UPC Operations Microbenchmarking Suite 1.0 User`s manual
Contents
1. CESGA Alliance UPC Operations Microbenchmarking Suite 1 0 User s manual Authors PhD Guillermo L pez Taboada Dami n Alvarez Mall n 1 taboada udc es 2 dalvarez cesga es FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA Contents 1 Contact 2 Files in this benchmarking suite 3 Operations tested 4 Customizable parameters 4 1 Compile time 00 ll ee AQ Rut INO si 2 aid arba a ee bw A a aki bl A San ab AL s l d l j k sa 5 Compilation 6 Timers used 7 Output explanation UOMS User s Manual April 2010 1 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA 1 Contact Vou can contact us at Galicia Supercomputing Center CESGA http www cesga es Santiago de Compostela Spain upc Acesga es PhD Guillermo Lopez Taboada Computer Architecture Group CAG http gac des udc es index en html University of A Corufia Spain taboada udc es 2 Files in this benchmarking suite e doc manual pdf This file User s manual e COPYING and COPYING LESSER Files containing the use and redistribution terms license e changelog txt File with changes in each release e src affinity upc UPC code with affinity related tests e src config make def template Makefile templates for HP UPC and Berkeley UPC e src config parameters h Header with some customizable parameters e src defines h Header with needed definitions e src headers h Header with HUCB functions headers e src mem_mana2. Defines the UPC compiler used to compile the suite e UPCFLAGS Defines the UPC compiler flags used to compile the suite Please note you should not specify any number of threads flag at this point e UPCLINK Defines the UPC linker used to link the suite e UPCLINKFLAGS Defines the UPC linker flags used to link the suite e THREADS SWITCH Defines the correct switch to set the desired number of threads It is com piler dependent and also includes any blank space after the switch Once you have set up your make def file you can compile the suite as following make NTHREADS NUMBER_OF_UPC_THREADS E g for 128 threads make NTHREADS 128 6 Timers used This suite uses high resolution timers in IA64 architecture In particular it uses the Interval Timer Counter AR ITC For other architectures it uses the hpupc ticks now if you are using HP UPC or bupc_ticks_now if you are using Berkeley UPC whose precision depends on the specific archi tecture If none of this requirements are met the suite uses the default gettimeofday function However the granularity of this function only allows to measure microseconds rather than nanosec onds UOMS User s Manual April 2010 7 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA 7 Output explanation This is an output example of the broadcast Benchmarking upc all broadcast processes 2 1 bytes repetitions t_min nsec t_max nsec t_avg nsec BW_aggregated MB sec 4
3. e upc_memget_asynci remote e upc_memput_asynci remote i e upc_memcpy_asynci local e upc memget asvnci local e upc_memput_asynci local e upc all alloc e upc free In bulk memorv transfer operations there are two modes remote and local Remote mode will copv data from one thread to another whereas local mode will copv data from one thread to another memorv region with affinitv to the same thread 4 Customizable parameters 4 1 Compile time In the src config parameters h file you can customize some parameters at compile time They are e NUMCORES If defined it will override the detection of the number of cores If not defined the number of cores is set through the sysconf _SC_NPROCESSORS_ONLN system call e ASYNC_MEM_TEST If defined asynchronous memory transfer tests will be built Default is defined e MINSIZE The minimum message size to be used in the benchmarking Default is 4 bytes e MAXSIZE The maximum message size to be used in the benchmarking Default is 16 megabytes UOMS User s Manual April 2010 4 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA 4 2 Run time The following flags can be used at run time in the command line e help Print usage information and exits e version Print UOMS version and exits e off cache Enable cache invalidation Be aware that the cache invalidation greatly increases the memory consumption Also note that for block sizes smaller t
4. 20 19942 48820275 2463315 85 0 00 8 20 19942 22922 21457 25 0 70 16 20 19942 22397 21420 10 1 43 32 20 19942 22235 21626 35 2 88 64 20 20277 33610 22886 00 3 81 128 20 20285 22812 21676 60 11 22 256 20 20767 22845 22230 50 22 41 512 20 20767 23020 22314 85 44 48 1024 20 22777 29255 24169 85 70 01 2048 20 23705 25425 24603 85 161 10 4096 20 24562 27097 26437 60 302 32 8192 20 29885 33205 32174 35 493 42 16384 20 42492 44735 43919 35 732 49 32768 10 68317 70052 69490 00 935 53 65536 10 121610 123837 122635 00 1058 42 131072 10 227550 231515 229323 50 1132 30 262144 10 437645 444740 441354 00 1178 86 524288 10 861287 871700 867619 70 1202 91 1048576 5 1702722 1704420 1703642 40 1230 42 2097152 5 3417170 3435637 3429128 40 1220 82 4194304 5 6830267 6839535 6834224 40 1226 49 8388608 2 13434382 13469047 13451715 00 1245 61 16777216 2 27310152 27343357 27326755 00 1227 15 33554432 1 54294385 54294385 54294385 00 1236 02 The header indicates the benchmarked function and the number of processes involved The first column shows the size used for each particular row It is the size of the data at the root thread or in any thread in a non rooted operation The second column is the number of repetitions performed for that particular message size The following three columns are respectively the minimum maximum and average latencies The last column shows the aggregated bandwidth calculated using the maximum latencies Therefore the bandwidth report
5. ed is the minimum bandwidth achieved in all the repetitions Moreover when 2 threads are used affinity tests are performed This way you can measure the effects of data locality in NUMA systems if the 2 threads run in the same machine This feature may be useful even when the 2 threads run in different machines E g Machines with non uniform access to the network interface like quad socket Opteron Nehalem based machines or cell based machines like HP Integrity servers The output of this tests is preceded with something like using cores 0 and 1 Number of cores per node 16 CPU Mask 1000000000000000 core 0 0100000000000000 core 1 UOMS User s Manual April 2010 8 FUNDACI N UNIVERSIDADE DA CORUNA ef L I CES A Ca All tests after these lines are performed using core 0 thread 0 and core 1 thread 1 until another affinitv header is showed UOMS User s Manual April 2010 9
6. ger upc Memory related functions for allocation and freeing e src UOMS upc Main file It contains the actual benchmarking code e src init upc Code to initialize some structures and variables e src Makefile Makefile to build the benchmarking suite e src timers timers c Timing functions e src timers timers h Timing functions headers e src utils data_print upc Functions to output the results e src utils utilities c Auxiliarv functions UOMS User s Manual April 2010 2 az FUNDACI N UNIVERSIDADE DA CORUNA e NA csa GY 3 Operations tested upc barrier upc all broadcast upc all scatter upc all gather upc all gather all upc all permute upc all exchange upc all reduceC upc all prefix reduceC upc all reduceUC upc all prefix reduceUC upc_all_reduces upc_all_prefix_reduces upc_all_reduceUS upc all prefix reduceUS upc all reducel upc all prefix reducel upc_all_reduceUI upc_all_prefix_reduceUI upc_all_reduceL upc_all_prefix_reduceL upc all reduceUL upc all prefix reduceUL upc all reduceF upc all prefix reduceF upc all reduceD upc all prefix reduceD upc_all_reduceLD UOMS User s Manual April 2010 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA e upc all prefix reducelD e upc memcpv remote e upc_memget remote e upc_memput remote e upc_memcpy local e upc_memget local e upc_memput local e memcpy local e memmove local e upc_memcpy_asynci remote
7. han the cache line size it will not work e warmup Enable a warmup iteration e reduce_op OP Choose the reduce operation to be performed by upc all reduceD and upc_all_prefix_reduceD Valid operations are UPC_ADD default UPC_MULT UPC_LOGAND UPC_LOGOR UPC_AND UPC_OR UPC_XOR UPC_MIN UPC_MAX e sync mode MODE Choose the synchronization mode for the collective operations Valid modes are UPC_IN_ALLSYNC UPC_OUT_ALLSYNC default UPC_IN_ALLSYNC UPC_OUT_MYSYNC UPC_IN_ALLSYNC UPC_OUT_NOSYNC UPC_IN_MYSYNC UPC_OUT_ALLSYNC UPC_IN_MYSYNC UPC_OUT_MYSYNC UPC_IN_MYSYNC UPC_OUT_NOSYNC UPC_IN_NOSYNC UPC_OUT_ALLSYNC UPC_IN_NOSYNC UPC_OUT_MYSYNC UPC_IN_NOSYNC UPC_OUT_NOSYNC e msglen FILE Read user defined problem sizes from FILE in bytes If specified it will override minsize and maxsize e minsize SIZE Specifies the minimum block size in bytes Sizes will increase by a factor of 2 e maxsize SIZE Specifies the maximum block size in bytes UOMS User s Manual April 2010 5 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA e input FILE Read user defined list of benchmarks to run from FILE Valid benchmark names are upc barrier upc all broadcast upc all scatter upc all gather upc all gather all upc all exchange upc all permute upc_memget upc_memput upc memcpy local upc
8. memget local upc memput local upc memcpy memcpy memmove upc all alloc upc free upc all reduceC upc all prefix reduceC upc all reduceUC upc all prefix reduceUC upc all reduces upc all prefix reduces upc all reduceUS upc all prefix reduceUS upc all reducel upc all prefix reducel upc_all_reduceUI upc all prefix reduceUl upc all reducel upc all prefix reduce upc all reduceUL upc all prefix reduceUL upc all reduceF upc all prefix reduceF UOMS User s Manual April 2010 6 FUNDACI N UNIVERSIDADEDA CORUNA 5 L 2 CESGA CA upc all reduceD upc all prefix reduceD upc all reducelD upc all prefix reducelLD upc_memget_asynci upc_memput_asynci upc_memcpy_asynci local upc memget asvnci local upc memput asvnci local upc memcpv asvnci 5 Compilation To compile the suite you have to setup a correct src config make def file Templates are provided to this purpose The needed parameters are e CC Defines the C compiler used to compile the C code Please note this does not involve the resulting C code generated from the UPC code if your UPC compiler is a source to source compiler e CFLAGS Defines the C flags used to compile the C code Please note this does not involve the resulting C code generated from the UPC code if your UPC compiler is a source to source compiler e UPCC
Download Pdf Manuals
Related Search