Date: Wed, 8 Dec 2010 13:05:11 -0500 From: Alejandro Imass <ait@p2ee.org> To: Mark Terribile <materribile@yahoo.com> Cc: FreeBSD Questions <freebsd-questions@freebsd.org> Subject: Re: Page faults and threads Message-ID: <AANLkTi=VTiLnj7QTHeTGYWz7H=0AE7AJa4NV_C2VgPq-@mail.gmail.com> In-Reply-To: <592756.11515.qm@web110301.mail.gq1.yahoo.com> References: <AANLkTi=E%2BCAA6%2BCTNKYLqbDvsxVG6rmR4Q9nE_TGcHau@mail.gmail.com> <592756.11515.qm@web110301.mail.gq1.yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Dec 8, 2010 at 12:27 PM, Mark Terribile <materribile@yahoo.com> wro= te: > [...] Copying the list for this comment... Well I can tell you this much. A while back I carried out some heavy load testing on a Catalyst app on mod_perl, mod_worker apache 2 on FBSD 6.2 and Debian with kernel 2.6.18 both on 64 bits same exact hardware: FreeBSD 6.2 Stable Apache 2.2.3, Server MPM: Worker mod_perl 2.0.4 Perl v5.8.8 built for amd64-freebsd-thread-multi Catalyst 5.7014 mysql Ver 14.12 Distrib 5.0.45, for portbld-freebsd6.2 (amd64) using 5.2 Debian 4.0 (Etch) Stable Apache 2.2.3, Server MPM: Worker mod_perl 2.0.4 Perl v5.8.8 built for amd64-debian-thread-multi Catalyst 5.7014 mysql Ver 14.12 Distrib 5.0.45, for Debian (amd64) using 5.2 I can't disclose the official results, but the bottom line was that Linux was generally somewhat faster (61 vs. 72 ms per request on average) but it's over-optimistic memory broke the OS completely when maxing out and consuming the swap, and we concluded that the VM in that version of Linux was particularly broke for our application. Coincidentally you could observe a dramatic difference in the Virtual and Resident size of each Apache process when in peak load: Linux 2.6.18 : 710/130 FreeBSD 6.2: 185/132 (it also proved in both cases that threaded mod_perl (an esoteric mix in the eyes of many an expert) saved tons of RAM and was generally very stable ;-)) Anyway what really caught our attention in these tests is that when the extreme load was removed, FBSD recovered almost immediately like nothing had happened and we had to reboot Linux almost every time, cause it remained very unstable afterwards. We did manage to play with Linux's vm settings especially the over-committal stuff to somewhat behave like FBSD, but vanilla FBSD was always more stable in every test. This told us that the FBSD folks must be doing something right in the standard thread / vm management in FBSD and we switched to it since for our heavy-weight Web deployments. > > I'd like to know if things how things are now on FreeBSD. > Hopefully this thread will catch the eye of some FBSD threading experts ! > =A0 =A0Mark Terribile > > --- On Wed, 12/8/10, Alejandro Imass <ait@p2ee.org> wrote: > >> From: Alejandro Imass <ait@p2ee.org> >> Subject: Re: Page faults and threads >> To: "Mark Terribile" <materribile@yahoo.com>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTi=VTiLnj7QTHeTGYWz7H=0AE7AJa4NV_C2VgPq->