Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 23 Apr 1999 11:08:10 -0700 (PDT)
From:      Matthew Dillon <dillon@apollo.backplane.com>
To:        Andrew Heybey <ath@niksun.com>
Cc:        Amancio Hasty <hasty@rah.star-gate.com>, Alex Zepeda <garbanzo@hooked.net>, Chris Piazza <cpiazza@home.net>, freebsd-hackers@FreeBSD.ORG
Subject:   Re: SMP nerd toy report 
Message-ID:  <199904231808.LAA23991@apollo.backplane.com>
References:   <199904231658.MAA13662@stiegl.niksun.com>

next in thread | previous in thread | raw e-mail | index | archive | help

:Well, I tried a couple of "make buildworld" on a single-processor P-II 
:450, 256MB ram, /usr/src and /usr/obj on separate 10000RPM IBM scsi
:disks, softupdates.  This is a build with an empty /usr/obj.  No
:changes to /etc/make.conf except CFLAGS="-O -pipe".  The OS
:is 3.1-STABLE as of 03/21/1999.
:
:Took about 1:15 (tried one with -j4 and one with -j8--no difference).
:This is about the same as my dual celeron 300A (at 300MHz/66MHz FSB)
:with a *single* IDE drive and 64MB of memory.  I expected it to be
:faster presuming that the build is IO or memory bound (faster disks
:(or at least two spindles) and faster memory).  This system is
:building 3.1-RELEASE if that makes a difference.
:
:It is noticably slower than my dual celery at 450 MHz/100MHz FSB
:(about 0:55), which seems to say that CPU does have something to do
:with it.
:
:So what conclusions can I draw?  Softupdates schedules the disk so
:well as to negate the effect of an extra spindle?  buildworld is at
:least partially CPU bound?  3.1-RELEASE takes less to build than
:3.1-STABLE (as of 03/21/1999)?  (I didn't think there were any major
:new chunks added but I could be wrong.)
:
:andrew

    Also, buildworld doesn't parallelize as well as it could.  There are
    so many small programs that one processor is often idle while the
    otherone is doing a link.  I only get 150% cpu utilization even with the
    entire contents of the disk cached.

    The 1:15 makes sense verses 50 minutes for my duel-PIII/450.  150% of
    50 is 50+25 = 1:15.

					-Matt
					Matthew Dillon 
					<dillon@backplane.com>


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199904231808.LAA23991>