[OpenAFS] UDP slowness remedies?

[Dan Pritts]

so first, i think your problem doesn't have much to do with the rate at 
which the machines can send packets and has everything to do with the 
chatty back-and-forth nature of the protocol.  I don't really know the 
protocol myself, actually, but this seems very likely; i've certainly 
run into it with other protocols/applications, e.g. TSM backups and 
windows file sharing.

I'm guessing you are going between Bloomington and Indianapolis so 
latency shouldn't be too high, but even 10ms surely will add up if the 
conversation goes back and forth a million times.

I'm pretty sure you can run multiple vos move's in parallel, which would 
help dramatically.

as far as your iperf results, my experience is that tuning UDP buffers 
generally is not necessary; the defaults are usually sufficient to get 
hundreds of megabits.

in UDP mode, iperf does not attempt to scale the bandwidth; it tries to 
send at whatever bandwidth you specify on the acommand line.   the 
default is 1Mbit/sec...is it possible that's where your 1mbit result 
came from?


Eric Chris Garrison wrote:
> So, I'm doing a mass migration of AFS volumes from one campus to
> another.  Both machines have Gigabit interfaces, and we have a 20 Gig
> link between the campuses that I understand to be not at all saturated.
>
> However, I'm seeing a transfer rate of only about 30 Mbits/sec
> (according to iftop) between the two machines.  I'm working on it with
> my local network people, but it seems to be a UDP issue... if I use
> iperf with defaults for UDP, I get about 1 Mbit/sec.  If I increase the
> buffer size to 1024M, it screams along at 900+ Mbit/sec.
>
> So how's this work with AFS/rx?  Is there something I can tweak to
> increase UDP transfer speed?
>
> Any advice would be appreciated.
>
> Thanks,
>
> Chris
> _______________________________________________
> OpenAFS-info mailing list
> OpenAFS-info@openafs.org
> https://lists.openafs.org/mailman/listinfo/openafs-info