Changeset 50ff1d0
- Timestamp:
- Aug 30, 2022, 3:18:14 PM (2 years ago)
- Branches:
- ADT, ast-experimental, master, pthread-emulation
- Children:
- 2c1b7de
- Parents:
- c702d21
- File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
doc/theses/thierry_delisle_PhD/thesis/text/eval_micro.tex
rc702d21 r50ff1d0 132 132 \caption[Cycle Benchmark on Intel]{Cycle Benchmark on Intel\smallskip\newline Throughput and scalability as a function of \proc count, 5 \ats per cycle, and different cycle count. 133 133 For throughput, higher is better, for scalability, lower is better. 134 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}134 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 135 135 \label{fig:cycle:jax} 136 136 \end{figure} … … 164 164 \caption[Cycle Benchmark on AMD]{Cycle Benchmark on AMD\smallskip\newline Throughput and scalability as a function of \proc count, 5 \ats per cycle, and different cycle counts. 165 165 For throughput, higher is better, for scalability, lower is better. 166 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}166 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 167 167 \label{fig:cycle:nasus} 168 168 \end{figure} … … 170 170 \subsection{Results} 171 171 172 Figure ~\ref{fig:cycle:jax} and \ref{fig:cycle:nasus} show the results for the cycle experiment.173 Looking at the left column on Intel first, Figure ~\ref{fig:cycle:jax:ops} and \ref{fig:cycle:jax:ns}, which shows the results for many \ats, in this case 100 cycles of 5 \ats for each \proc.172 Figures~\ref{fig:cycle:jax} and \ref{fig:cycle:nasus} show the results for the cycle experiment. 173 Looking at the left column on Intel first, Figures~\ref{fig:cycle:jax:ops} and \ref{fig:cycle:jax:ns}, which shows the results for many \ats, in this case 100 cycles of 5 \ats for each \proc. 174 174 \CFA, Go and Tokio all obtain effectively the same throughput performance. 175 175 Libfibre is slightly behind in this case but still scales decently. … … 178 178 As expected, this pattern repeats again between \proc count 72 and 96. 179 179 180 Looking next at the right column on Intel, Figure ~\ref{fig:cycle:jax:low:ops} and \ref{fig:cycle:jax:low:ns}, which shows the results for few threads, in this case 1 cycle of 5 \ats for each \proc.180 Looking next at the right column on Intel, Figures~\ref{fig:cycle:jax:low:ops} and \ref{fig:cycle:jax:low:ns}, which shows the results for few threads, in this case 1 cycle of 5 \ats for each \proc. 181 181 \CFA and Tokio obtain very similar results overall, but Tokio shows more variations in the results. 182 182 Go achieves slightly better performance than \CFA and Tokio, but all three display significantly workst performance compared to the left column. … … 188 188 Looking now at the results for the AMD architecture, Figure~\ref{fig:cycle:nasus}, the results show a story that is overall similar to the results on the Intel, with close to double the performance overall but with slightly increased variation and some differences in the details. 189 189 Note that the maximum of the Y-axis on Intel and AMD differ significantly. 190 Looking at the left column first, Figure ~\ref{fig:cycle:nasus:ops} and \ref{fig:cycle:nasus:ns}, unlike Intel, on AMD all 4 runtimes achieve very similar throughput and scalability.190 Looking at the left column first, Figures~\ref{fig:cycle:nasus:ops} and \ref{fig:cycle:nasus:ns}, unlike Intel, on AMD all 4 runtimes achieve very similar throughput and scalability. 191 191 However, as the number of \procs grows higher, the results on AMD show notably more variability than on Intel. 192 192 The different performance improvements and plateaus are due to cache topology and appear at the expected \proc counts of 64, 128 and 192, for the same reasons as on Intel. 193 Looking next at the right column, Figure ~\ref{fig:cycle:nasus:low:ops} and \ref{fig:cycle:nasus:low:ns}, Tokio and Go have the same throughput performance, while \CFA is slightly slower.193 Looking next at the right column, Figures~\ref{fig:cycle:nasus:low:ops} and \ref{fig:cycle:nasus:low:ns}, Tokio and Go have the same throughput performance, while \CFA is slightly slower. 194 194 This is different than on Intel, where Tokio behaved like \CFA rather than behaving like Go. 195 195 Again, the same performance increase for libfibre is visible when running fewer \ats. … … 253 253 \caption[Yield Benchmark on Intel]{Yield Benchmark on Intel\smallskip\newline Throughput and scalability as a function of \proc count. 254 254 For throughput, higher is better, for scalability, lower is better. 255 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}255 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 256 256 \label{fig:yield:jax} 257 257 \end{figure} … … 259 259 \subsection{Results} 260 260 261 Figures~\ref{fig:yield:jax} and ~\ref{fig:yield:nasus} show the results for the yield experiment.262 Looking at the left column on Intel first, Figure ~\ref{fig:yield:jax:ops} and \ref{fig:yield:jax:ns}, which shows the results for many \ats, in this case 100 \ats for each \proc.261 Figures~\ref{fig:yield:jax} and \ref{fig:yield:nasus} show the results for the yield experiment. 262 Looking at the left column on Intel first, Figures~\ref{fig:yield:jax:ops} and \ref{fig:yield:jax:ns}, which shows the results for many \ats, in this case 100 \ats for each \proc. 263 263 Note, the Y-axis on the graph is twice as large as the Intel cycle-graph. 264 264 A visual glance between the left columns of the cycle and yield graphs confirms my claim that the yield benchmark is unreliable. … … 276 276 This lack of communication is probably why the plateaus due to topology are not present. 277 277 278 Lookking next at the right column on Intel, Figure ~\ref{fig:yield:jax:low:ops} and \ref{fig:yield:jax:low:ns}, which shows the results for few threads, in this case 1 \at for each \proc.278 Lookking next at the right column on Intel, Figures~\ref{fig:yield:jax:low:ops} and \ref{fig:yield:jax:low:ns}, which shows the results for few threads, in this case 1 \at for each \proc. 279 279 As for @cycle@, \CFA's cost of idle sleep comes into play in a very significant way in Figure~\ref{fig:yield:jax:low:ns}, where the scaling is not flat. 280 280 This is to be expected since fewet \ats means \procs are more likely to run out of work. … … 312 312 \caption[Yield Benchmark on AMD]{Yield Benchmark on AMD\smallskip\newline Throughput and scalability as a function of \proc count. 313 313 For throughput, higher is better, for scalability, lower is better. 314 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}314 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 315 315 \label{fig:yield:nasus} 316 316 \end{figure} … … 318 318 Looking now at the results for the AMD architecture, Figure~\ref{fig:yield:nasus}, the results again show a story that is overall similar to the results on the Intel, with increased variation and some differences in the details. 319 319 Note that the maximum of the Y-axis on Intel and AMD differ less in @yield@ than @cycle@. 320 Looking at the left column first, Figure ~\ref{fig:yield:nasus:ops} and \ref{fig:yield:nasus:ns}, \CFA achieves very similar throughput and scaling.320 Looking at the left column first, Figures~\ref{fig:yield:nasus:ops} and \ref{fig:yield:nasus:ns}, \CFA achieves very similar throughput and scaling. 321 321 Libfibre still outpaces all other runtimes, but it encounter a performance hit at 64 \procs. 322 322 This suggest some amount of communication between the \procs that the Intel machine was able to mask where the AMD is not once hyperthreading is needed. 323 323 Go and Tokio still display the same performance collapse than on Intel. 324 Looking next at the right column, Figure ~\ref{fig:yield:nasus:low:ops} and \ref{fig:yield:nasus:low:ns}, all runtime effectively behave the same as they did on the Intel machine.324 Looking next at the right column, Figures~\ref{fig:yield:nasus:low:ops} and \ref{fig:yield:nasus:low:ns}, all runtime effectively behave the same as they did on the Intel machine. 325 325 At high \ats count the only difference was Libfibre's scaling and this difference disappears on the right column. 326 326 This suggest that whatever communication benchmark it encountered on the left is completely circumvented on the right. … … 395 395 \caption[Churn Benchmark on Intel]{\centering Churn Benchmark on Intel\smallskip\newline Throughput and scalability of the Churn on the benchmark on the Intel machine. 396 396 For throughput, higher is better, for scalability, lower is better. 397 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}397 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 398 398 \label{fig:churn:jax} 399 399 \end{figure} … … 402 402 403 403 Figures~\ref{fig:churn:jax} and Figure~\ref{fig:churn:nasus} show the results for the churn experiment. 404 Looking at the left column on Intel first, Figure ~\ref{fig:churn:jax:ops} and \ref{fig:churn:jax:ns}, which shows the results for many \ats, in this case 100 \ats for each \proc, all runtime obtain fairly similar throughput for most \proc counts.404 Looking at the left column on Intel first, Figures~\ref{fig:churn:jax:ops} and \ref{fig:churn:jax:ns}, which shows the results for many \ats, in this case 100 \ats for each \proc, all runtime obtain fairly similar throughput for most \proc counts. 405 405 \CFA does very well on a single \proc but quickly loses its advantage over the other runtimes. 406 406 As expected it scales decently up to 48 \procs and then basically plateaus. … … 422 422 In this particular benchmark, the inherent chaos of the benchmark in addition to small memory footprint means neither approach wins over the other. 423 423 424 Looking next at the right column on Intel, Figure ~\ref{fig:churn:jax:low:ops} and \ref{fig:churn:jax:low:ns}, which shows the results for few threads, in this case 1 \at for each \proc, many of the differences between the runtime disappear.424 Looking next at the right column on Intel, Figures~\ref{fig:churn:jax:low:ops} and \ref{fig:churn:jax:low:ns}, which shows the results for few threads, in this case 1 \at for each \proc, many of the differences between the runtime disappear. 425 425 \CFA outperforms other runtimes by a minuscule margin. 426 426 Libfibre follows very closely behind with basically the same performance and scaling. … … 461 461 \caption[Churn Benchmark on AMD]{\centering Churn Benchmark on AMD\smallskip\newline Throughput and scalability of the Churn on the benchmark on the AMD machine. 462 462 For throughput, higher is better, for scalability, lower is better. 463 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the m nimums.}463 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 464 464 \label{fig:churn:nasus} 465 465 \end{figure} … … 467 467 468 468 Looking now at the results for the AMD architecture, Figure~\ref{fig:churn:nasus}, the results show a somewhat different story. 469 Looking at the left column first, Figure ~\ref{fig:churn:nasus:ops} and \ref{fig:churn:nasus:ns}, \CFA, Libfibre and Tokio all produce decent scalability.469 Looking at the left column first, Figures~\ref{fig:churn:nasus:ops} and \ref{fig:churn:nasus:ns}, \CFA, Libfibre and Tokio all produce decent scalability. 470 470 \CFA suffers particular from a larger variations at higher \proc counts, but almost all run still outperform the other runtimes. 471 471 Go still produces intriguing results in this case and even more intriguingly, the results have fairly low variation. … … 484 484 I did not further investigate what causes these unusual results. 485 485 486 Looking next at the right column, Figure ~\ref{fig:churn:nasus:low:ops} and \ref{fig:churn:nasus:low:ns}, like for Intel all runtime obtain overall similar throughput between the left and right column.486 Looking next at the right column, Figures~\ref{fig:churn:nasus:low:ops} and \ref{fig:churn:nasus:low:ns}, like for Intel all runtime obtain overall similar throughput between the left and right column. 487 487 \CFA, Libfibre and Tokio all have very close results. 488 488 Go still suffers from poor scalability but is now unusual in a different way. … … 592 592 \label{fig:locality:jax:noshare:ns} 593 593 } 594 \caption[Locality Benchmark on Intel]{Locality Benchmark on Intel\smallskip\newline Throughput and scalability as a function of \proc count. For throughput, higher is better, for scalability, lower is better. Each series represent 15 independent runs, the dotted lines are extremes while the solid line is the medium.} 594 \caption[Locality Benchmark on Intel]{Locality Benchmark on Intel\smallskip\newline Throughput and scalability as a function of \proc count. 595 For throughput, higher is better, for scalability, lower is better. 596 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 595 597 \label{fig:locality:jax} 596 598 \end{figure} … … 621 623 \label{fig:locality:nasus:noshare:ns} 622 624 } 623 \caption[Locality Benchmark on AMD]{Locality Benchmark on AMD\smallskip\newline Throughput and scalability as a function of \proc count. For throughput, higher is better, for scalability, lower is better. Each series represent 15 independent runs, the dotted lines are extremes while the solid line is the medium.} 625 \caption[Locality Benchmark on AMD]{Locality Benchmark on AMD\smallskip\newline Throughput and scalability as a function of \proc count. 626 For throughput, higher is better, for scalability, lower is better. 627 Each series represent 15 independent runs, the dashed lines are maximums of each series while the solid lines are the median and the dotted lines are the minimums.} 624 628 \label{fig:locality:nasus} 625 629 \end{figure} 626 630 627 Figures~\ref{fig:locality:jax} and \ref{fig:locality:nasus} show s the results on Intel and AMD respectively.631 Figures~\ref{fig:locality:jax} and \ref{fig:locality:nasus} show the results for the locality experiment. 628 632 In both cases, the graphs on the left column show the results for the @share@ variation and the graphs on the right column show the results for the @noshare@. 629 630 On Intel, Figure~\ref{fig:locality:jax} shows Go trailing behind the 3 other runtimes. 631 On the left of the figure showing the results for the shared variation, where \CFA and Tokio slightly outperform libfibre as expected. 632 And correspondingly on the right, we see the expected performance inversion where libfibre now outperforms \CFA and Tokio. 633 Looking at the left column on Intel first, Figures~\ref{fig:locality:jax:share:ops} and \ref{fig:locality:jax:share:ns}, which shows the results for the @share@ variation. 634 \CFA and Tokio slightly outperform libfibre, as expected based on their \ats placement approach. 635 \CFA and Tokio both unpark locally and do not suffer cache misses on the transferred array. 636 Libfibre on the other hand unparks remotely, and as such the unparked \at is likely to miss on the shared data. 637 Go trails behind in this experiment, presumably for the same reasons that were observable in the churn benchmark. 633 638 Otherwise the results are similar to the churn benchmark, with lower throughput due to the array processing. 634 Presumably the reason why Go trails behind are the same as in Figure~\ref{fig:churn:nasus}. 635 636 Figure~\ref{fig:locality:nasus} shows the same experiment on AMD. 637 \todo{why is cfa slower?} 638 Again, we see the same story, where Tokio and libfibre swap places and Go trails behind. 639 As for most previous results, all runtime suffer a performance hit after 48 \proc, which is the socket boundary. 640 641 Looking next at the right column on Intel, Figures~\ref{fig:locality:jax:noshare:ops} and \ref{fig:locality:jax:noshare:ns}, which shows the results for the @noshare@ variation. 642 The graph show the expected performance inversion where libfibre now outperforms \CFA and Tokio. 643 Indeed, in this case, unparking remotely means the unparked \at is less likely to suffer a cache miss on the array. 644 The leaves the \at data structure and the remote queue as the only source of likely cache misses. 645 Results show both are armotized fairly well in this case. 646 \CFA and Tokio both unpark locally and as a result suffer a marginal performance degradation from the cache miss on the array. 647 648 Looking now at the results for the AMD architecture, Figure~\ref{fig:locality:nasus}, the results show a story that is overall similar to the results on the Intel. 649 Again overall performance is higher and slightly more variation is visible. 650 Looking at the left column first, Figures~\ref{fig:locality:nasus:share:ops} and \ref{fig:locality:nasus:share:ns}, \CFA and Tokio still outperform libfibre, this time more significantly. 651 This is expected from the AMD server, which has smaller and more narrow caches that magnify the costs of processing the array. 652 Go still sees the same poor performance as on Intel. 653 654 Finally looking at the right column, Figures~\ref{fig:locality:nasus:noshare:ops} and \ref{fig:locality:nasus:noshare:ns}, like on Intel, the same performance inversion is present between libfibre and \CFA/Tokio. 655 Go still sees the same poor performance. 656 657 Overall, this experiment mostly demonstrates the two options available when unparking a \at. 658 Depending on the workload, either of these options can be the appropriate one. 659 Since it is prohibitively difficult to detect which approach is appropriate, all runtime much choose one of the two and live with the consequences. 660 661 Once again, this demonstrate that \CFA achieves equivalent performance to the other runtime, in this case matching the faster Tokio rather than Go which is trailing behind. 639 662 640 663 \section{Transfer} … … 723 746 \end{tabular} 724 747 \end{centering} 725 \caption[Transfer Benchmark on Intel and AMD]{Transfer Benchmark on Intel and AMD\smallskip\newline Average measurement of how long it takes for all \ats to acknowledge the leader \at. DNC stands for ``did not complete'', meaning that after 5 seconds of a new leader being decided, some \ats still had not acknowledged the new leader.} 748 \caption[Transfer Benchmark on Intel and AMD]{Transfer Benchmark on Intel and AMD\smallskip\newline Average measurement of how long it takes for all \ats to acknowledge the leader \at. 749 DNC stands for ``did not complete'', meaning that after 5 seconds of a new leader being decided, some \ats still had not acknowledged the new leader.} 726 750 \label{fig:transfer:res} 727 751 \end{figure} … … 733 757 The semaphore variation is denoted ``Park'', where the number of \ats dwindles down as the new leader is acknowledged. 734 758 The yielding variation is denoted ``Yield''. 735 The experiment was only run for the extremes of the number of cores since the scaling per core behaves like previous experiments. 759 The experiment was only run for the extremes of the number of \procs since the scaling is not the focus of this experiment. 760 761 The first two columns show the results for the the semaphore variation on Intel. 762 While there are some differences in latencies, \CFA is consistenly the fastest and Tokio the slowest, all runtime achieve results that are fairly close. 763 Again, this experiment is meant to highlight major differences so latencies within $10\times$ of each other are considered close to each other. 764 765 Looking at the next two columns, the results for the yield variation in Intel, the story is very different. 766 \CFA achieves better latencies, presumably due to the lack of synchronization on the semaphore. 767 Neither Libfibre or Tokio complete the experiment. 768 Both runtime use classical work-stealing scheduling and therefore since non of the work-queues are ever emptied no load balancing occurs. 769 Go does complete the experiment, but with drastically higher latency: 770 latency at 2 \procs is $350\times$ higher than \CFA and $70\times$ higher at 192 \procs. 771 This is because Go also has a classic work-stealing scheduler, but it adds preemption which interrupts the spinning leader after a period. 772 773 Looking now at the results for the AMD architecture, the results show effectively the same story. 774 The first two columns show all runtime obtaining results well within $10\times$ of each other. 775 The next two columns again show \CFA producing low latencies while Libfibre and Tokio do not complete the experiment. 776 Go still has notably higher latency but the difference is less drastic on 2 \procs, where it produces a $15\times$ difference as opposed to a $100\times$ difference on 256 \procs. 777 736 778 This experiments clearly demonstrate that while the other runtimes achieve similar performance in previous benchmarks, here \CFA achieves significantly better fairness. 737 779 The semaphore variation serves as a control group, where all runtimes are expected to transfer leadership fairly quickly.
Note: See TracChangeset
for help on using the changeset viewer.
