bachelor-thesis

e99bf619c2 handle memory allocation outside of the cache, pre-allocate in benchmark and memset to hopefully guarantee no pagefaults will be encountered Constantin Fürst 2024-02-08 16:24:39 +0100
19ef2df856 update perf profile with manually disabled huge pages Constantin Fürst 2024-02-08 15:35:08 +0100
d4677b3c59 measure performance without huge pages on Constantin Fürst 2024-02-08 15:25:41 +0100
7afcffbefa set correct node in perf recording script Constantin Fürst 2024-02-08 15:24:31 +0100
87b71e6300 remove warings for weak access failure Constantin Fürst 2024-02-08 15:24:15 +0100
c86d517444 fix the published results for prefetching Constantin Fürst 2024-02-07 23:11:50 +0100
94b3576d5a publish measurements from benchmark Constantin Fürst 2024-02-07 23:07:13 +0100
8999fe4ca3 share current config for qdp bench from vampir Constantin Fürst 2024-02-07 23:05:37 +0100
79a7e9637c fix benchmark by waiting and not dropping barrier in aggrj Constantin Fürst 2024-02-07 23:05:19 +0100
01ecdce88f set execution engine to hardware Constantin Fürst 2024-02-07 22:37:15 +0100
bc1c3d0096 fix block size for access by cacher in scanb Constantin Fürst 2024-02-07 22:14:24 +0100
f4159d8149 fix old flag name from FLAG_TRIGGER_PAGES to FLAG_FORCE_MAP_PAGES Constantin Fürst 2024-02-07 22:11:41 +0100
99552b3de4 add option for forcing map of pages by touching each one with a write at its begin, required as somehow behaviour changed, cache was experiencing page fault errors and handling by dsa is simply too slow Constantin Fürst 2024-02-07 22:09:56 +0100
5044b4419c make load balancing thread-local to reduce atomic cost Constantin Fürst 2024-02-07 18:19:30 +0100
f9d47d3a45 add scanb back to the barrier, now other threads will wait for finish of work submission Constantin Fürst 2024-02-07 18:19:13 +0100
006b856c44 resolve issues from the recent reset of qdp benchmark Constantin Fürst 2024-02-07 17:48:15 +0100
de1de9134b reset benchmark Constantin Fürst 2024-02-07 17:42:48 +0100
c7b91e400f reset to using hardware path, sw was only temporarily used for debugging Constantin Fürst 2024-02-07 17:22:50 +0100
7abd83ca45 fix offloading cache bug, introduced by not using maxptr for secondary invalid value, more care needed to be taken on when the secondary invalid would be set and how, now we set in in cachedata:init for both the local task in access and the one in cache state Constantin Fürst 2024-02-07 17:20:30 +0100
4a587a36e2 remove overlap-execution barriers and run for the entire block Constantin Fürst 2024-02-07 16:03:36 +0100
73914d4cdd Merge branch 'master' of https://git.constantin-fuerst.com/constantin/bachelor-thesis Constantin Fürst 2024-02-07 15:54:51 +0100
8dfefe665e add warnings to the offloading cacher Constantin Fürst 2024-02-07 15:54:33 +0100
c393b8eb88 improve load balancing node assignment Constantin Fürst 2024-02-07 15:54:02 +0100
b9cfb818fd Merge branch 'master' of https://git.constantin-fuerst.com/constantin/bachelor-thesis Constantin Fürst 2024-02-07 15:35:08 +0100
21702d5309 remove sub and overchunking for scanb caching, use the per-iteration barriers again Constantin Fürst 2024-02-07 15:34:29 +0100
7d614769db remove forgotten access to load timer Constantin Fürst 2024-02-07 15:18:42 +0100
6a4eec37ca remove vector-load timing as its too expensive Constantin Fürst 2024-02-07 15:15:48 +0100
93a281fa26 improve debug output for relwithdebinfo in qdp, fix filename for record perf script, add perf.svg with better debug info Constantin Fürst 2024-02-07 15:04:13 +0100
624e8b55ea add script to record perf and make the flame graph Constantin Fürst 2024-02-07 15:01:38 +0100
942d7be7e9 redo benchmarks for qdp Constantin Fürst 2024-02-07 14:42:02 +0100
a83f208cd2 fix time evaluation for qdp bench Constantin Fürst 2024-02-07 14:31:52 +0100
cc35131df8 Merge branch 'master' of https://git.constantin-fuerst.com/constantin/bachelor-thesis Constantin Fürst 2024-02-07 05:52:38 +0100
cc8d203771 redo benchmarks for qdp, move previous results to old (folder) Constantin Fürst 2024-02-07 05:52:23 +0100
94669924c8 implement cache in aggrj for qdp Constantin Fürst 2024-02-07 05:38:04 +0100
b7559a18a6 remove cerr warning Constantin Fürst 2024-02-07 05:30:32 +0100
c7877ecdf6 remove skeleton of now defunct function in qdp Constantin Fürst 2024-02-07 05:27:41 +0100
20c6e54df7 remove broken implementation for non-divisible chunk-group-thread-counts Constantin Fürst 2024-02-07 05:26:41 +0100
a3a8dff1aa reset some changes to the aggregation and filter functions not quite needed Constantin Fürst 2024-02-07 05:19:49 +0100
69aec6fa48 add plotter for the results of qdp which turns them into a donut-graph Constantin Fürst 2024-02-07 04:26:16 +0100
122eab35b7 modify benchmarking code to measure time spent loading vectors too Constantin Fürst 2024-02-07 04:25:44 +0100
3fc9bc6145 finish the section on using cache with qdp, also add a section on possible usage of dwcas for aba problem solution Constantin Fürst 2024-02-07 00:54:23 +0100
98835799d4 also copy the invalid handlers pointer (see last commit) on CacheData copy constructor Constantin Fürst 2024-02-06 23:36:55 +0100
75d72ce556 dont use uint64::max as invalid pointer but allocate a valid but empty vector, using its address as invalid Constantin Fürst 2024-02-06 22:50:31 +0100
43254d0f3c note that the cost observed for swq submission is lower than what reese kuper saw Constantin Fürst 2024-02-06 21:17:03 +0100
8dcc248620 first pass of modifications to abstract and intro Constantin Fürst 2024-02-06 21:16:19 +0100
b55196a7a7 redo reference types, using misc with howpublished for content that is not scientific work Constantin Fürst 2024-02-06 21:15:11 +0100
94ee998c2e finalize draft of abstract and intro Constantin Fürst 2024-02-06 16:21:50 +0100
9bb68a7244 remove unnecessary cite by andres advice Constantin Fürst 2024-02-06 16:21:17 +0100
8b635f7e87 update bachelor.pdf with recent changes Constantin Fürst 2024-02-05 16:32:14 +0100
58f36279ec reformulate implementation chapter - our chocie has changed and we now use push-pull and not smart-copy anymore for load balancer Constantin Fürst 2024-02-05 16:32:01 +0100
eabba98972 include the new benchmarks for peak throughput and work division and modify their analysis Constantin Fürst 2024-02-05 16:31:23 +0100
8bdc3e1d76 start formulation of thesis introduction Constantin Fürst 2024-02-05 16:30:46 +0100
9d5fbe085b formulate abstract Constantin Fürst 2024-02-05 16:30:09 +0100
579494bc41 redo the plots for benchmarks of peak throughput and work division Constantin Fürst 2024-02-05 16:29:37 +0100
db8751afc7 run the modified peak throughput benchmarks Constantin Fürst 2024-02-05 14:00:09 +0100
1b6c60c49b benchmark copy throughput for using 1,2,4,8 dsas and remove brute cpu bench (we steal it from andre) Constantin Fürst 2024-02-05 13:46:54 +0100
b6f85ca202 redo benchmarks for pushpull Constantin Fürst 2024-02-05 13:23:31 +0100
21bbf53e55 use local and one remote node for pushpull on intranode Constantin Fürst 2024-02-05 13:22:29 +0100
c43bce3e13 add results for pushpull benchmark Constantin Fürst 2024-02-05 13:18:29 +0100
850aebc6b9 update bachelor.pdf with the bullet points for abstract and intro Constantin Fürst 2024-02-05 02:37:37 +0100
6b500c3396 cite figure source at end of caption and not mid-sentence for chapter2 Constantin Fürst 2024-02-05 02:31:40 +0100
25f55adcb2 formulate bullet points for abstract and introduction to the thesis Constantin Fürst 2024-02-05 02:31:11 +0100
18ec95e201 adapt chapter on performance to the corrected measurements Constantin Fürst 2024-02-05 01:06:59 +0100
23daabbd73 fix bad section references Constantin Fürst 2024-02-05 01:06:16 +0100
321a4fb0bd manually add data from communication with andre to the cpu benchmark plotters Constantin Fürst 2024-02-05 00:47:40 +0100
1021197009 do not use the float-force parameter H but move to a more dynmic one (h!tb), also dont use Subsection anywhere instead use Section, add a small paragraph to the implementation chapter stating how we used the cache in qdp Constantin Fürst 2024-02-05 00:45:03 +0100
27c57aa4ce add node for content of evaluation chapter Constantin Fürst 2024-02-05 00:42:05 +0100
9e329d39e4 add pushpul benchmark for peak throughput Constantin Fürst 2024-02-05 00:00:38 +0100
33730c4a99 add all images which for now were committed only with -f Constantin Fürst 2024-02-04 16:46:13 +0100
a580939d29 rewrite chapter 3 with the corrected benchmark results, add these corrected results, re-add structograms with different names Constantin Fürst 2024-02-04 16:45:29 +0100
89d2a6c71f redo the structograms for the benchmark code Constantin Fürst 2024-02-04 16:43:03 +0100
94818536cd improve the section on qdp in state chapter and use (redone) graphic for simple query Constantin Fürst 2024-02-04 16:42:16 +0100
4f27e9c9c0 remove pdf files from gitignore Constantin Fürst 2024-02-04 16:41:35 +0100
7572350b28 apply recommendations of andre to chapter 5 Constantin Fürst 2024-02-04 14:05:43 +0100
c6495b8b02 write introductory paragraph to design section Constantin Fürst 2024-02-04 14:05:24 +0100
d1cc3e3b0c modification to qdp benchmark, returns to per-chunk barrier wait, uses userspace semaphore for one-way barrier from scan_b to aggr_j as scan_b should submit asap but aggr_j should wait on submission from scan_b, contains TODO for modifying code to support chunkcount not divisible by 2 Constantin Fürst 2024-01-31 23:59:59 +0100
bf79435ff0 add new benchmark plots from the rewritten microbench Constantin Fürst 2024-01-31 23:03:16 +0100
8ab5eb4902 finish adapting plotters to new result style, add division by thread count to the throughput plotters, adjust figure sizes to be small (larger font when scaled up in latex) Constantin Fürst 2024-01-31 23:02:07 +0100
f3e89405a5 publish benchmark results from vampir for the redone microbench Constantin Fürst 2024-01-31 22:26:22 +0100
067a31e560 modify submission benchmark descriptors to have x10 internal repetitions Constantin Fürst 2024-01-31 21:59:23 +0100
326cf92af3 update benchmark plotters for changes made to benchmark and result format Constantin Fürst 2024-01-31 21:54:21 +0100
875098b258 remove plotters which are not in use anymore Constantin Fürst 2024-01-31 21:54:01 +0100
ed77e57e5f remove benchmark descriptors for unused engine location benchmark Constantin Fürst 2024-01-31 21:53:02 +0100
a216d96003 remove old benchmark plots Constantin Fürst 2024-01-31 21:52:33 +0100
8ac601fc07 add option for internal repetitions to benchmarks which allows the small copies of 1kib to run long enough for the timings to become usable (goal is about 1s runtime for each iteration) Constantin Fürst 2024-01-31 18:02:02 +0100
a963406f7c move mode selection to Configuration.hpp, adapt the CopyMethodPolicy-Function to return only src_node for task sizes under 16MiB which is now required to not cause high submission count which slows down small copies Constantin Fürst 2024-01-30 15:04:53 +0100
ef8286da17 unset the weak-wait-flag on deallocation for CacheData which is where completion guarantee is required, also extend some comments in CacheData Constantin Fürst 2024-01-30 15:03:38 +0100
8dc3827676 do not check in the debug benchmark result Constantin Fürst 2024-01-30 14:48:08 +0100
25451fa26a pretty-format the benchmarker script which got mangled from editing on vampir Constantin Fürst 2024-01-30 14:47:33 +0100
3bfbeca21f resize source and destination pointer holders properly before use and use path from template and not dml::software for cache flush in benchmark loop Constantin Fürst 2024-01-30 14:46:50 +0100
ef805244ac use 4gib as size and again 1 aggrj thread for qdp bench Constantin Fürst 2024-01-30 14:45:51 +0100
3c90e24bc1 revert the cacher to allow load balancing control through the copy placement policy function which now selects on how many nodes the task is split again, and not just which nodes the task MAY run on (which was done experimentally) Constantin Fürst 2024-01-30 14:14:59 +0100
ea978423c0 slightly modify the debug benchmark descriptor to contain multiple threads and also a batch Constantin Fürst 2024-01-30 13:59:55 +0100
cae0fbb56e remove benchmark descriptor handling helper scripts Constantin Fürst 2024-01-30 13:59:15 +0100
24bdccd1e3 rewrite the benchmarker to not allocate the memory regions each iteration but before the test runs, also flush cache each iteration using dml-operation, also set dsa-device using the parameter to submit and not using libnuma assignment Constantin Fürst 2024-01-30 13:52:51 +0100
e2c6fd8587 set iteration count for submission bench to 10 for the large and 100 for the small, also test 128mib which is over the size of the cache available on xeonmax (instead of 32mib) Constantin Fürst 2024-01-30 13:50:25 +0100
4f9abc911f make benchmark.hpp a cpp file to make it clear that it will have global variables Constantin Fürst 2024-01-30 13:06:32 +0100
f905ee77eb wait less for task launch and dont write iterations complete out Constantin Fürst 2024-01-29 14:29:37 +0100
72fb3764fc modify benchmarker script to require less parameters and sit in root dir Constantin Fürst 2024-01-29 14:28:43 +0100
c8b4f3d624 fix issues with benchmark.hpp Constantin Fürst 2024-01-29 14:17:27 +0100

Commit Graph Select branches Hide Pull Requests master Mono Color

Commit Graph

Select branches

Hide Pull Requests

master