httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From r..@ai.mit.edu (Robert S. Thau)
Subject Re: Nameserver cache
Date Wed, 22 Mar 1995 19:11:46 GMT
   Date: Wed, 22 Mar 1995 15:24:29 -0800 (PST)
   From: Simon Spero <ses@tipper.oit.unc.edu>

   mini data-point; using a 100 entry wide cache backed by a non-local nis 
   server (with dns caching enabled, and the cache warm), using the memory 
   cache was around 4 times as fast as always hitting the server. 

This is in line with my intuition at the time I wrote the code
originally, that grubbing through shared memory would be faster than
IPC even if the process on the other end had the remote hostname
immediately to hand.  That does seem to be the case with the standard
SunOS4 libraries, but I have to admit I didn't try -lresolv.

In any case, my intuitions are far less clear when it comes to
preestablished pipes, vs. local resolver (unless, of course, you're
using the whole scheme just to get NIS+ out of the picture --- it
seems that it's *always* a good idea to get NIS+ out of the picture
;-).  It could be that having httpd manage its own cache just isn't
worthwhile without shared memory.  (I believe this would still cover
the NIS+ afflicted).

In any case, whoever said that what we really need here is a bake-off,
they're right.

rst


Mime
View raw message