|
Hi, I wonder if any one is familiar with the following, and may
help explain what is happening: I have a Gigabit dedicated connection between two computers:
one running Enterprise Linux 3 (EL3), and the other running Fedora Core 5 (FC5).
I established an NFS connection in which FC5 exports a directory and EL3 mounts
it. I then start an application in EL3 that gets some “hard
real time” data and writes it out to FC5. It is a streaming process, so
it would be nice to do this for an extended amount of time. With this configuration, once the run starts I can see
network activity all the time, and activity on the drive in FC5 writing the
data. Really looks good. Unfortunately we have occasional events that take a
little bit longer than we can afford (hiccups), and the whole thing has to
stop. In trying to find out what the longer event is, I tried to
upgrade from EL3 to FC5, i.e. both computers running Fedora Core 5. All the
settings being exactly the same, I see a completely different behavior: Now as
I start the run, I don’t immediately see any network activity or
hard-drive write activity. When I finally see it, it happens at the same time
that I get a much longer hiccup. It appears that it wants to accumulate in a local buffer
long enough (about 100 MB) before it starts sending stuff, just that the buffer
is way bigger than before for some reason. EL3 is using kernel 2.14.21, FC5 is
the “bordeau” dist. I’ve tried to tinker with NFS parameters
(async, block sizes, etc) but wasn’t able to reproduce the EL3-FC5
scenario. Any idea what has changed? Thanks, Emanuel ---- Emanuel Machado, PhD. Senior Engineer, Project Leader Cytonome, Inc. Email: Emanuel-Machado@xxxxxxxxxxxx |
