Learning DCOM

Performance Results for ThreadingModel=“Apartment”

This section captures the performance results when the ThreadingModel of CLSID_Details is ThreadingModel="Apartment". To make the discussion easier, concentrate on the VoidCall column in each table. In Table B-4, note the following:

• For in-process method invocations, the client and the server share the same apartment. This implies that method calls are made directly, without marshaling.

• From the data shown in Table B-4, a direct method call is about 170 times more efficient than a local out-of-process method invocation. It is about 1800 times more efficient than a remote out-of-process method invocation.

For Table B-5 and Table B-6, note the following:

• The client apartment is different from the server apartment. This implies that marshaling will result.

• For all in-process method invocations, the client apartment marshals interface pointers between itself and a different apartment, a different STA.

• In-process and local out-of-process method invocations are essentially equivalent, when the client and server exist in different apartments.

Table B-4. Client Apartment: STA (Secondary STA Thread)

	FirstCall	VoidCall	ArrayCall	LongCall
In-process	0.000005	0.000006	0.000005	0.000007
Local	0.000812	0.000172	0.000183	0.001776
Remote	0.039747	0.001798	0.002399	0.017383

Table B-5. Client Apartment: MTA (MTAThread1)

	FirstCall	VoidCall	ArrayCall	LongCall
In-process	0.000103	0.000094	0.000103	0.000808
Local	0.000612	0.000091	0.000099	0.000795
Remote	0.037605	0.001695	0.001885	0.016981

Table B-6. Client ...