bachelor-thesis

52a026805e use different notation for glossary which results in easier reading and correct result Constantin Fürst 2024-01-07 14:40:48 +0100
c0c75aa51b use uniform colormap in the plots and use separate output folder for the plots Constantin Fürst 2024-01-07 14:40:13 +0100
5f72404508 re run changed smart peak throughput benchmarks Constantin Fürst 2023-12-13 13:53:41 +0100
1fca956a0a update the peak throughput plotter to show the difference of smart and allnodes too Constantin Fürst 2023-12-13 13:44:24 +0100
791184ff10 fix mistake in benchmark descriptors for smart peak performance Constantin Fürst 2023-12-13 13:42:46 +0100
f809eb5847 rerun copy benchmark with smart assignment Constantin Fürst 2023-12-13 13:17:17 +0100
6b9581b8d1 add results for the new 512mib internode hbm copy peak perf test Constantin Fürst 2023-12-13 12:49:27 +0100
0748826fcd use transfer size of 512mib for HBM intranode copy and modify plotter accordingly Constantin Fürst 2023-12-13 12:45:34 +0100
60a5ba5120 refactor the benchmark plotters and submit newly plotted graphs Constantin Fürst 2023-12-12 17:14:42 +0100
c0f2aa2b64 re-plot the benchmarks with the new data Constantin Fürst 2023-12-12 16:22:11 +0100
1c7369f20e add results for smart and allnodes peak throughput benchmark Constantin Fürst 2023-12-12 16:01:18 +0100
1e55565072 remove previous benchmark results for peak performance Constantin Fürst 2023-12-12 15:13:49 +0100
475b2d5b5a provide two benchmarks for peak performance, one that is brute force and one that uses smart node assignment and therefore lower utilization Constantin Fürst 2023-12-12 15:13:27 +0100
9cef69c33f add mdsa v3 benchmark results for peak performance Constantin Fürst 2023-12-12 15:03:09 +0100
7ced0bce4c fix an issue in the python script that lead to references being modified which caused bad node settings Constantin Fürst 2023-12-12 14:52:39 +0100
8787b441bc add second type of multi dsa benchmark results Constantin Fürst 2023-12-12 14:28:25 +0100
6d9002d1e7 use different engine configuration depending on whether intra socket (all 4 engines on the socket) or inter socket (src and destination engine, cross copy) is the copy type Constantin Fürst 2023-12-12 14:15:20 +0100
68a838f0d1 add final multi-dsa results Constantin Fürst 2023-12-12 14:01:09 +0100
c92bb28d9a add intermediate multi-dsa results Constantin Fürst 2023-12-12 13:41:14 +0100
f11bb710ae add intermediate multi-dsa results Constantin Fürst 2023-12-12 13:37:22 +0100
7f7230197c dont submit multiple copies - test takes too long and this has almost no effect at work size of 1gib Constantin Fürst 2023-12-12 13:32:21 +0100
3e102509a9 add intermediate multi-dsa results Constantin Fürst 2023-12-12 13:29:01 +0100
584c5bdfc4 add intermediate multi-dsa results Constantin Fürst 2023-12-12 13:21:23 +0100
e9807df09c use all 8 engines for each copy task, as the engine location does not affect performance Constantin Fürst 2023-12-12 13:17:06 +0100
5089936f30 prepare peak throughput plotter for multi-node results Constantin Fürst 2023-12-12 13:03:46 +0100
db11eb60e6 add 4e results from copy peak perf bench Constantin Fürst 2023-12-12 12:22:47 +0100
17264186a6 small changes to the plotter scripts for nicer display Constantin Fürst 2023-12-12 11:57:19 +0100
1682b84fb4 use a batch size of 8 to check whether multiple engines can increase throughput Constantin Fürst 2023-12-12 11:56:30 +0100
4598cedd40 re-run peak perf test Constantin Fürst 2023-12-11 20:14:42 +0100
575ff8cf82 turn node -1 into node 7 again - this got messed up by a hastily written modification script Constantin Fürst 2023-12-11 20:05:08 +0100
eb4ea5162d fix bugs that were introduced by changes to the plotter scripts Constantin Fürst 2023-12-11 19:57:37 +0100
cf675e37f5 dont pin the thread to hbm nodes but to the hbm src node minus 8 in peak perf benchmark Constantin Fürst 2023-12-11 19:57:06 +0100
d3e8fec087 re-run engine location bench Constantin Fürst 2023-12-11 19:28:13 +0100
5c3b008620 add results for peak, mt and submit benchmarks Constantin Fürst 2023-12-11 19:27:08 +0100
788b2f25d3 add new results for engine location benchmark Constantin Fürst 2023-12-11 18:58:21 +0100
808c8f3ae7 run the engine location benchmarks with size of 1gib only 10 times Constantin Fürst 2023-12-11 18:56:09 +0100
806f5f4f97 remove old benchmark results Constantin Fürst 2023-12-11 18:45:43 +0100
099f454f19 modify plotters to a more streamlined state, all now use the file-loop in main and have a function that processes one file into the dataset, also adds the peakthroughput plotter and removes the defunct opt-submitmethod plotter Constantin Fürst 2023-12-11 18:45:05 +0100
b37968dd3f rename engine location benchmarks, modify plotter to support missing configuration files as the new cases are not universally applicable to all configurations Constantin Fürst 2023-12-11 14:32:04 +0100
6cde7288e9 use total time in submitmethod benchmark too Constantin Fürst 2023-12-11 14:06:59 +0100
a548d9afe5 restructure mtsubmit benchmarks Constantin Fürst 2023-12-11 14:06:36 +0100
b7cae18b6d restructure the engine location bench, correct and update the plotter to use new total time Constantin Fürst 2023-12-11 14:05:19 +0100
405166cbe8 add peak perf benchmark descriptors Constantin Fürst 2023-12-11 13:34:17 +0100
148c4c213a re-run mtsubmit tests Constantin Fürst 2023-12-06 14:15:00 +0100
fb9164ae89 add script to disable dsa and then load a new config Constantin Fürst 2023-12-06 14:14:18 +0100
9886b20112 remove the multiple tests for mtsubmit and ensure that the wq will always receive the same elements to make the test fair Constantin Fürst 2023-12-06 14:11:06 +0100
01850cf97b add new mtsubmit test results Constantin Fürst 2023-12-06 13:54:03 +0100
3964da0d7a use ms10 instead of ms50 as ms50 with > 2 threads will overfill the wq of size 128 (max) and cause an error unhandled in the benchmark yet Constantin Fürst 2023-12-06 13:51:56 +0100
f9b00a5b32 modify mtsubmit to meassure both ssaw and ms50 submit methods Constantin Fürst 2023-12-06 13:39:42 +0100
846fa9be43 re-run mtsubmit benchmarks for 4e Constantin Fürst 2023-12-06 13:28:18 +0100
6f620d6461 re-run mtsubmit benchmarks for 1e Constantin Fürst 2023-12-06 13:26:56 +0100
59853ffc92 remove the mtsubmit performance plotter which was based on a false assumption, modify the mtsubmit cost plotter for the new task set size Constantin Fürst 2023-12-06 13:20:22 +0100
d72e28adb8 make mtsubmit a multisubmit Constantin Fürst 2023-12-06 13:18:28 +0100
8f68f41ba0 position legend for optimal submit method in bottom right to make the entire plot readable Constantin Fürst 2023-12-05 23:38:44 +0100
8ba84587be add plotter for submit method with logscale to show optimal Constantin Fürst 2023-12-05 23:33:21 +0100
4d4cc389c6 recreate the plots for mtsubmit with the new benchmark results Constantin Fürst 2023-12-05 23:18:06 +0100
d3020902f3 merge the new results from rerunning mtsubmit benchmarks Constantin Fürst 2023-12-05 23:16:44 +0100
10b1f9018c add plotted results from the new scripts Constantin Fürst 2023-12-05 23:16:41 +0100
c27514890e finish the move to save entire results and not condensed average in the plotter scripts Constantin Fürst 2023-12-05 23:15:34 +0100
9e28ca5a27 re-run mtsubmit benchmarks Constantin Fürst 2023-12-05 23:14:15 +0100
ee9119ea46 remove the 1gib benchmark results Constantin Fürst 2023-12-05 22:22:03 +0100
573e312a69 update the submit method benchmark plotter Constantin Fürst 2023-12-05 22:21:48 +0100
9c5b573beb re-run benchmarks with current configuration Constantin Fürst 2023-12-05 20:04:29 +0100
21a81c5aa0 do not pretty print the json files produced to save space Constantin Fürst 2023-12-05 19:52:54 +0100
151ddd79d5 set repetition count of tests to 1000 again and replace the 1 GiB tests with 32 MiB Constantin Fürst 2023-12-05 19:52:29 +0100
b7133346d8 add benchmark results Constantin Fürst 2023-12-05 14:33:57 +0100
b44b52b600 meassure nanoseconds instead of microseconds Constantin Fürst 2023-12-05 14:26:54 +0100
21194b7fde fix the benchmarker script by testing for the correct amount of parameters and obtaining the test name correctly Constantin Fürst 2023-12-05 14:25:36 +0100
e386370144 add dsa configuration files to repository Constantin Fürst 2023-12-05 14:19:50 +0100
d3a09181ce move the existing results into a new folder and modify them to be only benchmark descriptors Constantin Fürst 2023-12-05 14:14:46 +0100
167370dbf9 prepare submitmethod benchmark plotter for the new result type Constantin Fürst 2023-12-05 13:47:04 +0100
fb3cabb6e6 change naming of the json entries for timing, these are now not averages anymore but vectors Constantin Fürst 2023-12-05 13:46:30 +0100
e6df656845 use vector for timing - this update got lost somehow Constantin Fürst 2023-12-04 13:12:36 +0100
5b889b87e1 fix include naming Constantin Fürst 2023-12-04 13:12:13 +0100
d0631f18e0 write script to re-run existing benchmarks in a given folder Constantin Fürst 2023-12-04 13:01:29 +0100
bc8c4f8ab3 restructure of directory layout Constantin Fürst 2023-12-04 13:01:09 +0100
8065dd4345 collect entire duration vector and dont condense the information down Constantin Fürst 2023-12-04 11:48:29 +0100
d77230dece use throughput as unit for benchmarks Constantin Fürst 2023-12-04 10:30:54 +0100
b1213c1673 regenerate the mtsubmit plot with the new data Constantin Fürst 2023-12-04 09:47:30 +0100
1ba3669aad redo the benchmarks for mtsubmit that seem out-of-line Constantin Fürst 2023-12-04 09:43:47 +0100
1405ef38fe properly set the titles of the graphs and add parameter so they are not cut off Constantin Fürst 2023-12-03 14:34:56 +0100
e129638a87 add benchmark plotting scripts using seaborn to display the results, the plots are also added Constantin Fürst 2023-12-03 14:27:51 +0100
7c751aa2ce format benchmark-plan.md so that it is legible in rendered version too Constantin Fürst 2023-11-29 23:19:18 +0100
134076281c Merge branch 'dev/benchmarks' of https://git.constantin-fuerst.com/constantin/bachelor-thesis into dev/benchmarks Constantin Fürst 2023-11-29 23:15:07 +0100
9fa15d9045 add benchmark-findings.md which is a scratchpad for things that stand out to me with the benchmark results Constantin Fürst 2023-11-29 23:14:38 +0100
8705cd55ff add benchmark results for submit, mtsubmit and crosscopy Constantin Fürst 2023-11-29 23:04:16 +0100
151cadf3c7 formulate the benchmark descriptions more concisely, in benchmark-plan.md Constantin Fürst 2023-11-29 22:57:48 +0100
854cd6916c refrain from using a benchmarking script as this is too inflexible and a huge amount of work to create, manual mode it is ..., also create more concrete plan for benchmarks and the conclusions to draw from them in bechmark-plan.md (new) Constantin Fürst 2023-11-29 22:37:47 +0100
071ed9be18 add missing read of iteration count to json task description reader Constantin Fürst 2023-11-28 14:10:28 +0100
8c5a061343 remove buggy option for multiple sizes Constantin Fürst 2023-11-28 14:07:56 +0100
897e363f44 fix forgotten add to vector in AddToTaskVector of ReadTaskData Constantin Fürst 2023-11-28 13:33:47 +0100
1bfb1f316c ignore first five runs to reduce influence of warmup, add non-batch-descriptor batch loop for testing, calculate standard deviation for all three messurements Constantin Fürst 2023-11-28 13:23:50 +0100
9265ce6f65 add empty benchmarker-script which will orchestrate all benchmarks and print the results in the future, add comments about planned tests to this script and remove the benchmark notes markdown file which is now obsolete Constantin Fürst 2023-11-27 14:26:15 +0100
5fa12feb7d add required explicit block_on_fault-option to the task submission so that the DSA can handle page faults Constantin Fürst 2023-11-27 14:02:49 +0100
4e9688224b create a custom barrier structure that allows synchronization of each iteration of the meassurement loop Constantin Fürst 2023-11-27 13:29:38 +0100
80d1b5f543 remove fractional average calculation in favour of std::accumulate and remove option to set thread affinity to core while keeping support for node affinity assignment Constantin Fürst 2023-11-27 11:36:36 +0100
42f892e70b restructure CMakeLists.txt slightly Constantin Fürst 2023-11-27 11:26:46 +0100
659883a765 check the status of the batch operation too - forgot this Constantin Fürst 2023-11-26 18:50:33 +0100
da1da2062a remove brackets around status code string representation Constantin Fürst 2023-11-26 18:50:10 +0100
22f3ed8956 clean up naming of structs, functions and files Constantin Fürst 2023-11-26 18:41:51 +0100

Commit Graph Select branches Hide Pull Requests master Mono Color

Commit Graph

Select branches

Hide Pull Requests

master