A short study about Shibboleth performances.

[Etienne Dysli-Metref]

On 10/06/16 18:09, David Verdin wrote:
>> What does the response times distribution look like?
> According to which parameter? Relatively to the users count, it is on
> the graphs.

Yes, relative to the number of logins per second. I see now you have the
average there in orange (that graph is rather crowded). Did you also
record the quartiles? For my tests, I decided that 95% of requests
should be served in less than 5s.

>> In my own load tests, I couldn't get more than 25 logins/second without
>> response times exceeding the 5s threshold (unsolicited SSO, 1 CPU,
>> PostgreSQL, external LDAP for authn & attributes)
> And did you, by any chance use one of the optimizations we found?

No, I did not try to optimise anything. I mainly wanted to benchmark a
"small VM" on our platform to see what we could get out of it.

   Etienne

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP digital signature
URL: <http://shibboleth.net/pipermail/users/attachments/20160613/1089bb82/attachment.sig>