nmon for AIX - 15 years on and still learning
nagger 100000MRSJ Visits (5817)
I have been running the nmon (nigel's monitor) for AIX project for 15 years now but I am still learning.
I initially wrote it to help me run AIX 3.2 then AIX 4 benchmarks on RS/6000 machines as running the older "down the screen" and "line by line" tools like vmstat and iostat was driving me mad (literally making my eye hurt) and so using screen curses enabled the number to stop moving on the screen. nmon's second most important feature was to make graphing the results simpler as a graph includes the time dimension, it gives us lots of clues about what is going on and how events tie together like a peak in the CPU use and disks. But as every benchmarker knows the final benchmark report needs at least 6 graphs. The number is sufficient to make managers at IBM and the customer's eyes glaze-over with too much detail they don't understand and just stop reading further - Oops!! Am I being too cynical again :-) We started taking the nmon output data file into Lotus 1-2-3 and then manually creating the graphs at first. Then Steve took pity on us and automated the graphing for us using 1-2-3 scripts and eventually the Excel macros that we use now.
Stephen (Mr nmon Analyser) Atkins and I recently had email from a nmon user in India asking about Weighted Averages - to which my response has always been - Weighted Average is the average when the resource is not zero and ... that is wrong! So ignoring any very quiet periods. This particularly helps with disk stats when some disks (and the filesystem in them) are only used for some periods, say parts of a batch run or cycled through like rotating RDBMS log files or benchmarks ramp-up and down periods.
Well, I was wrong! And Steve put me right - Weighted Average is the sum of squares divided by the sum - so large values have a larger influence. This is a better calculation as we don't often have genuinely have zero values, just very small ones and you don't have to invent some ignore threshold. This just shows that Steve is a clever chap and yet again adds value. He also pointed to the nmon analyser documentation (NA_Userguide v33.doc) - which you get when you download the Analyser from nmon Analyser wiki page. If you are not using the nmon Analyser version 3.3 then I am embarrassed for you! Please update ASAP. And don't forget to get your car brakes serviced - as you probably forgot that too.
I have always rated the Analyser as 50% of the reason why nmon has become so popular over the years. It is a pure guess but as I had 1500 on my email list at one time I guessed there were 10,000 users and each running it on say 10 machines on average so that makes 100,000 copies running. My biggest mistake was note adding technology that some how charge $1 each time it starts :-) But then the zero cost was perhaps the second most attractive feature and a charge would have killed off nmon completely. The emial list is now defunct in favour of new community technology like ... this blog and Twiiter.
For the record,
I hope this helps, thanks, Nigel