67 lines
3.2 KiB
Plaintext
67 lines
3.2 KiB
Plaintext
rpcrt - RPC runtime performance tests:
|
|
Null Calls, Buffer I/O, Binding, Context Handles, Etc.
|
|
bin\*\rpcsrv -h // The server controls which test(s) get run, read
|
|
the usage for info.
|
|
bin\*\rpcclnt -t <protseq> [-e <endpoint>] [-s <server>]
|
|
|
|
workset - Rpc client and server applications used to measure working set.
|
|
start bin\*\wssrv [-t protseq] [-e endpoint]
|
|
start bin\*\wsclnt [-t protseq] [-e endpoint]
|
|
Client and server applications pause to allow working set measurement:
|
|
Server: startup, after listen
|
|
Client: startup, after first call, after flush and another call,
|
|
after flushing the server and making another call.
|
|
|
|
lpc - Raw LPC client and server
|
|
start bin\*\lpcsrv
|
|
start bin\*\lpcclnt [-i iterations] -n case -n size
|
|
Cases:
|
|
1: Uses the message for data transfer, size limited to ~225b
|
|
2: Uses Nt{Read,Write}RequestData to move data, unlimited size.
|
|
In case 2, its behavior is different (faster) than LRPC.
|
|
|
|
context - Measures context switch time using a variety of mechanisms
|
|
start bin\*\context [-n type] [-n case]
|
|
Types:
|
|
1: Both client and server, context switch between two threads
|
|
2: Server of process-to-process context switches
|
|
3: Client of process-to-process context switches
|
|
Cases:
|
|
1: Uses an NT eventpair
|
|
2: User an NT eventpair, but client uses two steps (sethigh, waitlow)
|
|
2: Uses two win32 events and WaitForSingleObject.
|
|
3: Uses two win32 events and WaitForSingleObject w/ timeout.
|
|
|
|
Note: Use lpc test to measure pp context switch via LPC.
|
|
Note: Use pmsg test to measure pp and tt context switch via PostMessage()
|
|
|
|
postmsg - Measures time to do process-to-process (pp) and thread-to-thread
|
|
PostMessage() calls.
|
|
Note: The server is a windows app which doesn't display a window,
|
|
it's running, really it is!
|
|
start bin\*\pmsgsrv
|
|
bin\*\pmsgclnt [-n case]
|
|
Cases:
|
|
1: PostMessage (thread-to-thread)
|
|
2: PostMessage (procress-to-process)
|
|
3: PostMessage and win32 event. (tt)
|
|
4: PostMessage and win32 event. (pp)
|
|
Note: The postmessage and win32 event version forces a context
|
|
switch on every post message. This is more OLE-like. Otherwise,
|
|
in cases 1 and 2, the messages get qued-up.
|
|
|
|
local - A variety of system/machine/processor/memory measurements.
|
|
bin\*\local -n case
|
|
Cases:
|
|
1: -n start -n end - memcpy calls (maximum cacheing)
|
|
2: -n start -n end - strcpy calls (maximum cacheing)
|
|
3: -n start -n end - strlen calls (maximum cacheing)
|
|
4: -n size - allocate and free 'size' memory blocks.
|
|
5: SetEvent and ResetEvent calls (kernel traps)
|
|
6: GetTickCount calls
|
|
7: Open and query a registry key
|
|
8: Query a registry key
|
|
9: Create and set a key (* must run first, before 7, 8 and 10)
|
|
10: Updating (setting) a key
|
|
|