This is a discussion on Last week ... - Solaris Rss ; Wasn't a really good week. A lot of work to do. And at the end i've got ill ... a ruined stomach killed my plans for the weekend. My thanks goes to the colleague which brought his germs into the ...
Wasn't a really good week. A lot of work to do. And at the end i've got ill ... a ruined stomach killed my plans for the weekend. My thanks goes to the colleague which brought his germs into the office
I used the free day on Thursday to look at numbers. Millions of rows gathered DIM_stat records. This gig showed me that the most important tool to analyze such masses of data is a good statistical tool. Call me weird but R is an invaluable tool to get to conclusions. The problem when diagnosing performance problems is either you have no or the wrong data or you have to much data. Or it was measured differently and you have make sense out of it.
An interesting finding: There are professional data collecting tools showing massive peaks on all metrics when you measure them with a second tool. I'm pretty opinionated that a monitoring tool shouldn't introduce such large transient to the monitored subject. However it didn't disturbed the normal operation.
At least i got to some acceptable conclusions out of the data and thus i was able to make some suggestions to the customer.
Other conclusions of this week:
- Commands like trapstat, busstat and cputrack/cputrack deserve more attention
- An own pastebin system on you own server is really useful to archive data when you aren't allowed to connect your own notebook. I'm using codetrunk for this task at the moment
- No matter how long you work with Solaris, there are still surprises and moments of thinking "wow, cool design" in it for you
Read More about [Last week ......