Manuel MSB
7/12/2004 7:41:00 AM
Hi,
I am currently evaluating the performance of .NET Remoting. I build some tests to
measure the performance of the different .NET Remoting features. (Different
formatters, channels, CAO, SAO â?¦). I do this by simple taking the time span
which a method call on a remote object takes.
The tests work fine but the problem is I get unusual results. I think I should explain
you first how my tests are working. I have 9 test methods. These test methods are
executed on a CAO, SAO Singleton and SAO Single Call. Which are running in the
4 combination of protocol and formatter (e.g. HTTP â?? SOAP).
The tests methods range from methods with no return values and no parameters
to methods which get and return huge serialize able objects. I run these tests on
different combinations of 3 PC s which produced 6 results (PC1 as Server PC2 as
Client, PC3 as Server PC1 as Client etc).
But the results which I get seem very unusually. One combination seems to create correct 3 seems to create half correct and 2 seems to create incorrect results. The
part that looks incorrect is, that the method that takes nothing is surprisingly much
slower than the other one which gets and returns huge objects (I call them on the same remote object). This canâ??t be!?
I m thinking that the easy methods (not much data to transfer) are taking
much too long time. Interesting is also that in some Client â?? Server
combination this is only on HTTP SOAP CAO and HTTP BINARY CAO the case
and in some never and in some in all channel â?? formatter combination.
Here some results:
HTTP BINARY CAO:
PC 1 to PC 2
Normal Method: 5ms (No return no parameter)
Big Parameter Method: 287ms (Huge parameter ~20k)
Big Return Method: 305ms (Huge return value ~25k)
(Seems to be correct)
PC 2 to PC 1
Normal Method: 200ms (No return no parameter)
Big Parameter Method 50ms (Huge parameter ~20k)
Big Return Method: 176ms (Huge return value ~25k)
(Strange)
I am sure the tests are correct as I check the parameter, the return values and if they are running on the server. I also run them several times in the same application as I know the overhead of the first call. The problem is that one and the same test produces different results on different machines (By the way the machines can't be the problem as the fastest combination brings the worst result. No firewall is installed and no other processes are running!).
Perhaps it has something to do with the configuration of the PCs or with caching.
However it is very important for us to figure out what the problem is because
performance is very significant and 5ms or 200ms is a huge different. Perhaps
one of you have some idea what goes wrong.
It would be very helpfull if someone could give me some advice on that.
Thanks,
Manuel