Blueback IO load
From PTAGISWiki
Input/Output load for the machine blueback has been captured by a cron job since mid-2003.
The data was collected through the following command:
/usr/bin/sar 1 2 | grep Average
Which generates output that looks like this:
Average 0 0 0 100
That results are then manipulated by the cron job to produce output like this:
Report on server Administration
HGS.SMS.402-A
Account Name: carters Location:/home
Memory statistics for server: blueback on 1 Jan 2004
|------------------------------------------------------------+---------------------------------------------------------------|
| Memmory Statistics | CPU statistics: Portion of time the CPU running in .... |
|----------+------------------------+------------------------|---------------+---------------+---------------+---------------|
| TIME |Real Memory (free/total)|Swap Memory (inuse/free)| User mode | System mode | Blocked i/o | Idle |
|----------|------------------------|------------------------|---------------|---------------|---------------|---------------|
| 00:00:06 | (8271M/12G) | (3180M/16G) | 85 | 10 | 0 | 4 |
| 00:10:02 | (8491M/12G) | (2942M/16G) | 45 | 17 | 5 | 32 |
The log files for the data include the date of the first run and data from subsequent years is concatenated to the end of the file. So a file named "blueback.Jan1" contains a record for every 10-minute interval on Jan 1 2004, 2005, and 2006, assuming that the cron job successfully ran on each of those days.
With that assumption in mind, I generated the following graph which represents the average time the CPU spent blocked in an i/o wait state over a two-minute interval, with samples taken every 10 minutes.
(Note: The holes in the data set began when the carters home account was moved off of blueback and onto some other machine, so carters cron on blueback couldn't find the code to execute when the cron wanted to start the job.)
