On Thu Oct 26, 2006 at 19:18:03 -0700, Aaron Glenn wrote: >On 10/26/06, Oliver Hookins <[EMAIL PROTECTED]> wrote: >> >> I guess my question at this point is, what do you do when you have say 1GBps >> of traffic to record and you need 100% accuracy? >> > >you use the SNMP octet counters and poll them at regular intervals. >but guess what - you'll generally end up with the same value. what >application do you have that requires precisely 100% precision? I >can't think of any. Nearly all ISP billing these days is, and has >been, done by sampling interface octet counters at 5 minute intervals, >and then chucking the top 5% of the sampled values (roughly). using >sampled flow data for billing is actually statistically *more* >accurate (depending on your sampling method, of course).
Problem with that is that we calculate billing based on the inbound/outbound data to/from an IP address, so we can't use counters based on physical switch ports. So big ISPs and companies actually use sampled data and then extrapolate to provide billing information? I had no idea. Unfortunately the lowest sample rate our switches go down to is 20 but this may still turn into large amounts of data with lots of flows... I'll keep looking into it. Thanks everyone for your help. -- Regards, Oliver Hookins _______________________________________________ pmacct-discussion mailing list http://www.pmacct.net/#mailinglists
