Shibboleth Developers

[Scott Cantor <>]

One of these days I'll get it into my head that any hardware at OSU can be 
assumed to be 50% slower than whatever somebody would run
to track their recipe database with.

I'm Sparc-challenged, so I didn't really have any good perspective until I 
got to do some Intel testing. My numbers on a 2x666 P3
show HS response time of under 2 seconds for 20 users. 50 users still stays 
under 6-7 seconds (which sucks, but this is not a
high-end box and that's a lot of logins).

After learning how to find out what the shib machines are (cat /proc/cpuinfo) 
I noticed shibprod0 is a 1x1.2Ghz or so. It turns out
that that doesn't do all that much better, so clearly having the 2 or more 
engines helps a lot, as you would expect.

A really high end Intel server would scream, and clearly a couple of 2 ways 
load balanced would handle a sizable amount of login
traffic.

I have a new motherboard and CPU coming at home (2.5Ghz), so I'm inclined to 
benchmark again on that, under Windows, and see what
that looks like.

-- Scott

------------------------------------------------------mace-shib-design-+
For list utilities, archives, subscribe, unsubscribe, etc. please visit the
ListProc web interface at 

    http://archives.internet2.edu/

------------------------------------------------------mace-shib-design--