RE: 2.6 Kernel Userspace Daemon 0.1-0602131225 (Hopefully attached)

Thanks for the comprehensive feedback.
 
The "omkmigd <defunct>" issue (which I have also seen) is due to the om kernel, totally independent of the userspace daemon and well beyond my abilities to sort, I will leave that one to the experts :>).  The ACK's you observe should occur every 15 min, and serve as a fallback for the autodiscovery mechanism should a node be missed. ( Start the userspace daemon up on one node with the network cable unplugged, reconnect it and wait, it should "find"/"be found by" all the other nodes within 15 min of reconnecting the cable).
 
It is an intention of mine to support clusters of "peers", and I am trying to make sure that the source will support this feature in the future. At the moment the
2.6 kernel is still very much in development, it will better support the complexity of "peer" clusters when the new om admin filesystem is available.
 
I am currently working on a far superior incantation userspace daemon with less design mistakes (I have learnt a lot recently) which will be some months in the writing. The version you have is "network fragile" and will occasionally fail under heavy network load (such as booting diskless workstations, or clusters in excess of 64 nodes).  When it is alpha ready I will contact you with a view to testing it.

-----Original Message-----
From: Deadpan110 [mailto:[email protected]]
Sent: 28 February 2006 11:23
To: Carl Radford (DC0854-IT)
Cc: openmosix-general-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
Subject: Re: [openMosix-general] 2.6 Kernel Userspace Daemon 0.1-0602131225 (Hopefully attached)


Hi,

I have decided to use the linux 2.6.15 kernel patched with:
http://openmosix.snarc.org/files/releases/2.6/patch-2.6.15-om-1.bz2
as it is easier for me to compile one for use in gentoo.
(I eventually decided to build the kernel and the root for i686)

On the controller, I run ./runme.sh and on the drones, I run make all; ./omuserd -t3490 -u1334 -d
Then from the controller, I run several instances of timewaster. (8 instances for use on 4 nodes all with dual cpus)

Automatic migration works fine and I can see all the drone nodes running 2 or 3 instances of omkmigd each with a nice significant cpu usage.

Also on the network, I have 4 boxes running gentoo linux with kernel-2.4.26-om1 and these are correctly identified as bad nodes.

When the timewaster instances finish, i recieve:
omuserd(udp)[7628]: Received ACK from 192.168.1.19 (c0a80113)
omuserd(udp)[7628]: Received ACK from 192.168.1.18 (c0a80112)
etc

But on the actual drone nodes, I get: 
# ps aux | grep omkmigd
root       898  0.0  0.0      0     0 ?        S    07:03   0:00 [omkmigd]
nobody    7418  1.5  0.0      0     0 ?        Z    08:45   2:09 [omkmigd] <defunct>
nobody    7426  2.0  0.0      0     0 ?        Z    08:50   2:47 [omkmigd] <defunct>
nobody    7477  3.7  0.0      0     0 ?        Z    09:17   4:03 [omkmigd] <defunct>
etc

(no doubt I am proberly doing something wrong)

Anyways, my nodes will continue to be available for anything you wish me to try on them. :)

Also, I know Its early days yet, but will the Userspace Daemon support being a controller and also a drone on a single node to work like oM 2.4.26 userspace tools currently do?

Martin