SIGUnix

Minutes for 2000.10.26

NCSA Discussion

Possible Projects

High performance networking

can't use TCP; heavy performance degradation
drivers need to be multi-processor compliant
empahasis on latency, moreso than bandwidth

Filesystems for clusters

bandwith matters
NFS has bottlenecks

Administration issues

GLUNIX provides a single-node interface for cluster
desire is to have utilities such as ps and top work across the nodes of the cluster
MOSIX has too high of an overhead
desire to update patches across all the nodes

Current setup

using PBS (portable batch system)

Discussion after representatives leave

Things to look at

GLUNIX
parallel-fork call (pfork?)
one member will write up proposal over weekend
Beowulf FAQ

Projects

NJAMD

Mike Perry will continue to work on NJAMD instead of NCSA cluster

Bugfixing

file-dialog team met
other groups didn't meet; these will problem drop and be replaced by NCSA projects

Dreamcast

found docs on programming for the Dreamcast
close to getting GCC to compile; definitely possible
flashy demos are available

pftp

met
c-version unbroken
Perl version coded a lot more classes

LimitGuard

met
got source from feathers
got reply from person at CCSO; said he was interested, and forwarded infor to someone else
won't meet this week

Other topics

Choosing a chair

Frank was chair before Lars
no suggestions presented

EOH

Cuecat functionalities

You've been here, you might be interested in here
provide maps
To get throughput, possibly have incentives
one station per building; need to talk to EOH representiative to see if we can install machines in buildings, and get network accessibility
possibly give prize to every 500'th scanner
on other hand, focus on getting useful information to the users, and incentives will not be needed