Subject: Re: rpc xid randomness
To: None <fvdl@netbsd.org>
From: Jun-ichiro itojun Hagino <itojun@itojun.org>
List: tech-userlevel
Date: 09/09/2003 07:15:18
> On Mon, Sep 08, 2003 at 07:50:58PM +0900, Jun-ichiro itojun Hagino wrote:
> > to summarize,
> > - the currently-committed code is not good. it is not resistant to
> > number reuse/duplication.
> > - sequential number with time.tv_sec initialization is resistant to
> > number reuse/duplication, if we don't set date(1).
> > - niels' generator is resistant to number reuse/duplcation, and probably
> > there's no chance for duplication on reboot (due to the use of random
> > number as initialization)
>
> I just want to see some kind of benchmark, like lat_rpc from lmbench.
> Or maybe just spray(8).
i ran lat_rpc with patch. i don't have a system without patch
(i'm on a research retreat to rural area), so if there's someone with
similar machine (machine: IBM thinkpad X31, 2672-JHJ) without patch
can take measurement i would be greatful.
itojun
% work.i386/lmbench-2alpha11/bin/i386-netbsd1.6Z/lat_rpc -s
% work.i386/lmbench-2alpha11/bin/i386-netbsd1.6Z/lat_rpc localhost
RPC/tcp latency using localhost: 27.1248 microseconds
RPC/udp latency using localhost: 29.1774 microseconds