Hi list,

Small follow-up for completeness-sake:

I decided to ditch the (My)SQL storage backend all together and go with 
a "plain" RRD setup.

One RRD for each IP, with DS's for bytes_in, bytes_out, packets_in and 
packets_out. I keep 5 minute samples for a year and hourly avg/min/max 
samples for 5 years. This results in 7MB per RRD file. I might even increase 
the history to 10 years (for now I also keep the raw data in text files).

As the box has a fair ammount of RAM, most of the RRD stuff happens in memory, 
yielding pretty decent performance. The ammount of memory and/or more recent 
rrdtool versions account for a much improved performance compared to earlier 
tests.

On a side-note regarding SQL partitioning:
The way postgresql handles partitioning isn't all that different from what I 
had implemented myself (using MySQL) manually. In fact, with postgresql most 
of the partitioning stuff is "manual" (eg: no dynamic creation of new 
partitions when needed). In the end it turned out more complex for my setup 
as well.

-- 
Regards,

Ruben Laban
Senior Systems and Network Administrator
ISM eCompany

_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists

Reply via email to