I have attached a patch to stat.cpp and stat.hpp that:
(1) Changes the type of M_counters to unsigned long long from unsigned 
long
(2) Instead of tracking the count , average, and average of the squares in 
M_counters and the sum and sum of squares separately in individual 
instance variables just tracks the count, sum, and sum of squares in the 
M_counters array.
(3) Creates a new function, computeMean that calculates the mean when 
required (i.e., when it is displayed) rather than on every single update.

This has a few advantages:
(1) Because it uses unsigned long long exclusively there is no reason to 
worry about lost precision anywhere.
(2) We only need to do the division for averages when it is needed, the 
rest of the time we can just add.
Also, I couldn't locate where the periodic sums were reset; so either I 
missed that spot in the original code or there was a bug.

I also have a few questions/notes that I want to get some opinions on:
(1) The -trace_rtt option uses a different method of calculating the RTTs 
than the RTD partitions and averages, would it make sense to unify the two 
so that the observed and computed values are in sync?
(2) The update_nb option seems to take great care to avoid calling 
getmilliseconds() on every loop.  Is there a particular system that this 
call is very expensive on?
(3) The -trace_rtt option provides sub millisecond resolution, would it 
make sense to provide that same resolution for the timing loop, and RTD 
counters?

Charles

--
Dr. Charles P. Wright
Research Staff Member
Network Server Systems Software
IBM T.J. Watson Research Center

Attachment: stdev.diff
Description: Binary data

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Sipp-users mailing list
Sipp-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/sipp-users

Reply via email to