Date: Mon, 2 Dec 1996 11:30:13 -0500 (EST) From: John Dyson <dyson@dyson.iquest.net> To: dennis@etinc.com (dennis) Cc: kpneal@pobox.com, hackers@freebsd.org, davem@caip.rutgers.edu Subject: Re: TCP/IP bandwidth bragging Message-ID: <199612021630.LAA09409@dyson.iquest.net> In-Reply-To: <3.0.32.19961202100230.00b7b890@etinc.com> from "dennis" at Dec 2, 96 10:02:36 am
next in thread | previous in thread | raw e-mail | index | archive | help
> At 01:21 AM 12/2/96 -0500, you wrote: > >At 01:15 PM 12/2/96 +1000, Stephen Hocking wrote: > >>A Linux zealot has the following in his sig - what's our current ability? > >> > >>---------------------------------------------//// > >>Yow! 11.26 MB/s remote host TCP bandwidth & //// > >>199 usec remote TCP latency over 100Mb/s //// > >>ethernet. Beat that! //// > >>-----------------------------------------////__________ o > >>David S. Miller, davem@caip.rutgers.edu /_____________/ / // /_/ >< > > > >*sigh*. > > > >Hope this doesn't get out onto the net again. > > > Yeah, like those FTP reported statistics are really accurate. I've seen > physically impossible numbers....... > I have been holding quiet until now, esp since I was one of the culprits in the above mentioned flame-war. My take on the whole thing is that there is always an attempt to show parity or superiority of one party over another (ego or money thing or whatever.) However, when comparisons are made, the context or situation associated with the comparisons should be as fully disclosed as possible (especially when the situation might not be "real world" for many of the users who are attempting to compare.) Micro-level benchmarks are not the applications that end-users normally run and should be interpreted very carefully. Frankly, many end-users can be fooled by looking at the micro-benchmark results. Application benchmarks are more accurate, especially when run in the same kind of environment that the user will encounter. Even then, you should not blindly trust those. One thing that is missing in many micro-level benchmarks is the testing of scalability. Running one tcp connection between machines is very different than running 2000. Running one process on a machine is very different than running 2000 processes (even if the system is not paging.) In fact, the impact of micro-benchmark perf is different depending on a machines application. For example, fork/exec time on a single user workstation with a forking www server that is servicing 100 hits per day isn't very important. Even if the fork/exec time is 10msecs, it isn't going to impact the performance of the system very much. (Note that FreeBSD fork/exec time is 650usecs on my machine, but in my application -- writing/compiling software, it wouldn't appear to be measurably faster than a machine that fork/exec's in 10msecs.) Other factors will be more important on a system like that. John
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199612021630.LAA09409>