Re: another laundry or shopping list

["Albert D. Cahalan" <acahalan@cs.uml.edu>]

Jordi Polo writes:

>>> Meaning in CPID terms, that a signal for a cpid would get redirected
>>> by the home node, which is obvious from the CPID.
>>>
>>> That's for process migration.
>>
>> This is so... wrong. Now you have two nodes that must work OK,
>> increasing the chance of failure. You increase network traffic
>> and reduce performance.
>>
>> There are two good ways to do process migration. The first way
>> is with SSI. The second is totally userspace. Doesn't Condor
>> support this? Obviously the userspace solution has limits, but
>> it ought to be the best performing.
>
> I agree we double the chance of fail but this about issues you can't
> take away home node. You need some of the home node syscalls. Think
> in a 5 nodes cluster,all of them connected with his own modem to
> internet.

OK, that would be: /dev/ttyS0, /dev/ttyS1, /dev/ttyS2...
Any modem can be used from any node.

> 2 of them with telnet and ftp open,

OK. Note that they have the same local IP address.

> as the processes has
> that sockets open you need a little of it being bound to that node.

No, although migration away from IO might be a dumb idea.

> will be all the machines in the cluster have the same time ? maybe
> you don't want it and you want the time of your home node not
> others.  If you find a way to avoid this (not in userspace) i'll be
> really happy .

All nodes follow POSIX time, which is similar to UTC.
(time zones are done with libc and environment variables)

Time skew causes trouble even with plain NFS, so what
else is new? Add delays as desired to avoid time warps.



Linux-cluster: generic cluster infrastructure for Linux
Archive:       http://mail.nl.linux.org/linux-cluster/