From: <mos...@t-...> - 2002-12-30 07:38:24
|
Hi Kevin, On Montag, 30. Dezember 2002 01:13, Kevin N. Carpenter wrote: > Hi all. > > I'm currently playing around with a 5 node system. Node 1 exist at the > opposite end of a 14mbs network connection. Nodes 2-5 are connected via > a 100mbs hub (half duplex). > > If I crank up (5) CPU intensive processes on Node 1, they migrate nicely > to the other 4 nodes. However, apparently due to the slow interconnect, > the processes swap around a lot and vary in CPU usage from 40-96%. > > If I shutdown node 1, and crank up (4) processes on the remaining nodes, > they run at 99.9%. > > FWIW - I'm using the ECCP2 DC executable. This process has a 496KB > working set size per TOP, when running locally. When run under > Openmosix, TOP shows 204KB for the remote tasks (shown as sleeping) and > 316KB for the local task (shown as running). > > What can I do to use all 5 nodes efficiently? All (5) are the same with > regards to CPU & motherboard. Node 1 has a bit more memory and a hard > disk. Nodes 2->5 have less memory and are diskless. node1 is "less" used by openMosix because of the slow network link. You can increase the openMosix-speed (mosctl setspeed xxxx) on node1 and "fake" the load-balancing algorithm. That will cause that processes easier migrate to your node1. > > Thanks, > > Kevin C. <snip> many greets, Matt -- E-mail : mos...@t-... www : http://www.openmosixview.com an openMosix-cluster management GUI 1. First they ignore you 2. Then they laugh at you 3. Then they fight you 4. Then you win ! |