Well since you are using S3 it's not like adding more nodes adds more spindles, you only really distribute the HBase-side processing. Also, a count is a sequential operation so it's not distributed (you only hit one node at a time). To speed up scans you can increase scanner caching.
J-D On Tue, Oct 20, 2009 at 9:23 AM, Ananth T. Sarathy <[email protected]> wrote: > So, I have added 3 region servers, and I am not getting any real performance > increases.. Does anyone have any other tips? I checked the wiki, but we are > doing what is applicable to us. Is there any way check latency on s3, and > are the any jobs that could be run to increase performance for things like > count? (we only have like 27k rows right now, and it's terribly slow, which > is weird, because it was fast for a while, and suddenly slowed down ) > Ananth T Sarathy >
