Significance of all-zero checksum counts

Vernon Schryver vjs@calcite.rhyolite.com
Mon Jan 5 18:54:28 UTC 2004


> From: Gary Mills <mills@cc.UManitoba.CA>

> ...
> Doesn't the flooding between DCC servers keep then synchronized?
> Maybe it's the time lag between updates that contributes to this
> problem? 

Only reports of bulk mail are flooded among DCC servers.  The network
of DCC servers is now hearing about more than 70,000,000 mail messages
on busy days.  If every report were flooded, then each "flood" would
involve more than 7 GBytes/day.  See
http://www.dcc-servers.net/dcc/graphs/big.cgi?BIG=all-spam

>           In any case, if it only results in the counts being zero
> instead of one, it's not a big deal.

Close.  It only results in counts being zero instead of a value less
than your `dccd -t` threshold.

> > The best way to avoid such zero counts is to make DCC clients
> > prefer a particular server with RTT adjustments in /var/dcc/map.
>
> Won't this defeat the load balancing?  We sometimes get wide variations
> in server loading.  Having multiple DCC servers is essential in those
> situations.

Whether the effects of RTT adjustments defeat the load balancing
depends on how completely you want to balance the load.

> How much RTT adjustment do you recommend?

I use "RTT-500 ms" or "RTT-1000 ms" to bias the load balancing.  

I think I recall that you use Solaris.  `dbclean -F` and perhaps `dccd -F`
in 1.2.26 might help your installation.  As far as I can tell, Solaris
likes to do the equivalent of an asynchronous fscyn() on every
modification of a file mapped with mmap().


Vernon Schryver    vjs@rhyolite.com



More information about the DCC mailing list

Contact vjs@rhyolite.com by mail or use the form.