From: Thorbjoern Ravn Andersen
How will you handle if the system simply cannot handle
more requests per second?
I change the subsystem class for all the ODBC/JDBC database server jobs to a
time slice of 1 millisecond, and knock the run priority down to 99, and allocate
just 1 meg RAM to the subsystem. We only need those jobs for serving .Net and
JEE clients. It's fun to irritate those people. That frees more resources for
all our RPG CGI programs ;-)
Of course that's facetious, but it illustrates a point. How do you gracefully
accommodate increasing workloads? You find a system that is essentially self
tuning. Every now and then you use it's workload management functions to
address pressing needs. Maybe you're hosting New York and Montana as clients,
but New York is late paying their bill. Adjust a few tuning parameters, and all
of a sudden Montana is getting great responses, at the expense of New York.
Okay, if your box is pegged at 100% CPU utilization 24X7 and customers begin
complaining about performance, then you call IBM and order the activation key
for a capacity on demand upgrade. You enter the key, and suddenly your server
is running on 4 sockets instead of 2. CPU goes back down to 50%.
How do .Net operators manage their workloads? Tune Windows? Stop the virus
checker? Tune IIS? Tune asp.Net? Add another server to the farm? Set up
another VMware partition? Deploy applications across more asp.net instances?
Is that someone's idea of gracefully supporting higher workloads?
-Nathan
midrange.com
