[zeromq-dev] PGM at high throughput

[Robin Weisberg]

Another warning message I'm getting, this looks more serious as it results
in the test not completing. Linux to linux again. Here I'm testing larger
message sizes.

root at dev03:/local/home/data/robin/zmq-1.0.0/perf/tests/zmq# ./pgm_local_lat
wimp "eth0;226.0.0.1:7500"  "eth0;226.0.0.1:7500" 800000 100000
local_exchange network: zmq.pgm://eth0;226.0.0.1:7500
iface to connect to remote_exchange: eth0;226.0.0.1:7500
message size: 800000 [B]
roundtrip count: 100000
Start pgm_remote_lat on remote host and pres enter to continue.


** (process:3895): WARNING **: dropped 203 messages due to invalid NLA.

whats an NLA? What would cause this?

Thanks!
Robin


On Wed, Jul 22, 2009 at 2:12 PM, Robin Weisberg <robin.weisberg at gmail.com>wrote:

> I've been testing sending pgm from linux to windows and linux to linux
> using the pgm_remote_thr/pgm_local_thr and have a few questions maybe
> issues.
>
> 1) the pgm_remote_thr reports message gaps even for tests that last < 3
> seconds. I thought there was recovery being done at the PGM level for 10
> seconds? (pgm_secs is set to 10 in config.hpp). This is true for windows and
> sometimes linux. FYI, I previously changed pgm_max_rte to 125000000 in
> config.hpp
>
> C:\ework\zmqsvnv1.0.0\windows\Release>pgm_remote_thr.exe wimp
> "192.168.0.195;226.0.0.1:7500" 100 100000
> iface to connect to local_exchange: 192.168.0.195;226.0.0.1:7500
> message size: 100 [B]
> message count: 100000
>
> msg# 16742 GAP
>
>
> 2) On the linux side while publishing for extended periods of time I see
> this message. I'm assuming this means that the io thread is stuck sending
> because of full OS buffers and doesn't indicate a real problem that could
> cause message loss. Can someone confirm that?
>
> root at dev01:/local/home/data/robin/zmq-1.0.0/perf/tests/zmq#
> ./pgm_local_thr wimp "eth0;226.0.0.1:7500" 100 100000
> local_exchange network: zmq.pgm://eth0;226.0.0.1:7500
> message size: 100 [B]
> message count: 1000000
> Start pgm_remote_thr on remote host and pres enter to continue.
>
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
>
> .... (lots of these)
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
>
> ** (process:28101): WARNING **: sendto 226.0.0.1 failed: 105/No buffer
> space available
> Pres enter when pgm_remote_thr exits.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20090722/db60899e/attachment.htm>