not that I did not mention "according to nominal performance"....
fortunately there is an easy temporary solution
I ran my little benchmark on bud19 (root -l -b -q charge.C ####)
_ the becnhmark was from spud8 : took 61 s
_ from bud19, all /net/spuds mounted: took 204 s
_ removed all /net/spuds and the /dblast07 nfs mount: took 27 s !!
The temporary solution is to only mount /net/data/4, and bud24 on the
buds. Then they are ~2 times faster as you would expect. You can
still access the analysis/data directories (as I did in the above
benchmark). I have just done the change on all machines
Note: I tried also experimenting with different nfs options (i.e. "soft"
instead or "hard" mounts). We can also play with allowing also "explicit
mounts" with the user,noauto option but over time that may build to the
same problems seen above since you need to umount on edisk, before
mounting another (I guess it will be the way to go...)
-- ________________________________________________________________________________ Tancredi Botto, phone: +1-617-253-9204 mobile: +1-978-490-4124 research scientist MIT/Bates, 21 Manning Av Middleton MA, 01949 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This archive was generated by hypermail 2.1.2 : Mon Feb 24 2014 - 14:07:30 EST