On Mon, Jun 16, 2008 at 2:50 PM, Nicolas Williams <[EMAIL PROTECTED]> wrote: > It doesn't get substantially worse unless the customer adds many entries > to /etc/services -- but at ~11,000 entries there's not many more that > they could add, so the time to merge is really bounded at some small > multiple of 1.5 minutes.
Oh, but now there are plenty to conflict with. Whenever I've needed a port for local use I've always grabbed it and not registered it with IANA. There's a pretty good chance if I've been using it for several years, the IANA registered use is irrelevant to me. If any upgrade path tacks my entries onto the end of the file, things like "netstat -a" will give different output. There's some potential that this may throw off a variety of homegrown tools that rely upon such output. Do we have an idea of how often /etc/services is read under various workloads? If it is really frequently, can you confirm that nscd caches this to minimize the performance impact of linear file searches? (I've not kept up on the impact of post S10 nscd changes.) Now I am done advocating for the devil. If IANA has set aside a range for private use, having that clearly documented in the services(4) man page and/or comments in /etc/services file would be a good idea. -- Mike Gerdts http://mgerdts.blogspot.com/ _______________________________________________ networking-discuss mailing list [email protected]
