[NBLUG/talk] thought I'd ask here also - Anyone have any experience with LTSP?

Jordan Erickson jerickson at logicalnetworking.net
Sat Jan 24 09:25:46 PST 2015


Take a peek at this page I wrote a few years back - might still be
applicable to your situation:

https://help.ubuntu.com/community/UbuntuLTSP/Firefox3Optimize

In my experience, web browsers will always be the hogs of an LTSP server
(especially if you choose to use flash or java). If your thin clients
can handle it, consider installing browsers as a 'localapp' (you install
browser+plugins in client chroot and configure localapps in lts.conf for
each architecture on server. Browser then gets executed locally on TC
and won't hog server resources). You can also go full 'fat client' setup
(*all* apps run on thin client and LTSP server is basically used as a
fancy boot server), your thin clients can probably handle it if they're
all 3GHz w/2GB, but it's more upkeep and maintenance (which LTSP tries
to minimize because of its purpose). I've always been an LTSP purist -
everything *should* run on server, with a single config, but in many use
cases that's not possible.

Open/LibreOffice is another issue I've seen, especially when working
with heavy graphics / clipart stuffs (as local thin client video memory
is used and there is/was a bug where every time you drag it to another
portion of the document, it doesn't release the memory properly and
eventually exhausts it all). I had a bug report on this many moons ago,
not sure if it ever got fixed.

But now, X (and mir/wayland/...) development is moving quickly away from
being a network enabled display server and LTSP is probably going to be
rendered incompatible at some point. =( Maybe I'm wrong (I hope I am)
because I really love the benefits it provides.

Cheers,
Jordan


On 01/23/2015 04:54 PM, gandalf at sonic.net wrote:
> It's reportedly Gigabit. I'm rather new here so I'm learning as I'm
> going. There is a three LTSP cluster setup and a separate server with
> the /home share which is about two terabytes. There are about a
> hundred forty LTSP users but perhaps 40 of them are currently active.
> The client LTSP machines seem to be Core2 Duos 3ghz with 2gigs of
> memory although I haven't seen all of them. They are running Ubuntu 12.
>
> We experienced some system slowdown today with the load average going
> into the 30-60 range and CPU running to 50% or so if I remember right.
> The two main culprits were chrome and firefox. I did some
> experimentation afterwards while the users were at a party and I was
> able to push the system to about 3 all by myself by watching youtube
> videos or scrolling facebook. Ten bored people checking facebook or
> watching videos could easily produce the results I observed.
>
> I'm thinking about experimenting with turning the thin clients into
> fat clients where they run applications on their own CPUs and memory
> but have only a day or two experience with LTSP. I also think that
> it's rather odd to have the home directories on a separate server.
> Unless it was a dedicated network connection it would compete with the
> server/client connection for bandwidth in a game where everyone
> looses. It would make much more sense to have some sort of fast raid
> array set up directly to the LTSP server.
>
> I'm the new guy on the job here and there is a lot of stuff that was
> set up by people who no longer work here.
>
>
> On 2015-01-23 16:00, Jordan Erickson wrote:
>> Can you give us a bit more detail as to your network/server setup? You
>> say you're running 24cpus and 128gigs. Might be a logical explanation to
>> house NFS on its own server.. What's your network speed? Thin (or fat)
>> client specs? Distro?
>>
>>
>> Cheers,
>> Jordan
>>
>>
>> On 01/23/2015 03:52 PM, gandalf at sonic.net wrote:
>>> It's actually a NFS to a separate server. I'm not sure why it was done
>>> that way. To me this would probably further congest the network while
>>> not being nearly as fast as a high performance direct disk interface.
>>>
>>> On 2015-01-23 15:00, Omar Eljumaily wrote:
>>>> Not speaking from experience with LTSP server, but my guess would be
>>>> that it is a disk io bottleneck.
>>>>
>>>> http://stackoverflow.com/questions/21617500/understanding-load-average-vs-cpu-usage
>>>>
>>>>
>>>>
>>>> I believe that web browsers cache on disk most of their content and
>>>> media before rendering it.  That's a lot of users hitting a single
>>>> disk drive or array.  Do you have a large RAID array?  SSD?
>>>>
>>>> Omar
>>>>
>>>> On 1/23/2015 2:19 PM, gandalf at sonic.net wrote:
>>>>>
>>>>> I'm seeing very high load averages with low CPU usage. The two main
>>>>> culprits are chrome and firefox. I've got about a hundred users
>>>>> sharing the LTSP server which has some rich resources (24cpus and
>>>>> 128gigs).
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> talk mailing list
>>>>> talk at nblug.org
>>>>> http://nblug.org/cgi-bin/mailman/listinfo/talk
>>>>
>>>> _______________________________________________
>>>> talk mailing list
>>>> talk at nblug.org
>>>> http://nblug.org/cgi-bin/mailman/listinfo/talk
>>> _______________________________________________
>>> talk mailing list
>>> talk at nblug.org
>>> http://nblug.org/cgi-bin/mailman/listinfo/talk
>>
>> _______________________________________________
>> talk mailing list
>> talk at nblug.org
>> http://nblug.org/cgi-bin/mailman/listinfo/talk
> _______________________________________________
> talk mailing list
> talk at nblug.org
> http://nblug.org/cgi-bin/mailman/listinfo/talk




More information about the talk mailing list