| Breck Carter [TeamSybase] 2005-06-19, 8:23 pm |
| Please implement some kind of CONNECTION_PROPERTY performance counter
for CPU time (or processor time, or timer ticks, or whatever is
available).
Neither high precision nor high accuracy is absolutely necessary.
However, it should be possible to identify which connection is taking
most of the CPU time when the overall usage spikes.
At one client, for example, there is a problem with rogue connections
that cause blocks and/or consume high amounts of CPU time. Everyone
knows we should fix these problems at the application level but the
reality is that, at the moment, other issues take priority. The client
wants to automatically drop connections that cause problems. We can do
this for blockers (by enhancing our Block Sniffer program), but it's
hard to identify the Queries From Hell on a real-time basis... temp
file space usage helps but isn't good enough.
We need to increase throughput by a factor of 10, and could probably
increase it by a factor of 100 if it weren't for the rogue
connections; they are a real showstopper. Yes, we are considering
BACKGROUND_PRIORITY,
but that requires application changes, plus the
result (sloooooowing down the big queries instead of killing and
rerunning them later) is not popular with management.
Even a crude measurement of processor effort would satisfy this
requirement.
Breck
--
SQL Anywhere Studio 9 Developer's Guide
Buy the book: http://www.amazon.com/exec/obidos/A...7/risingroad-20
bcarter@risingroad.com
RisingRoad SQL Anywhere and MobiLink Professional Services
www.risingroad.com
|