Date: Thu, 5 Dec 1996 13:48:41 -0500 (EST) From: Mark Mayo <mark@quickweb.com> To: Michael Smith <msmith@atrad.adelaide.edu.au> Cc: hackers@freebsd.org Subject: Re: New benchmarks to design / lmbench results Message-ID: <Pine.BSF.3.94.961205133119.7282A-100000@vinyl.quickweb.com> In-Reply-To: <199612050218.MAA19340@genesis.atrad.adelaide.edu.au>
index | next in thread | previous in thread | raw e-mail
On Thu, 5 Dec 1996, Michael Smith wrote: > Mark Mayo stands accused of saying: > > > > > > Do you have '-pipe' in CFLAGS in /etc/make.conf? Enough memory? SCSI? > > > A more reasonable load average for 'make world' is 1.2 or so, and normally > > > less than 10% idle in my experience. > > > > Always wanted to know what -pipe did; neato! No I didn't have it in > > there.. I don't really want to use the suggested -O2 -m486 however, so I > > think I'll add -pipe. I don't have the time to do another make world to > > compare however.. I've got 32 MB or RAM - is this considered enough? > > If you're not doing anything else on the machine, then 32M is fine. > Run 'systat -vmstat' while you're running the build and keep an eye > on whether you're paging. > > > Any idea how much faster -pipe generally speeds up the compilation? I'm > > assuming I might need more RAM for the PIPES to talk though.. I'll be up > > to 48MB soon, then I'll be happy :-) > > I wouldn't want to make a guess, but with your hardware I'd expect it to > make a moderately significant improvement. (10-20%?) Well, I compared a kernel build with and without -pipe. It took 4 minutes, 35 seconds without, and dropped to 3 minutes, 27 seconds with the -pipe CFLAG!! Wow! That's a significant improvement. Now I can't wait to do a make world with -pipe! I did notice, however, that adding CFLAGS= -pipe to /etc/make.conf didn't automagically add a -pipe flag to the kernel build procedure (I had to add the -pipe manually to the kernel Makefile) Also, I ran lmbench out of curiosity! Wow, does FreeBSD rock on Pipe bandwidth!! I get 147.20 MB/s pipe bandwidth. Also, the context switches were blazing (166667 - whatever that number is, or 6 usecs) on 2 and 8 "small processes". Pipe transactions were at 33 usec. The fork/exec times were also very impressive, at 1203 usec (831/s). Basically, everything involving TCP/UDP/socket/Pipe had amazing results - the highest of the EXAMPLE group by factors. I'll have to do searching and see what newer machines are doing. The results are quite impressive, and fun to look at :-) My machines seems to have very high memory read times (according to lmbench anyways) with the memory sum and read bandwidth hovering around 160 MB/s and mmap reads at around 108 MB/s. The writes weere quite slow (~50MB/s) and bcopy results slow as well. It was the first time I've ever ran lmbench, and it was quite fun. Maybe I'll go and get 1.1 and tweak it to pieces and see what numbers I can squeeze out to poke fun at my Linux friends ;-) cya, -mark > > > -Mark > > -- > ]] Mike Smith, Software Engineer msmith@gsoft.com.au [[ > ]] Genesis Software genesis@gsoft.com.au [[ > ]] High-speed data acquisition and (GSM mobile) 0411-222-496 [[ > ]] realtime instrument control. (ph) +61-8-8267-3493 [[ > ]] Unix hardware collector. "Where are your PEZ?" The Tick [[ > --------------------------------------------------- | Mark Mayo mark@quickweb.com | | RingZero Comp. vinyl.quickweb.com/mark | --------------------------------------------------- "To iterate is human, to recurse divine." - L. Peter Deutschhelp
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.94.961205133119.7282A-100000>