hello,
I have encountered a situation where on one server, the
gigabit network interface has been saturated because a couple of files
are being read a lot (first time I see that, usually all the servers are
stressed a lot in that kind of cases). I switched to a setting like this
for olbd:
olb.sched io 100 maxload 80
which is not exactly working as I was expecting (or I was dreaming :-) ):
indeed when the load reach or is above 80, the server is not eligible
anymore ("olb_client defered;: eligible servers overloaded for
..../foo.root"). However, the new clients which are trying to connect on
this server to open one of its files, are ending with an error as the
server was not able to open the file. I would have thought that in that
case, an other server would have been chosen and the file being staged
from there if necessary.
I could replicate the most active files on other servers but I would
like to avoid that, as an other day it can be an other server which is
the target of the load.
thanks in advance for any hints,
JY
ps: using xrootd version 20071101-0808p1
|