[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Physics application using xmlBlaster
I'm evaluating xmlBlaster as a candidate for a message passing system for
a data acquisition system to be used by a variety of high energy physics
experiments. The publish/subscribe model fits perfectly this kind of
application, but it often requires message rates peaks of 100KHz. Size of
messages vary from a few Bytes up to a MByte. We have successfully used
commercial publish/subscribe software in the past, but as experiments grow
larger the licensing price becomes a big problem.
So far, on a few PCs plugged to a switch using 100Mb ethernet I've got
rates from 50Hz to 200Hz, where each application (xmlBlaster, publisher
and subscribers) run on a separated node. These rates were measured with
one publisher and three subscribers at most, sending 100 bytes long
Have someone already done this kind of performance tests with xmlBlaster?
On the web page I see only numbers from a single machine.
does 100KHz mean to send 100000 messages per second?
Well, you are a lucky guy, to have so many data :-)
xmlBlaster can't handle this with today hardware.
Even with raw socket connection this is almost impossible.
Sending only one byte and this 100000 times a second
will occupy on an ethernet with 1500 bytes of package size
If you collect for example half a second the messages
on client side and send them in a bunch (we call it burst mode)
could help, but still your requirements are
difficult to solve.
xmlBlaster has the burstmode for callbacks not yet implemented
(it is scheduled - but this can be in years ...).
Marcel + Heinrich
mailto:ruff at swand.lake.de