GitLab

placing a "-1" anywhere in the latency or bandwidth matrices will result in
huge penalties when that connection is used in the mapping.

Also added neat (possibly novel?)
"mutation spike" which, when the standard deviation of the population gets
beneath a certain value, spikes the mutation rate from 1% to 50% for a generation,
thus keeping the deviation (and the population diversity) up.
Results from testing look promising.. in my "standard" 6-node mapping,
the solver finds the "best" solution in 95% of the runs.
(vs about 50% or less the old way.)

version of the data in widearea_delays . Schema for the two are
identical.

Also cleaned up the schema for both tables, so that they have proper
primary keys, etc.

at the user when things go south.

named all over the place).

the right.

but I've been pummeled into a mere (perl) shell of my former self.

longer necessary.

Added "crossover rate" to improve population stability.
Changed termination behavior to stop if, after, 256 rounds,
no better solution has been found (changable with the -r switch.)

(smaller population, non-in-place child building,
 proper 'roulette' selection function, etc.)

*<name of pnode>
to
<name of vnode> <name of pnode>

This means that fixed vnodes may have unique names
(not just the name of the pnode they are fixed to.)

Added "-v -v" mode,
"-v" now just outputs progress without "friendly" prompts/output
or error analysis.

Fixed bizarre bug in mutation engine that was wedging the solver.
(Should also allow for expts with only 1 non-fixed node.)

Added check to skip solving altogether if there are 0 non-fixed nodes.
(this also handles the edge case of an input with no virtual nodes.)

Added check to exit with errorcode if there are more virtual nodes than
can be mapped.

When we put nodes into the reloadpending experiment, don't put
an entry into last_reserve - to be safe, delete such an entry
if it is there. This prevents the double-reload problem we've
been having, found by Leigh.

in /etc/testbed for use by NSE

Whee!

- Halved penalty for overshooting bandwidth (undershooting remains the same.)
- Added penalty argument to -m for assessing multiplexing penalties.
- Made a few optimizations.

Small change for expts in TBOPSPID (emulab-ops for us): Instead of vname (which isn't usually defined), show when it was reserved to the expt. Great for checking on reloading nodes, hwdown nodes, etc.

if it fails, we exit gracefully from the node setup instead of
screwing stuff up even worse. Basically, I converted all close($TM) to
CLOSETMCC($TM) which does the proper checks and dies if error.
This is probably not ideal either, but not really sure what the ideal
solution is when tmcc fails.

So, the length of time is set in the .cnf file when signing a cert
with the CA. However, the length of time the CA is good for is not set
in the .cnf file (the entry is ignored). Rather, it has to be on the
command line. So, the certs really were good for 3 years; it was the
CA that had expired, and once that happens the certs are no longer any
good. Very bogus.

for a node, we start up NSE, even if no FullTcp traffic generators
are configured for the node

nodes that get moved to the reload pending experiment.  Without it,
tmcd won't free them from the reloading experiment.

Second, fix the case of freeing from the reloading or reloadpending
experiments.  The old code would get nodes stuck in their holding
experiment.