[Planetlab-users] Node availability
llp at CS.Princeton.EDU
Mon Jun 14 15:07:46 EDT 2004
It would be great to define a widely-accepted definition of node
and more generally, PlanetLab goodness. A single integer would be great;
then I could program my ambient device to glow in a meaningful way.
Much of what we depend on today uses remote probes: can I ping the
can I ssh into it, does DNS resolve its name, and so on. Keeping those
from falling too low is harder than any of us anticipated, partly
we're constantly battling 160+ local network administrators, but Neil
that there's much more to it. The node manager needs to be correctly
vservers, users need to be congestion controlling their packet streams,
files need to not consume the entire disk, cron jobs need to not be
on themselves, and so on. Part of the problem is that our support for
isolation has lagged resource demand; we need to monitor (and probably
to) the obvious, sooner rather than later.
You've also raised the next interesting problem we'll need to address:
nodes with the sensors (services) you need.
P.S. When I say "we" I mean everyone, not just the support team.
On Jun 14, 2004, at 2:22 PM, Sean C. Rhea wrote:
> On Jun 14, 2004, at 11:05 AM, Neil Spring wrote:
>> I can get 260 nodes to run scriptroute, but that may be an
>> overestimate of node availability because a few of these hosts are
>> running my code even though I can't log in. It looks to me that
>> codeen http://codeen.cs.princeton.edu/status/ has 150+ hosts
>> participating (though I believe they're more interested in having all
>> sites than having all nodes; I'm sure Vivek and crew would have more
>> nodes than I do if that was a goal).
> FYI, I ignore Internet2-only machines, and I have Bamboo/OpenHash
> running on about 195-205 hosts at any given time these days. The
> other day I got 230 or so up, but that's the most I've ever seen.
> Men who have discovered the limits of arrogance make better
> company: You notice more when you're not running around
> imposing your will on everything.
> -- Virginia Vitzthum
> Users mailing list: Users at lists.planet-lab.org
More information about the Users