|
Dear All,
I found an imbalance case on RR scheduler.
I have three real servers and their hardware & software are
identical.
I think RR scheduler would be the best choice for this situation.
I use ab(Apache Benchmark) sending http requests of the same page to
dispatcher continuesly. But after a long time test the load of them are not
same as I expect. One of them is get much more pending connections
than others.
I explain this by
If one of there servers get a bit higher load than others(in a dynamic
environment, it happens usually) once, the time of it to finish a request
is longer than others. That means it have one more pending request to
finish
than other. But the RR scheduler still treats they have the same ability to
finish a
request and still dispatch the same propertion of requests to them. So
this poor server will get the the same quantity of requests as others but it
never
finish these requests in the same time as others.
The result is that the server with a bit higher load at initial time will
get
more and more loaded.
Any one have the same experience?
|
