Random reads are 5x faster in 0.96.0! Or not

[Jean-Daniel Cryans <jd...@gmail.com>]

Hey devs,

I was testing performance of 0.96.0RC3 vs 0.94.12 and found that random
reads are 5x faster out of the box in the former, party! But wait, no,
HBASE-7008 disabled Nagle's algo and it was only applied to 0.95, so if you
configure that in 0.94 it becomes even faster than 0.96 by a small margin.

Just a reminder when doing your own testing,

J-D

Re: Random reads are 5x faster in 0.96.0! Or not

[Demai Ni <ni...@gmail.com>]

hi, J-D,

It is a very exciting improvement. If we'd like to make 0.94 user benefit
from this, will backport HBASE-7008 be good enough, code wise? Certainly,
more testing is needed to guard it. Thanks.

Demai


On Mon, Sep 23, 2013 at 5:10 PM, Jean-Daniel Cryans <jd...@gmail.com>wrote:

> Hey devs,
>
> I was testing performance of 0.96.0RC3 vs 0.94.12 and found that random
> reads are 5x faster out of the box in the former, party! But wait, no,
> HBASE-7008 disabled Nagle's algo and it was only applied to 0.95, so if you
> configure that in 0.94 it becomes even faster than 0.96 by a small margin.
>
> Just a reminder when doing your own testing,
>
> J-D
>