First of all, thanks for this extremely useful tool.
We use it to copy hd-images to a pool of 48 PCs, and it works
surprisingly well. At least, last semester it did :).
We use PXE to boot all clients into a small debian-system with r/o
nfs-roots, and cluster-ssh to control them all at once (eg., start
udp-receiver). Network is 100mbit half-duplex, the switch can handle
102gbit internally. Unfortunately, we have no physical access to the
On the server, we use this command to cast the complete harddisk:
cat /dev/hda | lzop -c - | udp-sender --half-duplex --max-bitrate 90m
On the clients, the opposite is:
udp-receiver | lzop -d - | dd of=/dev/hda
hda has about 29gb of used data, and a capacity of 80Gb. We do about 16
clients at the same time, and the first round went very well (it took
about 105 minutes to write the whole disk).
Now, the problem for the second 16 (which are the same hardware than the
first 16) is, that the speed of udpcast has dropped down from about
50mbit/s (with which it transferred about 24Gb) to right now about
50kbit/s (with which it has transferred about 5gb yet) (that's ok for
now, as they can finish writing the image over the easter-weekend), with
many timeoutnotanswered-messages on the sender.
These messages came from all clients every few seconds when the speed
was about 50mbit/s, but the client number 2 was overrepresented there
and the speed dropped down to about 50kbit/s, so I cut it off.
When udpcast reduces the speed to that of the slowest receiver, and that
receiver gets cut off, would it be possible that the speed gets up
again? At the moment it doesn't seem so ...
Thanks in advance, Lukas