[Orca-dev] Two levels of logging for disk and tape devices

[Dmitry Berezin]

Hi Blair,

 

Here is the second question about changes to the orcallator.se.

 

Implement two levels of logging for disk and tape devices -

 

There were several requests regarding collecting more or less statistics for
io devices. I would like to implement the following method for collecting
these data:

By default, collect only minimal (TBD) statistics for each disk/tape and
only minimal (possibly different, again TBD) totals for disk/tape
statistics. Define WATCH_DISK_EXTENDED (not defined by default) to collect
full set of statistics for each disk device (this will generate a lot of
data), define WATCH_DISK_EXTENDED_TOTALS to collect full set of totals for
all disks. Do the same for tape devices.

This approach enables fine control of how much data is collected without
modifying code at all - simply add -Ds at the command line in the startup
script.

Here is the set of available statistics:

 

 1. Percent of time transactions in progress

 2. Average service time in milliseconds

 3. Reads per second

 4. Writes per second

 5. Kilobytes read per second

 6. Kilobytes written per second

 7. Average number of transactions waiting for service

 8. Average number of transactions actively being serviced

 9. Average service time in wait queue in milliseconds

10. Average service time active transactions in milliseconds

11. Percent of time there are transactions waiting for service

 

At this time, 1 and 2 are collected for each disk as well as totals of
3,4,5,6, average of total of 1, and maximum of 1.

We can keep it this way or change it. It would be very helpful to hear
different opinions on this subject.

Also, there are two possibilities for dealing with totals of 1,7,8,9,10, and
11 - compute mean or median values (or both?). Again, any comments?

 

  -Dmitry.

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/orca-dev/attachments/20041013/1b6c2801/attachment.htm