diff options
Diffstat (limited to 'TAO/performance-tests/Cubit/TAO/MT_Cubit/README')
| -rw-r--r-- | TAO/performance-tests/Cubit/TAO/MT_Cubit/README | 215 |
1 files changed, 215 insertions, 0 deletions
diff --git a/TAO/performance-tests/Cubit/TAO/MT_Cubit/README b/TAO/performance-tests/Cubit/TAO/MT_Cubit/README new file mode 100644 index 00000000000..6a9a0ff666c --- /dev/null +++ b/TAO/performance-tests/Cubit/TAO/MT_Cubit/README @@ -0,0 +1,215 @@ +// $Id$ + +THE MT_CUBIT EXAMPLE +-------------------- + +This directory contains the source code for the TAO "MT_Cubit" +benchmark, which enables us to measure the real-time performance of +TAO. MT_Cubit measures the latency, jitter, determinism, CPU +utilization, and degree of priority inversion incurred by TAO twoway +operations invoked under different workloads and priorities. For +more information and empirical results, please see + +http://www.cs.wustl.edu/~schmidt/PDF/RT-perf.pdf +http://www.cs.wustl.edu/~schmidt/PDF/RT-OS.pdf + +COMPILING +--------- + +Makefile is provided. Use make to compile. + +You can either run the server in the background in the same window as +the client or open a separate window for the client and server. + +OPTIONS +------- +Options are case sensitive, e.g., "-U" has a different meaning than "-u". + +(1) server: + +./server [-e <endpoint>] // Endpoint to use (URL style endpoint) + [-t <number_of_servants>] // # of servant threads to create + [-f <ior_file> ] // specify a file to output all ior's + [-m ] // Use multiple priorities for threads + [-r ] // Run the thread-per-rate test + +[-e <endpoint>] indicates the endpoint the ORB should bind to. + This is useful when a machine has more than one network interface. + + Default Value: Whatever the current active pluggable protocol + uses as its default endpoint. + +[-t <number_of_servants>] tells the server how many servants to create. + + Default Value: 2 + +[-f <ior_file> ] Use to specify a file to write all the iors from the + different servants out to disk. + + Default Value: nil, which means not to write the iors to disk. + +[-m ] For the 1 to n low priority servants, this parameter indicates + the use of a range of priorities for the threads. + + Default Value: Disabled. One priority will be used for the + low priority threads. + +[-r] turn on the thread-per-rate tests. (Priority inversion tests) + + Default Value: Disabled. + +The server will print the IOR of the servants, but it is not required +by the client. + +(2) client: + +./client [<ORB OPTIONS>] // ORB options, e.g., "-ORBObjRefStyle url" + [-d <datatype>] // what datatype to use for calls + [-n <num_calls>] // number of CORBA calls to make. + [-t <num_of_clients>] // number of client threads to create + [-f <ior_file>] // specify the file from which we read the object references (iors), if any. + [-r] // run thread-per-rate test. + [-o] // makes client use oneway calls. By default, twoway calls are used. + [-x] // makes a call to servant to shutdown + [-u <requests> ] // run the client utilization test for a number of <requests>. + [-1] // run the one-to-n test. + [-g <granularity>] // choose the granularity of the timing of CORBA calls + [-c] // run the number of context switches test. + [-m] // use multiple priorities for the low priority clients. + +[-d <datatype>] <datatype> is one of the following: + + 0 Octet + 1 Short + 2 Long + 3 Struct of the above 3 types + + Default Value: 0 + +[-n <num_calls>] is the number of CORBA twoway calls to be made + + Default Value: 1000 + +[-t <num_of_clients>] is the total number of concurrent clients to be + created. The clients will have successively decreasing + priorities if "-m" is specified. + + Default Value: 2 + +[-f <ior_file> ] Use to specify a file to read the object references (iors) of the + different servants from disk. + + Default Value: nil, which means not to read the iors from disk. + +[-r] run the thread-per-rate test. (Priority inversion tests) + + Default Value: Disabled. + +[-o] enables oneway calls instead of two-way calls. The oneway call + is a noop call (for now). + + Default Value: Disabled + +[-x] makes the client call the shutdown() method on the servant which in turn + invokes the shutdown() method on the ORB and causes it to exit gracefully. + + Default Value: Disabled + +[-u <seconds>] runs the client/server utilization test for a number of <requests>. + The output is the number of seconds for the test to make such number of requests. + Now run the same test but with the -l option to make the same number of requests, + this time not using CORBA, instead directly making function calls. The difference + in time for each test to make the same number of requests is the utilization of + the OS/ORB combined. + (Note: This option overrides any specified "-t" option) + + Default Value: Disabled + +[-1] runs one-to-n test. One servant services all low priority clients. + + Default Value: Disabled + +[-g <granularity>] choose the granularity of timing the CORBA calls. + This option should disappear sometime in the future, together with + all the modifications done to support this. + + What happens when this options is not used, is that (1) we start a + timer, (2) invoke the CORBA call, (3) stop the timer after call + returns. Some platforms have a very coarse grain timer (e.g. 10 + msec) and timing individual calls does not report very accurate + latency results. + + This option enables a way to (1) start the timer, (2) invoke + <granularity> calls, and (3) stop the timer. This way, we get a + better sense of the average latency of the CORBA calls. + + The downside is that the mechanism to enforce the frequency of calls + (i.e., 20 Hz, 10Hz, etc) is in conflict with this option. Such + mechanism depends on the individual time for each call to enforce + the calling frequency. + + Default Value: 1 (time each CORBA call) + +[-c] Obtain number of context switches. + + Default Value: Disabled + +[-m] Use multiple priorities for the low priority client + + Default Value: Disabled. Will use one priority for all low priority clients. + +Running on VxWorks +------------------ + +A typical run of the server and client on a single VxWorks target: + + -> iam "<your login>" + -> ld < server + -> ld < client + -> spa server, "-f", "ior.txt", "-t", "2" + -> spa client, "-f", "ior.txt", "-t", "2", "-n", "1000" + +RESULTS +------- + +The client prints the latency observed by the high priority client, +and the average latency observed by the low priority clients. Jitter +(standard deviation of the latency) is also printed by the client. + + +Running MT_Cubit on CHORUS: +-------------------------- + +1.You should make sure that you set the HOST environment variable on the +Chorus VME Board to the IP address of that Board before you run +MT_Cubit test. + +e.g +$rsh tambora1 setenv HOST 128.252.165.82 + +2. You should make sure that the path to IOR file has write permission +for ALL. + +server: +------ +$rsh host_name arun path_to_server -e iiop://DOTTED_DECIMAL_ADDRESS:0 + -f path_to_ior_file -ORBDottedDecimalAddresses 1 <server_options> + +Since Chorus doesn't support DNS, you should pass +dotted_decimal_address of the server machine to -e option. + +You DONT have to use -ORBDottedDecimalAddresses 1 if you have +TAO_USE_DOTTED_DECIMAL_ADDRESSES defined in your config.h file. + +e.g +$rsh tambora1 arun /MT_Cubit/server -e iiop://128.252.165.82:0 -f /ior +-ORBDottedDecimalAddresses 1 -t 10 -ORBObjRefStyle url + +client: +------- + +$rsh host_name arun path_to_client -f path_to_ior_file <client_options> + +e.g + +$rsh tambora2 arun /MT_Cubit/client -f /ior -t 10 -n 300 |