directory-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Enrique Rodriguez <erodrig...@apache.org>
Subject Re: Proposing some changes to the ProtocolProvider API
Date Fri, 10 Dec 2004 15:07:38 GMT
Berin Loritsch wrote:
> Alex Karasulu wrote:
> 
> <snip>

There has been some talk of SEDA adding overhead for "under 50 users." 
Chances are, this overhead will be so far below acceptable response time 
thresholds, that no one will notice.  So, there is no problem that 
requires us to use different approaches for different numbers of users. 
  The caveat here is that this applies to the type of protocols we plan 
to add, namely LDAP and Kerberos/changepw.

Put another way, if SEDA increases overhead for 1 user from 3ms to 30ms, 
on the one hand that's 10X slower ... on the other hand, that's 27ms 
slower.  There is no problem to be solved here.  The problem is, and 
always has been, exclusive to large numbers of users.

I believe typical network operations against a directory server are the 
result of human interaction, in which case we can use human interface 
guidelines in our performance testing.  As an example, the "maximum 
acceptable response time" according to the Gnome HIG is 100ms and for 
the web a couple seconds.  As long as we keep below acceptable response 
times, we will be perceived as having a fast server.

Gnome HIG:
http://developer.gnome.org/projects/gup/hig/draft_hig_new/feedback-response-times.html

Insights from Human Factors International:
http://www.humanfactors.com/downloads/apr01.asp

-enrique


Mime
View raw message