click_chatter v/s performance

[Douglas S. J. De Couto]

well here is an anecdotal result, from userlevel.  we have a ~17 node 
(computer) wireless network running a protocol written using click 
userlevel.  every 1.3 secs or so, every node sends a routing update 
packet.  each node logs every received routing packet (about 20 short 
lines of data) to a chatter channel.  i.e. nothing gets printed unless 
you connect to the chatter socket.  when we use netcat (nc) to dump the 
chattersocket output to a file on every node, throughput on the network 
drops by aout 50%.  this is on celeron boxes using aironet 11Mbps 
wireless cards.

the best way to find out how your particular logging demand affect 
performance is to try it out.  if you have a problem, work on reducing 
your logging overhead.  for example, i have managed to reduce our 
protocol logging byte-overhead by almost 2 orders of magnitude by 
switching to a binary format.  i also switched from using the chatter 
socket to using write() to dump data directly to the filesystem, 
avoiding the overhead of the network stack and anothe rprogram (nc) 
dumping the data.  this is all in preparation for logging on ipaqs (200 
MHz strongarm)....

Arvind Venkatesan wrote:

> Hello everybody,
> 
> I was just wondering about how much do
> print outputs through click_chatter (say one click_chatter for every 10
> data packets for statistics gathering) affect a router
> performance/introduce noise, since it does take system resources to do
> so.
> 
> 		Actually, I tried using StringAccum object and the
> append methods it provides, so that I can do one click_chatter for every
> 100 data packets(instead of once every 10 packets) and printout the
> buffered statistics. For some reason it prints out the buffered
> statistics and some junk(that confuse the awk scripts) after every
> cycle. I am taking care to clear this StringAccum object after every
> cycle of 100 data packets. How do I get rid of that junk? Am I missing
> anything here?
> 
> 	Is there any better way of collecting statistics (in case
> click_chatter has some kind of impact 
> on the router performance) ?
> 
> Thanks for the response!!!
> 
> -arvind
> 
> 
> 
> 
> 	
> 
> 


-- 
Douglas S. J. De Couto   decouto at lcs.mit.edu