[mpich-discuss] MPICH over WAN

[Nicolas Rosner]

I agree, with Michael; this sounds like the proverbial hammer-driven screw.

I'm not saying it's impossible, and I see why it may look like
distance should not matter. The smaller obstacles (DNS, tunneling,
port fwding, etc) would be many, but each shouldn't be too hard to
work around, with time and patience. And one could learn a lot while
at it, and we all like a challenge anyway, don't we? However...

Think of the factors beyond your control. Latency, delays, buffers
filling up; write error handlers for every call or watch your jobs
die. Debugging MPI code is hardly easy even under ideal conditions
(dedicated, identical hardware). Writing code for distributed,
heterogenous, volatile clients is tricky even with middleware that
actually focuses on that. Both at once is seriously harder than their
sum.

Take a look at the About (the "What is[n't] ___?" 2-paragraph thing, I
mean) pages for MPI, MPICH2, Condor, BOINC or so. The design goals,
intended audiences, what each will do for you. What do you want? If
you want to explore the MPI way, perhaps you could borrow someone's
old PC, or find a lab somewhere else that will take you as a intern,
or run an additional virtual box or two on your own machine? Just
thinking, anyway. Good luck!

N.




On Sat, Feb 5, 2011 at 7:37 PM, michael bane
<michael.bane at manchester.ac.uk> wrote:
> Maybe condor would be more appropriate.
> http://en.m.wikipedia.org/wiki/Condor_High-Throughput_Computing_System?wasRedirected=true
>
> Michael
> Http://www.Twitter.com/mkbane_mcr
>
> ----- Reply message -----
> From: "Eugene N" <neverov.biks.07.1 at gmail.com>
> Date: Sat, Feb 5, 2011 16:29
> Subject: [mpich-discuss] MPICH over WAN
> To: <mpich-discuss at mcs.anl.gov>
>
> Hi
>
> I a have a bunch of friends all round the town, whom i can
> persuade(blackmail, threats, usual stuff) to donate me their cpu time . But
> they all use different ISP, and into the bargain few have real IPs, most
> connect through NAT and their external IP changes every day or even hour.
>
> Is there a scenario wich will enable me to hook them machines up in to a
> MPICh ring? Perhaps I should use a VPN (hamachi for instance) to emulate the
> LAN?
>
> I wont be able to utilise uni network any time soon becouse of all the red
> tape. Right now it would be the only way for me to mitch up a cluster, i
> really hope some guru will help me out with advice.
>
> Thanks
> Eugene
>
>
>
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
>
>