[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [K12OSN] Re: MOSIX



Everyone,

>From my testing 8 nodes running seti home went pretty good.  100Mbit switch should be fine for smaller installations.  A switch with a 
gigabit link to the server should keep most people happy for some time.  The idea of course is to make the client display responsive.  
You may want to try traffic shaping.  Have the server kernel give the highest traffic priority to the Xwindows protocols.  Your client display 
refresh rate will rely on how loaded the server is.

The next challenge is migrating processes off the server.  The easiest way I've found is to tell mosix that the CPU speed is slower than it 
really is.  This has worked for me at least.

The other issue is process execution time.  The longer the process runs the more likely it will migrate to another node.  The next time you  
compile the kernel do the following.

Take the number of nodes you have and add one to it.  If you have any nodes that have multiple CPUs, count each CPU as a node.  My 
example cluster gives me 6.  Type

make -j6 bzImage
make -j6 modules

Watch the load on the cluster.  Your mileage will vary on this one:)

Good luck
Richard Camp

> I know one thing for sure, with the 4-5 PC's I had running on 10Mbit, that
> poor little hub was smokin' when I would stress test the cluster--solid
> green activity lights across the board.  The collision light wasn't solid
> but it was flashing *a lot*.  Watching the lights is not a very accurate
> measure of traffic of course, LOL, but with no management facility on it,
> that's the best I tell you.
> 
> When the cluster was under load it still appeared that the clients Xwindows
> refreshes were working well (even better then  with non-OpenMosix kernels.
> Especially on that old 486-66mhz.  It really made it feasible & worthwhile
> to use that old dog as a client.).
> 
> As you mentioned it will be very interesting to see what happens on a large
> set up.  I'm guessing that a Fast Ethernet (100Mbit Full Duplex) network
> would be able to handle the load even on a larger install.
> 
> The big concern, to me, remains--what happens when Client #1 process A is
> running on Client #5, and Client #5 powers off.  What happens to process A?
> The failover factor...
> 
> It may just be that OpenMosix would need to be a cluster of servers only at
> this point.
> 
> Valid questions all, and it would be great fun to test this stuff in a real
> lab.
> 
> James
> 
> ----- Original Message -----
> From: "Guy Lessard" <glessard coll-outao qc ca>
> To: "James Jensen" <jmsjnsn linuxfreemail com>
> Cc: <K12OSN redhat com>
> Sent: Thursday, April 04, 2002 8:23 AM
> Subject: Re: [K12OSN] Re: MOSIX
> 
> 
> > Boy all this mosix-k12ltsp stuff is very interesting.
> > I was wondering what will be the impact of having processus eating up the
> > bandwith of the network, when will the screen-keyboard refresh start
> hurting
> > because of all these processus flying around?
> > I guess this is a early question but a large installation of this
> mosix-k12ltsp
> > setup would start giving us answers.
> >
> 
> 
> 
> _______________________________________________
> K12OSN mailing list
> K12OSN redhat com
> https://listman.redhat.com/mailman/listinfo/k12osn
> For more info see <http://www.k12os.org>
> 
> 

-- 





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]