[Planetlab-users] Down Nodes / Memory Usage
knauer at jf.intel.com
Wed May 12 17:01:58 EDT 2004
In an E-mail message, Emil Sit wrote:
>Could be interpreted to mean, "woah! isdl_tbh is using 40% CPU, 13% of RAM and
>3Mbps of bandwidth (for at least 15 minutes) by running 10 processes!"
OK, so I'm (along with some un-indicted co-conspirators) idsl_tbh.
We've -- quite literally -- only recently gotten to the point where
there has been the luxury of measuring and optimizing resource
consumption. So rest assured that that is the next step.
However, I think you must be including localhost/loopback traffic in
that number (if you are certain you aren't, PLEASE let me know! :) as
much of the intra-system communication happens that way. This is not
as much a design choice as something forced on us by the infrastructure
Lastly, if it's any consolation, I'm not running on anywhere near all
the nodes, nor at all the sites, and am generally only on one machine
per site. I could probably, in fact, get by with fewer (that's also
part of the research), but needed to have a herd available in order to
handle the down-machine and high-load-machine syndromes that started
this thread. Instances of idsl_tbh commit suicide, in fact, to be a
good citizen when other slivers make a machine's load too great.
Rob Knauerhase [knauer at jf.intel.com] Intel Labs / PlanetLab SRP
"I have been a happy man ever since January 1, 1990, when I no longer had an
email address. I'd used email since about 1975, and it seems to to me that
15 years of email is plenty for one lifetime." -- Donald Knuth
More information about the Users