for Hadoop+HBase, c1.xlarge is the minimum requirement in my experience. I've been testing HBase on EC2 for over a year. Best regards, - Andy Problems worthy of attack prove their worth by hitting back. - Piet Hein (via Tom White) ----- Original Message ----- > From: Roman Shaposhnik <[email protected]> > To: [email protected]; Andrew Purtell <[email protected]> > Cc: > Sent: Thursday, November 10, 2011 7:33 PM > Subject: Re: HBase 0.92/Hadoop 0.22 test results > > On Thu, Nov 10, 2011 at 7:22 AM, Andrew Purtell <[email protected]> > wrote: >>> That's a small cluster running on EC2. >> >> What instance type? > > m1.large > >> Should use c1.xlarge or m4.4xlarge, they won't see the possibility of > noisy neighbors. > > I know, but these are expensive. I'm lucky enough Cloudera is being > gracious footing > the bill for m1.large instances and not requiring me to run spots. > >> Another thing that can happen, is if you use spot instances your spot > instances can >> be taken back by AWS at any time. We had clusters in us-west-1 last week > that were >> abruptly terminated without notice like this. (We use on-demand master and > spot slaves, >> only the masters remained running... several times last week...) > > That's not a problem. At least for now it isn't. > > Thanks, > Roman. > > P.S. I was talking to EC2 AWS folks trying to see whether they would > be in a position > to donate credits for Apache Projects, but these talks are not progressing > well. >
