Re: [patch] Re: [PATCH RESEND] sched: prefer an idle cpu vs an idle sibling for BALANCE_WAKE

[Mike Galbraith <umgwanakikbuti@gmail.com>]

On Tue, 2015-07-07 at 09:40 -0400, Josef Bacik wrote:

> The WAKE_WIDE_IDLE run was basically the same so I'm good with the KISS 
> version.  I'll run that through the load tests this morning and let you 
> know how it goes.  I'm still seeing a slight regression at lower RPS, 
> but it's like 1-2%, compared to ~15%.

If I'm to believe pgbench, and configs really are as close as they
should be, there may be something between 3.12 and master which could
account for your delta and then some.  I may go hunting.. heck, if I
trusted pgbench I would be hunting.

patched SLE (resembles 3.12 if you squint)
postgres@nessler:~> pgbench.sh
clients 12      tps = 134795.901777
clients 24      tps = 156955.584523
clients 36      tps = 161450.044316
clients 48      tps = 171181.069044
clients 60      tps = 157896.858179
clients 72      tps = 155816.051709
clients 84      tps = 152456.003468
clients 96      tps = 121129.848008

patched master
postgres@nessler:~> pgbench.sh
clients 12      tps = 120793.023637
clients 24      tps = 144668.961468
clients 36      tps = 156705.239251
clients 48      tps = 152004.886893
clients 60      tps = 138582.113864
clients 72      tps = 136286.891104
clients 84      tps = 137420.986043
clients 96      tps = 135199.060242