Re: wlus and timings
On Fri, Sep 24, 2004 at 10:16:45AM +0200, Patrick Willam wrote:
> Finn-Arne Johansen wrote:
> >Alot of our users have reported that it takes "forever" to add users to
> >the ldap using wlus and file import. Well, I guess there is good news
> >for the future.
> And IMHO at least there should be a progressing feedback in/for the
> But as far as I've looked into und understood the source code of WLUS
> this cannot be achieved with some simple change(s) in the fileimport
> function(s). :-(
Yes, but big file-imports need some other means. If you are running
schools with thousands of users, you should really make other methods
for adding users. And you should contribute the method back to the
community. If you dont contribute back, you will have the maintenance
> >Using the sarge-based debian-edu image i build at random times, I've
> >tested some adding:
> > ....
> >- Then I tried to add 1000 users, which took about 40 minutes.
> Ok this is faster, but IMHO still to slow.
> I think the fileimport has to be rewritten with optimisation for speed.
> >but what
> > I forgot, was that there is a sizelimit in the response from
> > openldap, which by default is 500, but that we have set to 1000. i
> > guess raising that value to 200 would have speeded up the process.
> > The first 500 of the last thousand was significant faster than the
> > last 500.
> your proposal = 200 < default = 500 < your setting = 1000
> < unchanged default config of my vm-tjener = 2000
> "raising" from 500 to 200 ??
> So what is "higher"? And is higher=better?
Sorry, i ment 2000
is the default in skolelinux now 2000 ? I dont think so.
okay, just checked, and found it's recently raised to 2000
> >Ralf: how many users did you have in advance? Have you tuned ldap to
> >give you more users when you query?
> On behalf of Ralf...
> users in advance: ~850
> ldap: not tuned, just defaults from installation (sizelimit=2000)
Then this should be enough then.
> >if not, change the sizelimit, so
> >that it might give you all the users when you query.
> we are talking about the "sizelimit" in slapd.conf, right?
> >Also there is some other parameters that you could tune in openldap,
> >and also, remember to edit nscd.conf on the clients.
> Where is the relation between "nscd.conf on the clients" and openldap &
> webmin & wlus on tjener?
If you have the numbers in nscd.conf to low, the cace wont hold all
your users, and you end up querying the ldap and stress that component.
I have not been able to give openldap enough thread/memory/resource to
server all the users at Ulsrud without using nscd.
> >Another thing that we have noticed at Ulsrud, is that the NFS mounting
> >some times has problems. We have hard-coded changed the use of
> >netgroups in /etc/exports into dns-names, and since then we have not
> >had any problems, I think. basically what we did was change the line
> >that looked like this:
> > /skole/tjener/home0 @ltsp-server-hosts(rw) @workstation-hosts(rw)
> > @server-hosts(rw)
> > /skole/tjener/home0 ltspserver*(rw) @workstation-hosts(rw)
> > @server-hosts(rw)
> Do only the ltspservers have NFS-problems or why not change all to DNS?
Well, we are in a debug modus. I think maybe the problem is that we
have turned of famd on the ltsp-server, and this also causes some extra
traffic. but the famd efficiently hang our servers when it was there.