[RC5] Stats suggestion

Pyke, David David.Pyke at CIBC.com
Fri Jan 5 14:41:15 EST 2001


But a simple decay factor of .0111 will allow for a weighted average.
Yesterdays value gets multiplied by .9889 (technically 89/90) and the added
to today's multiplied by .0111 (1/90) .  This gives today's a full weight
and the past 89 days a reducing factor with each specific day declining over
time.

This gives a much more faster reaction to changes in daily rate and gives an
average that doesn't take years to get to current average rate.

(This from me who now averages 4Mkeys but only has an "average" rate of
1.7Mkeys

> -----Original Message-----
> From: Ben Clifford [mailto:benc at hawaga.org.uk]
> Sent: Friday, January 05, 2001 3:40 AM
> To: rc5 at lists.distributed.net
> Subject: Re: [RC5] Stats suggestion
> 
> 
> 
> > The problem we face is a technical one:  Much of the speed 
> increase I've
> > attained in the rewrite of stats came from eliminating the need to
> > reference the huge "historical work" table.  Instead, the 
> total work and
> > total rate are calculated from a running total. 
> 
> It is possible to created an exponentially weighted running 
> average from
> just the previous days exponential average and todays total.
> 
> Exponential averages are like unix load averages from the 
> uptime command,
> if anyone has seen those.
> 
> The exponential average gives more bias to more recent 
> results, with old
> results gradually fading in significance.
> 
> This average should achieve a lot of what is needed - it 
> stops taking into
> account very old results, and can provide a reasonable summary of the
> recent past.
> 
> Ben
> 
> 
> --
> To unsubscribe, send 'unsubscribe rc5' to 
> majordomo at lists.distributed.net
> rc5-digest subscribers replace rc5 with rc5-digest
> 
--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list