Date: Wed, 21 Dec 2011 13:28:13 +0000 From: Tom Evans <tevans.uk@googlemail.com> To: Randy Schultz <schulra@earlham.edu> Cc: freebsd-performance@freebsd.org, Matthew Tippett <matthew@phoronix.com> Subject: Re: Benchmark (Phoronix): FreeBSD 9.0-RC2 vs. Oracle Linux 6.1 Server Message-ID: <CAFHbX1J-7AQpC=qGD5K9wpcoz0qK=RksWyyP34H-U2%2BpLosJsw@mail.gmail.com> In-Reply-To: <alpine.BSF.2.00.1112210801590.49042@tdream.lly.earlham.edu> References: <4EE1EAFE.3070408@m5p.com> <CAJ-FndDniGH8QoT=kUxOQ%2BzdVhWF0Z0NKLU0PGS-Gt=BK6noWw@mail.gmail.com> <4EE2AE64.9060802@m5p.com> <4EE88343.2050302@m5p.com> <CAFHbX1%2B5PttyZuNnYot8emTn_AWkABdJCvnpo5rcRxVXj0ypJA@mail.gmail.com> <4EE933C6.4020209@zedat.fu-berlin.de> <CAPjTQNEJDE17TLH-mDrG_-_Qa9R5N3mSeXSYYWtqz_DFidzYQw@mail.gmail.com> <20111215024249.GA13557@icarus.home.lan> <4EE9A2A0.80607@zedat.fu-berlin.de> <CADWvR2hpooV4Sdkzor6oHpXWkLCz0z0UxBB_-nnrnYLeUY5oow@mail.gmail.com> <CANY-Wm8DmFxm7xwXfRcyBSu4z6fTOap%2By=Mhx%2BuBP=uJN4g3fw@mail.gmail.com> <4EF1121F.9010209@zedat.fu-berlin.de> <4EF11B57.7090007@phoronix.com> <alpine.BSF.2.00.1112210801590.49042@tdream.lly.earlham.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Dec 21, 2011 at 1:16 PM, Randy Schultz <schulra@earlham.edu> wrote: > On Tue, 20 Dec 2011, Matthew Tippett spaketh thusly: > > -}There are still possible issues with those benchmarks. =C2=A0The Xeon h= as known > -}problems scaling from 6 to 12 cores (well enabling the hyperthreading),= so you > -}may find that some platforms are penalized in performance if HT is turn= ed on. > -}See the scaling that Phoronix has done in > -} > -}http://openbenchmarking.org/result/1112166-AR-1112153AR03 > -} > -}Most systems are good with scaling on real cores, the hyperthreading (a= nd for > -}that matter the Bulldozer thread affinity) can really break performance= . > -}Different platforms have different behaviours. =C2=A0Benchmarking is a = mucky > -}business.. > > This brings up a good point. =C2=A0While I don't have any hard #'s, I sus= pect the > vast majority of SA's do not have/spend much time tweaking this and tunin= g that. > Order the box, drop the OS on it, install needed bits and go. =C2=A0Sayin= g "oh for > app X you need to tune these sysctl's", while it may be entirely true, ki= nda > throws things out the window. =C2=A0It seems that once one starts down th= at slippery > slope, it merely becomes a game of how much time to you have to "tune 1 m= ore > thing". =C2=A0;> =C2=A0I think Phoronix has the right idea of just grabbi= ng a stock box > and not looking into what needs to be tweaked for a specific app. > I think that a good SA will at least consider how drives are arranged. We don't just slap ZFS on a single disk and expect magic to happen, we consider how write heavy a system will be and consider a dedicated ZIL, we consider what proportion of files will be re-read and how much application memory will be required and adjust ARC and L2ARC accordingly. Tuning and foresight are important. Cheers Tom
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAFHbX1J-7AQpC=qGD5K9wpcoz0qK=RksWyyP34H-U2%2BpLosJsw>