-
Notifications
You must be signed in to change notification settings - Fork 12
/
Status
3010 lines (2272 loc) · 109 KB
/
Status
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# ############################################################################
# K42: (C) Copyright IBM Corp. 2000.
# All Rights Reserved
#
# This file is distributed under the GNU LGPL. You should have
# received a copy of the License along with K42; see the file LICENSE.html
# in the top-level directory for more details.
#
# $Id: Status,v 1.237 2005/08/25 18:37:07 rosnbrg Exp $
# ############################################################################
This is for SDET results. If you want postmark or dbench results, look
at file StatusOtherBenchs.
************************ not all the 270s work well. Numbers
************************ should come from k4 or an equivalent
fast machines k4
(as of 8/28/2003 MAA cannot reproduce k4 numbers on k8)
known slow machines k3, k9
This file is for tracking the performance status of k42.
******************* RUNNING EXPERIMENT USING script ***********
Example of running with script on k4:
k42console -R -B killsteal -m k4 -- K42_INITSCR=/home/regress-hw/sysinit
(this assumes you're invoking k42console from the powerpc/noDeb/os containing
the image you want to run, and that you know the machine is free :-)
output (1-way and 4-way numbers) to be pasted in is in kitchroot/tmp/status
******************* RUNNING EXPERIMENT MANUALLY ***********
The experiment is:
Boot noDeb 1 and 4 way
at test prompt:
0
C0xfff
rlogin -l root to system (SDET depends on doing a setuid for each script)
on 1 way: /knfs/home/spec/bin/runSDET -c 0xf7f -s "1 1 1 1 1 1 1 1"
on 4 way: /knfs/home/spec/bin/runSDET -c 0xf7f -s "4 4 4 4 4 4 4 4"
on 24 way: /knfs/home/spec/bin/runSDET -c 0xf7f -s "24 24 24 24 24 24 24 24"
output to be pasted in is in kitchroot/output
***************************************************************
Note: If you hard reboot a machine, it will probably come up with a single
processor. You need to use the console to configure 4 processors and
fast reboot. Put your new results at the bottom of the file.
If you change the "best" flags story, update the above instructions when
you post the performance change that goes with the flag change.
If your changes slow the system down, please consider consulting with the
group before dropping them.
Please head your changes status with a one liner about what the change is.
***************************************************************************
Current floor system:
SDET_THROUGHPUT: For 2 script(s) throughput is 1682.2 scripts/hour
Use ref bits to improve page table eviction choice.
SDET_THROUGHPUT: For 2 script(s) throughput is 1718.4 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1722.5 scripts/hou
-------------
Added FCMPrimitiveKernelMultiRep and ensure cachesync and dirty is set on zero
fill faults.
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
Was stable at 1800.0 for 4 runs then degraded to 1786.6 on run 5
(recompiled rebooted and reran again with no functional change and found
more variance but peek still aht 1800.0)
Played a bit more and got a peek of:
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
-------------
Fixed FCMPrimitiveKernelMultiRep, FCMCommonMultiRep* and DHashTable to allow
a specification of an initial number of pages for the HashTable to avoid
resizes during kernel startup. This eliminated the problems on using
the FCMPrimitiveKernelMultiRep on k0. Have NOT seen previous peek of 1813.6
but change is too important not to commit.
Average seems to be about 1800.0
First Boot
SDET_THROUGHPUT: For 2 script(s) throughput is 1791.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
Second Boot
SDET_THROUGHPUT: For 2 script(s) throughput is 1791.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1791.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
-------------
Changes in shared versus non-shared code. The code is still off
by default. The numbers look like hte ones Jonathan has reported
above, with avarage of 1802 (Dilma, May 12 7:51pm)
-------------
Improvements in FileLinuxFile.C (moving towards getting the code
for small file opitmization in)
SDET_THROUGHPUT: For 2 script(s) throughput is 1545.1 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
(Dilma, May 13, 10:15am)
--------------
Improve non blocking hash growth logic, release lock more quickly, and add
interface to request a larger process set in servers. add call to
ramfs to request a larger process set.
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
Changes in the linux.k42 (decreased lock contention on a linux
lock on 12way, 24-way).
SDET_THROUGHPUT: For 2 script(s) throughput is 1782.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1795.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
(On k1 the average os 1714)
=============================================
No relevant change in terms of functionality, but here are the
numbers anyway:
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
=============================================
Changed ramfs token-polling interval to 100 ms.
Changed 5ms. delay to a thread-yield when waiting for an MP msg buf.
(Bryan, 5/14 08:45)
SDET_THROUGHPUT: For 2 script(s) throughput is 1791.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1718.4 scripts/hour
============================================
Mark established pages as cached synced and dirty
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
============================================
Slightly improved loop that clears referenced bits.
(Bryan, 5/14, 13:45)
SDET_THROUGHPUT: For 2 script(s) throughput is 1791.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1804.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
============================================
Dilma (5/14 2:15pm) changed linux.k42 so that allocation of temporary
struct file data structures in the linux emulation layer bypass the slab
allocator (they use __ramfs_alloc and __ramfs_free)
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1827.4 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1822.8 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1818.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1822.8 scripts/hour
============================================
Dilma (5/14 9:16pm) change in ramfs code. Throughput from
k2 is 1818
============================================
Bryan (5/15 8:00am) No changes. I just want it on the record that I'm
unable to get better than 1813 scripts/hour on a clean floor system
on k4.
SDET_THROUGHPUT: For 2 script(s) throughput is 1800.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1813.6 scripts/hour
============================================
Bob (5/15 3:30pm) mainly changes for hot-swapping should all be
ifdefed out. The one that might affect SDET performance is placing
a printf in TransEntry under the Kernel Info Run Silent flag.
SDET_THROUGHPUT: For 2 script(s) throughput is 1809.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1818.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1832.1 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1822.8 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1827.4 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1822.8 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1827.4 scripts/hour
============================================
Marc (5/15 5:45) eliminate unneeded tlbie calls
SDET_THROUGHPUT: For 2 script(s) throughput is 1832.1 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1836.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1855.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1846.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1841.4 scripts/hour
============================================
Jonathan (5/15 18:10) ugly hackery to remove xhandles from resMgr
External interface is completely static.
SDET_THROUGHPUT: For 2 script(s) throughput is 1846.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1855.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1860.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1850.9 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1855.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1860.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1860.5 scripts/hour
============================================
Orran (5/16 00:16) Paging infrastructure committed, only has impact if
down to 2000 pages free, tested on k8
SDET_THROUGHPUT: For 2 script(s) throughput is 1850.9 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1855.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1865.3 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1850.9 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1860.5 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1855.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1870.1 scripts/hour
============================================
Marc (5/16) Better stats in Interted Page Table
Numbers not as good - ran on k4. I'll verify and back out if its
really worse - dropping now so it gets into Bryan's morning run.
(It can't be worse - but ...)
SDET_THROUGHPUT: For 2 script(s) throughput is 1827.4 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1846.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1846.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1836.7 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1846.2 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 1836.7 scripts/hour
============================================
Bryan (5/16) Current baseline numbers on k4. Clean floor system. Booted
3 times. Ran SDET 8 times on each boot:
boot 1: 1850.9 1841.4 1846.2 1841.4 1850.9 1836.7 1841.4 1836.7
boot 2: 1841.4 1850.9 1850.9 1846.2 1846.2 1846.2 1855.7 1841.4
boot 3: 1800.0 1832.1 1841.4 1832.1 1832.1 1836.7 1846.2 1836.7
============================================
Bryan (5/16) Moved ramfsServer into the server resource domain.
boot 1: 1832.1 1841.4 1836.7 1836.7 1832.1 1832.1 1836.7 1832.1
boot 2: 1836.7 1832.1 1832.1 1827.4 1832.1 1832.1 1832.1 1832.1
boot 3: 1822.8 1822.8 1832.1 1822.8 1827.4 1827.4 1832.1 1827.4
============================================
Bryan (5/16) Added lock-contention tracing for spin-locks.
boot 1: 1827.4 1846.2 1841.4 1841.4 1846.2 1841.4 1846.2 1841.4
boot 2: 1822.8 1827.4 1832.1 1822.8 1827.4 1822.8 1827.4 1827.4
boot 3: 1822.8 1818.2 1827.4 1818.2 1818.2 1809.0 1818.2 1818.2
============================================
Marc (5/16) Improve eviction logic - at least on paper
boot 1: 1827.4 1836.7 1841.4 1832.1 1841.4
boot 2: 1850.9 1850.9 1850.9 1846.2 1850.9
============================================
Bryan (5/16) Picked up Jimi's noDeb nroff, but not Marc's latest change.
boot 1: 1956.5 1961.9 1972.6 1961.9 1961.9 1961.9 1972.6 1961.9
boot 2: 1961.9 1967.2 1978.0 1972.6 1978.0 1967.2 1961.9 1961.9
boot 3: 1961.9 1967.2 1967.2 1956.5 1961.9 1956.5 1956.5 1951.2
==============================================
Dilma (5/16, 3pm)
Reduced locking in linux code
boot just once, I saw this 5 times in a row (k4)
SDET_THROUGHPUT: For 2 script(s) throughput is 1994.5 scripts/hour
============================================
Bryan (5/16) Set FairBLockSpinCount in ramfsServer to 10000.
Measurements are on k8.
boot 1: 1956.5 1967.2 1967.2 1956.5 1961.9 1961.9 1961.9 1961.9
boot 2: 1967.2 1961.9 1972.6 1967.2 1972.6 1967.2 1967.2 1967.2
boot 3: 1956.5 1978.0 1972.6 1972.6 1983.5 1972.6 1978.0 1961.9
============================================
Marc (5/16) Avoid many cache sync operations by adding first swag at
map for execute logic.
boot 1: 2016.8 2022.5 2028.2 2028.2 2033.9
boot 2: 2016.8 2016.8 2028.2 2022.5 2022.5 2033.9 2039.7 2039.7
boot 3: 2022.5 2033.9 2028.2 2022.5 2033.9 2028.2 2039.7 2028.2
============================================
Bryan (5/17) Made kernel cleanup daemon run on a 100ms. interval by default.
Necessary for sustained operation on k0, and for honest
reporting of results (even 2-way).
May have slowed 2-way performance slightly. It's hard to say.
Measurements are on k4.
boot 1: 2028.2 2028.2 2033.9 2005.6 2033.9 2028.2 2028.2 2022.5
boot 2: 2022.5 2022.5 2028.2 2011.2 2011.2 2022.5 2022.5 2016.8
boot 3: 2028.2 2022.5 2028.2 2022.5 2022.5 2022.5 2022.5 2022.5
============================================
Marc (5/17) Redo numbers, change runSDET to do a single make for the list
(Numbers above had file name in FR debug code turned on!)
boot 1: 2022.5 2028.2 2045.5 2028.2 2039.7 2028.2 2028.2 2022.5
============================================
Bryan (5/17) Changed runSDET to disable thinwire polling during experiment.
Measurements are on k4.
boot 1: 2039.7 2039.7 2039.7 2039.7 2045.5 2039.7 2051.3
boot 2: 2039.7 2039.7 2039.7 2033.9 2039.7 2045.5 2045.5
boot 3: 2051.3 2039.7 2051.3 2051.3 2057.1 2045.5 2051.3
============================================
Bob (5/17) most of the infrsatructure needed to gather additional lock
contention results - more options to trace menu
Measurements are on k4.
boot1
SDET_THROUGHPUT: For 2 script(s) throughput is 2057.1 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2063.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2063.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2057.1 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2051.3 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
boot2
SDET_THROUGHPUT: For 2 script(s) throughput is 2063.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2063.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2063.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2051.3 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 2069.0 scripts/hour
============================================
Bob (5/18) put in rest of lock infrastructure for detecting handoff time
and if we ever delay for 1msec in lock timing window
2080.9 2080.9 2080.9 2080.9 2080.9 2080.9 2069.0 2074.9
2057.1 2063.0 2063.0 2063.0 2063.0 2057.1 2069.0 2057.1
2063.0 2074.9 2057.1 2063.0 2069.0 2074.9 2074.9 2080.9
============================================
Marc (5/18) convert XObject free list accesses in XHandleTrans to non-blocking
2063.0 2063.0 2074.9 2051.3 2074.9 2063.0 2069.0
2057.1 2063.0 2069.0 2057.1 2069.0 2063.0 2063.0
2057.1 2057.1 2057.1 2051.3 2057.1 2057.1 2057.1
============================================
Marc (5/20) convert XObject free list accesses in XHandleTrans to non-blocking
with fewer bugs/stupidities
2039.7 2039.7 2051.3 2051.3 2051.3 2045.5 2045.5
2045.5 2051.3 2063.0 2051.3 2057.1 2057.1 2063.0
2045.5 2057.1 2063.0 2063.0 2074.9 2063.0 2063.0
============================================
Bryan (5/20) Allocate per-vp structures in SegmentHATPrivate only as needed.
2074.9 2069.0 2074.9 2069.0 2080.9 2074.9 2087.0 2074.9
2080.9 2057.1 2074.9 2057.1 2074.9 2057.1 2080.9 2074.9
2074.9 2069.0 2080.9 2080.9 2099.1 2087.0 2099.1 2087.0
============================================
Bryan (5/20) Disabled MICHALKLUDGE in PageAllocatorKernPinned.H.
2074.9 2099.1 2111.4 2093.0 2111.4 2105.3 2111.4 2105.3
2080.9 2057.1 2074.9 2057.1 2074.9 2057.1 2080.9 2074.9
2093.0 2105.3 2117.6 2099.1 2111.4 2111.4 2111.4 2105.3
============================================
Bryan (5/20) Set lock spin counts to 10000 in the kernel, baseServers, & ramfs.
2099.1 2093.0 2105.3 2087.0 2111.4 2105.3 2105.3 2105.3
2099.1 2099.1 2105.3 2093.0 2105.3 2099.1 2099.1 2093.0
2105.3 2099.1 2111.4 2105.3 2130.2 2123.9 2123.9 2117.6
============================================
Bob (5/20) Added capability to compile out tracing should have no perf effect
2105.3 2105.3 2111.4 2105.3 2111.4 2111.4 2099.1 2105.3
2123.9 2117.6 2123.9 2111.4 2123.9 2123.9 2123.9 2130.2
2123.9 2111.4 2123.9 2117.6 2123.9 2117.6 2123.9 2111.4
============================================
Bob (5/28) Cleanup - added tracing dbg, kernelInfo dbg - no perf effect
2136.5 2123.9 2130.2 2117.6 2130.2 2111.4 2117.6 2111.4
2130.2 2111.4 2117.6 2117.6 2136.5 2123.9 2123.9 1223.9
============================================
Marc/Jonathan (5/28) New clustered object translation table management.
We accept the performance degradation, at least
for now.
2057.1 2063.0 2074.9 2057.1 2063.0 2057.1 2057.1
2074.9 2063.0 2074.9 2069.0 2080.9 2069.0 2074.9 2069.0
============================================
Orran (5/30) Clean up of paging infrastructure
2063.0 2063.0 2063.0 2069.0 2069.0 2051.3 2063.0
2069.0 2057.1 2063.0 2069.0 2057.1 1972.6 2051.3 2000.0
============================================
Bob (6/3) added traceXxxEnable function and other cleanup - no perf effect
2074.9 2069.0 2093.0 2069.0 2080.9 2074.9 2087.0 2069.0
2069.0 2039.7 2057.1 2063.0 2063.0 2057.1 2057.1 2045.5
============================================
Dilma (6/3) use of reader-writer locks for DirLinuxFS objects
2051.3 2045.5 2063.0 2045.5 2063.0 2051.3 2045.5 2057.1 2057.1
2051.4 2057.1 2051.3 2057.1 2063.0 2051.3 2063.0 2051.3
============================================
Dilma (6/4) (1) changes allowing derived classes from ServerFile
to have their own lock definition; for now we're using normal
(BLock) locks for everybody;
(2) moved stuff in ServerFileBlockRamFS::handleXObj() to base class
ServerFileBlock, so that we don't have code repetition.
2045.5 2057.1 2069.0 2039.7 2051.3 2045.5 2051.3 2051.3
2057.1 2045.5 2063.0 2045.5 2045.5 2045.5 1951.2 2000.0 1935.5
k0 numbers:
SDET_THROUGHPUT: For 1 script(s) throughput is 1730.8 scripts/hour
SDET_THROUGHPUT: For 2 script(s) throughput is 3412.3 scripts/hour
SDET_THROUGHPUT: For 3 script(s) throughput is 5046.7 scripts/hour
SDET_THROUGHPUT: For 4 script(s) throughput is 6605.5 scripts/hour
SDET_THROUGHPUT: For 5 script(s) throughput is 8144.8 scripts/hour
SDET_THROUGHPUT: For 6 script(s) throughput is 9686.1 scripts/hour
SDET_THROUGHPUT: For 7 script(s) throughput is 11454.5 scripts/hour
SDET_THROUGHPUT: For 8 script(s) throughput is 13150.7 scripts/hour
SDET_THROUGHPUT: For 9 script(s) throughput is 14594.6 scripts/hour
SDET_THROUGHPUT: For 10 script(s) throughput is 16216.2 scripts/hour
SDET_THROUGHPUT: For 11 script(s) throughput is 17522.1 scripts/hour
SDET_THROUGHPUT: For 12 script(s) throughput is 19547.5 scripts/hour
SDET_THROUGHPUT: For 13 script(s) throughput is 20259.7 scripts/hour
SDET_THROUGHPUT: For 14 script(s) throughput is 22600.9 scripts/hour
SDET_THROUGHPUT: For 15 script(s) throughput is 23478.3 scripts/hour
SDET_THROUGHPUT: For 16 script(s) throughput is 24615.4 scripts/hour
SDET_THROUGHPUT: For 17 script(s) throughput is 26266.1 scripts/hour
SDET_THROUGHPUT: For 18 script(s) throughput is 28296.9 scripts/hour
SDET_THROUGHPUT: For 19 script(s) throughput is 29106.4 scripts/hour
SDET_THROUGHPUT: For 20 script(s) throughput is 30252.1 scripts/hour
SDET_THROUGHPUT: For 21 script(s) throughput is 31500.0 scripts/hour
SDET_THROUGHPUT: For 22 script(s) throughput is 33000.0 scripts/hour
SDET_THROUGHPUT: For 23 script(s) throughput is 34644.4 scripts/hour
SDET_THROUGHPUT: For 24 script(s) throughput is 35850.6 scripts/hour
SDET_THROUGHPUT: For 24 script(s) throughput is 35850.6 scripts/hour
============================================
Bryan (6/4) - Miscellaneous changes:
Synchronize MPMsgMgr initialization explicitly.
Eliminate KernelInfo::TRACE_LOCK_CONTENTION and instead guard
trace-related work with a test of the relevant trace bit.
Yield once in fork() before delaying when waiting for active
threads to go away.
2045.5 2051.3 2063.0 2045.5 2063.0 2051.3 2057.1 2051.3
2045.5 2039.7 2057.1 2039.7 2063.0 2051.3 2057.1 2045.5
2039.7 2051.3 2063.0 2045.5 2069.0 2045.5 2063.0 2057.1
============================================
Bryan (6/5) - Create CPU-intensive resource domains lazily.
2087.0 2130.2 2136.5 2123.9 2136.5 2130.2 2130.2 2136.5
2123.9 2123.9 2123.9 2123.9 2130.2 2123.9 2123.9 2123.9
2130.2 2123.9 2136.5 2123.9 2142.9 2136.5 2130.2 2123.9
============================================
Dilma (6/5, 2:30pm) - ServerFile use BLock and DirLinuxFS (and its
derived classes) use RWBLock.
2099.1 2105.3 2111.4 2111.4 2099.1 2093.0 2099.1 2093.0 2111.4
2093.0 2123.9 2111.4 2117.6 2111.4 2111.4 2111.4 2117.6 2111.4
===========================================
Dilma (6/6) - cosmetic change: parameter in FileLinuxServer::_dup
declared as _inout should be _in only
2111.4 2111.4 2117.6 2111.4 2111.4 2111.4 2111.4 2111.4 2087.0
2111.4 2111.4 2111.4 2117.6 2105.3 2105.3 2105.3 2117.6 2117.6
============================================
Bryan (6/6) - Fix timing problems that have crept in. Initialize trace mask
earlier and synchronize resource-manager initialization.
Results obtained on k8.
2117.6 2142.9 2142.9 2130.2 2142.9 2130.2 2136.5 2136.5
2123.9 2130.2 2136.5 2130.2 2130.2 2130.2 2130.2 2130.2
2117.6 2123.9 2136.5 2123.9 2136.5 2130.2 2130.2 2123.9
============================================
Dilma (6/6) Changed control flags so that if you specify C 0xfff, you
have both small file and non-shared files optimization on. (The
flags NON_SHARING_FILE_OPT and SMALL_FILE_OPT are off by default.)
2149.3 2181.8 2181.8 2175.2 2181.8 2181.8 2175.2 2181.8 2188.4
2181.8 2175.2 2181.8 2175.2 2168.7 2181.8 2155.7 2195.1 2188.4
============================================
Bryan (6/6) - Initialize the trace mask even earlier. Should have no
impact on performance.
Results obtained on k8.
2136.5 2175.2 2175.2 2175.2 2188.4 2175.2 2181.8 2136.5
2142.9 2162.2 2175.2 2168.7 2188.4 2162.2 2168.7 2162.2
2142.9 2155.7 2168.7 2168.7 2168.7 2162.2 2136.5 2162.2
============================================
Bryan (6/6) - Added trace events in Delay's. No impact on performance.
Results obtained on k8.
2142.9 2181.8 2149.3 2188.4 2188.4 2175.2 2188.4 2188.4
2142.9 2175.2 2181.8 2181.8 2188.4 2175.2 2181.8 2181.8
2142.9 2142.9 2188.4 2181.8 2188.4 2181.8 2195.1 2188.4
============================================
Michal (6/7) - New virtfs and some FR debugging stuff.
Changes should not have any performance impact, but they do.
A closer look shows that we are sensitive to innocuous
changes that may affect code/data layout. (k8)
2087.0 2123.9 2136.5 2136.5 2123.9 2130.2
===================================================
Bob (6/9) - locking and tracing structure infrastructure no perf impact
2149.3 2175.2 2175.2 2181.8 2188.4 2175.2 2181.8 2181.8
2181.8 2201.8 2201.8 2201.8 2162.2 2195.1 2188.4 2188.4
============================================
Bob (6/9) - removal of mips - no perf impoact
2181.8 2188.4 2175.2 2175.2 2188.4 2181.8 2181.8 2181.8
2175.2 2175.2 2188.4 2181.8 2181.8 2175.2 2181.8 2181.8
============================================
Michal (6/10) - Disk support re-enabled.
Overhauled devfs.
fdisk (from util-linux) inclded.
StreamServer locking overhaul (to handle socket deadlock).
2136.5 2155.7 2149.3 2168.7 2162.2 2162.2 2155.7 2149.3
(My numbers always tend to be a bit low, don't know why.)
============================================
Bryan (6/10) - Changed eviction algorithm in mapping-fault handler to match
InvertedPageTable's algorithm.
Results obtained on k2.
2123.9 2149.3 2162.2 2162.2 2162.2 2155.7 2149.3 2142.9
2136.5 2142.9 2142.9 2149.3 2155.7 2155.7 2155.7 2155.7
2142.9 2155.7 2162.2 2162.2 2175.2 2168.7 2168.7 2162.2
============================================
Bryan (6/10) - Changed sense of the IO/CPU migration control bit. Now
migration is on by default but off with our normal SDET args.
Results obtained on k2.
2162.2 2168.7 2175.2 2162.2 2175.2 2168.7 2175.2 2162.2
2155.7 2168.7 2181.8 2175.2 2181.8 2168.7 2181.8 2168.7
2155.7 2168.7 2175.2 2175.2 2181.8 2175.2 2188.4 2175.2
============================================
Orran (6/11) - just a single tty object, no separate K42 stdin and stdout obj
2149.3 2181.8 2181.8 2181.8 2181.8 2175.2 2181.8 2168.7
2168.7 2168.7 2175.2 2175.2 2175.2 2162.2 2188.4 2175.2
2168.7 2162.2 2175.2 2181.8 2175.2 2168.7 2175.2 2175.2
============================================
Bryan (6/12) - Changed lock spin counts to 1000000 in kernel, ramfs, and
baseServers.
Results obtained on k8.
2149.3 2195.1 2195.1 2201.8 2195.1 2208.6 2201.8 2208.6
2155.7 2181.8 2188.4 2168.7 2188.4 2195.1 2201.8 2188.4
2162.2 2195.1 2201.8 2188.4 2201.8 2195.1 2195.1 2201.8
============================================
Dilma (6/13) Added support for intercepting NFS activity
during SDET.
2168.7 2201.8 2208.6 2195.1 2208.6 2195.1 2195.1 2195.1 2201.8
============================================
Marc (6/13) Added shared segment support to both shared and multirep
FCM's, and changed runProcessCommon to map text into a segment sized
region.
2250.0 2257.1 2264.2 2264.2 2271.3 2257.1 2278.5 2264.2
2250.0 2236.0 2257.1 2250.0 2264.2 2250.0 2250.0 2257.1
2257.1 2271.3 2278.5 2271.3 2285.7 2278.5 2271.3 2264.2
============================================
Marc (6/14) Fix bug - was unmapping text in fork because fork child
had wrong length for text region - not rounded up to segment size.
2352.9 2384.1 2384.1 2384.1 2392.0 2384.1 2392.0 2392.0
2360.7 2360.7 2376.2 2368.4 2376.2 2368.4 2368.4 2368.4
2360.7 2376.2 2384.1 2368.4 2392.0 2376.2 2376.2 2368.4
============================================
Marc (6/17) Fix bugs - get tlbie right (how did it ever work?)
Fix shared seg code for MP runs. Turn MICHALKLUDGE back on bug fix it.
Performance was same as above until turned MICAHKLUDGE on - I don't believe
degradation is cause by that but who knows? (numbers from k4) Note one
"good" run!
2345.3 2360.7 2360.7 2360.7 2368.4 2360.7 2360.7 2345.3
2345.3 2360.7 2360.7 2360.7 2384.1 2360.7 2360.7 2376.2
============================================
Bryan (6/18) - Restructured activate/deactivate. Should be no performance
impact.
Results obtained on k4.
2307.7 2368.4 2360.7 2352.9 2368.4 2345.3 2360.7 2360.7
2307.7 2345.3 2368.4 2337.7 2352.9 2352.9 2352.9 2345.3
2315.1 2360.7 2360.7 2360.7 2352.9 2352.9 2360.7 2360.7
============================================
Marc (6/18) - Remove Frame Array
2322.6 2352.9 2360.7 2352.9 2376.2 2368.4 2360.7 2360.7
2337.7 2360.7 2376.2 2360.7 2384.1 2376.2 2384.1 2368.4
============================================
Bryan (6/19) - Mechanism for pro-active generation-rolling. Not used yet.
Results obtained on k8.
2337.7 2345.3 2368.4 2360.7 2352.9 2352.9 2352.9 2368.4
2322.6 2360.7 2376.2 2352.9 2352.9 2345.3 2337.7 2352.9
2330.1 2368.4 2368.4 2352.9 2360.7 2360.7 2368.4 2360.7
============================================
Dilma (6/19) - Added k42 ramfs server. Not used yet, so it
has no impact on performance. Numbers from k4.
2345.3 2384.1 2384.1 2376.2 2376.2 2376.2 2376.2
2368.4 2392.0 2368.4 2360.7 2376.2 2384.1
============================================
Orran (6/21) - Couple of bug fixes, and ifdefed out lazy stuff still
not working, some additional work for lazy path is being done...
2345.3 2352.9 2352.9 2330.1 2345.3 2352.9 2360.7
2345.3 2337.7 2330.1 2352.9 2360.7 2345.3 2337.7 2337.7
============================================
Dilma (6/24) - Numbers from Friday 6/21 with k42 ramfs
as the default file system server in runSDET. Numbers were
collected on k4.
2408.0 2449.0 2432.4 2440.7 2449.0 2449.0 2449.0
2457.3 2449.0 2449.0 2457.3 2449.0 2449.0
==============================================
Marc (7/4) - Perfect except for evictions unmap info - and don't unmap
whole segs by abandoning vsid - really unmap them. On k4
2440.7 2457.3 2465.8 2465.8 2465.8 2474.2 2474.2 2465.8
2449.0 2457.3 2474.2 2465.8 2465.8 2465.8 2482.8 2474.2
============================================
Bryan (7/8) - Use PPC instead of MPMsgs for communication between dispatchers
on the same VP. Should have no impact on SDET performance.
Results obtained on k8.
2432.4 2465.8 2474.2 2474.2 2482.8 2474.2 2491.3 2482.8
2440.7 2465.8 2465.8 2465.8 2465.8 2474.2 2465.8 2474.2
2440.7 2474.2 2474.2 2482.8 2491.3 2482.8 2500.0 2491.3
===========================================
Dilma (7/9) - Finished lazy i/o reinitialization on fork, but
it is not turned on by default yet (I want to test
it on k0 before turning it on), so this commit should
have no impact on SDET performance.
Results from k4.
2457.3 2474.2 2457.3 2474.2 2474.2 2465.8 2465.8 2465.8
2474.2 2474.2 2474.2 2474.2 2465.8 2424.2 2449.0 2465.8
2457.3 2465.8 2465.8 2465.8 2474.2 2482.8 2465.8 2465.8
2482.8 2474.2
===========================================
Michal (7/9) - Restructured sysctl (more modular, easier to expand).
Swap-to-disk support.
Page-fault path timing measurement facility (disabled).
2449.0 2449.0 2449.0 2449.0 2457.3 2449.0 2449.0 2449.0
===========================================
Michal (7/17) - Async disk support for pager.
(Yes, I really did get identical results as last time.)
2449.0 2449.0 2449.0 2449.0 2457.3 2449.0 2449.0 2449.0
===========================================
Michal (7/26) - Fixup of I/O blocking/EOF/error code semantics
2432.4 2424.2 2432.4 2432.4 2424.2 2432.4 2424.2 2432.4
============================================
Dilma (7/31) - lazy i/o reinitialization is on by default (it's
on since yesterday). Numbers from k4
2500.0 2508.7 2508.7 2517.5 2500.0 2517.5 2517.5 2508.7 2508.7
2500.0 2508.7 2508.7 2508.7 2508.7 2517.5 2508.7 2517.5 2517.5
(these are two consecutive executions of runSDET)
============================================
Dilma (8/1) - cosmetic changes.
2500.0 2508.7 2508.7 2491.3 2491.3 2508.7 2517.5 2508.7 2517.5
2526.3 2526.3 2526.3 2535.2 2535.2 2526.3 2535.2 2526.3 2526.3
2526.3 2535.2 2535.2 2526.3
(numbers from a single runSDET command)
============================================
Dilma (8/2) - returning error on setFilePositon when operation
results in offset < 0
2500.0 2535.2 2535.2 2526.3 2526.3 2526.3 2535.2 2535.2 2544.2
2544.2 2544.2 2535.2 2535.2 2535.2 2544.2 2535.2 2544.2 2553.2
2544.2 2544.2 2553.2 2544.2
============================================
Dilma (8/5) - dealing with protected/public issues in FileLinux.C
Certainly not to have any impact on performance
(these numbers include also changes on StreamServerPty
done this morning (by Michal, I guess)).
2482.8 2508.7 2508.7 2526.3 2517.5 2517.5 2526.3 2517.5
2535.2 2535.2 2526.3 2535.2 2517.5 2526.3 2535.2 2535.2
2535.2 2535.2 2535.2 2535.2 2526.3 2544.2
============================================
Jonathan (8/8) - Added code to delete the old object after a swap.
A disposition parameter is used to control the behaviour.
The actual destruction is if defed out until Dilma can test.
K8 Numbers
2482.8 2517.5 2508.7 2526.3 2526.3 2517.5 2526.3 2526.3
2526.3 2526.3 2526.3 2526.3 2535.2 2526.3
============================================
Bryan (8/19) - New baseline numbers. Here are the results of six consecutive
boot sessions on k8 and six on k4:
k8:
2482.8 2517.5 2517.5 2517.5 2508.7 2517.5 2526.3 2526.3
2491.3 2500.0 2508.7 2500.0 2500.0 2517.5 2526.3 2517.5
2482.8 2500.0 2500.0 2508.7 2500.0 2508.7 2508.7 2508.7
2491.3 2491.3 2500.0 2500.0 2500.0 2508.7 2508.7 2508.7
2482.8 2526.3 2535.2 2526.3 2508.7 2535.2 2535.2 2535.2
2491.3 2508.7 2508.7 2526.3 2517.5 2517.5 2526.3 2517.5
k4:
2474.2 2508.7 2508.7 2508.7 2508.7 2508.7 2500.0 2508.7
2491.3 2517.5 2508.7 2517.5 2517.5 2526.3 2526.3 2508.7
2482.8 2500.0 2500.0 2491.3 2508.7 2517.5 2508.7 2508.7
2491.3 2508.7 2508.7 2508.7 2517.5 2508.7 2517.5 2508.7
2482.8 2517.5 2517.5 2508.7 2517.5 2526.3 2535.2 2517.5
2491.3 2508.7 2517.5 2500.0 2517.5 2508.7 2517.5 2508.7
============================================
Marc (8/23) - remove small sillyness in fork logic which CANNOT make
things slower - number from k8
2474.2 2500.0 2526.3 2517.5 2526.3 2535.2 2517.5 2517.5
2474.2 2500.0 2508.7 2500.0 2491.3 2500.0 2500.0 2500.0
2482.8 2474.2 2508.7 2508.7 2517.5 2517.5 2500.0 2508.7
Orran (9/10) - got console out of picture
2508.7 2508.7 2508.7 2517.5 2517.5 2517.5 2508.7 2508.7
2508.7 2517.5 2508.7 2508.7 2526.3 2517.5 2508.7 2517.5
===========================================
End of XCOFF toolchain.
===========================================
Dilma (9/30) - after being able to run SDET again: (from k3)
3185.8 3257.9 3257.9 3272.7 3287.7 3272.7 3185.8 3257.9
3287.7 3228.7 3228.7 3302.8 3302.8 3302.8 3243.2 3302.8
3287.7 3272.7
============================================
Bryan (10/8) - Clone and signal restructuring.
Results obtained on k8.
3348.8 3428.6 3428.6 3412.3 3428.6 3318.0 3364.5 3428.6
3302.8 3364.5 3380.3 3364.5 3380.3 3287.7 3364.5 3380.3
3364.5 3412.3 3380.3 3380.3 3428.6 3412.3 3396.2 3412.3
============================================
Bryan (10/9) - Switch to using glibc <signal.h> from <linux/signal.h>.
Should be little or no performance impact.
Results obtained on k8.
3287.7 3396.2 3412.3 3396.2 3461.5 3412.3 3428.6 3428.6
3318.0 3380.3 3380.3 3396.2 3396.2 3380.3 3396.2 3396.2
3333.3 3396.2 3396.2 3333.3 3428.6 3428.6 3412.3 3380.3
============================================
Bryan (10/30) - Sanity check. No one has made measurements for awhile.
Results obtained on k4.
3333.3 3272.7 3380.3 3396.2 3412.3 3364.5 3364.5 3272.7
3333.3 3380.3 3396.2 3364.5 3380.3 3348.8 3412.3 3364.5
3333.3 3380.3 3396.2 3348.8 3287.7 3380.3 3364.5 3396.2
3287.7 3333.3 3333.3 3396.2 3396.2 3396.2 3380.3 3380.3
3333.3 3380.3 3318.0 3396.2 3302.8 3380.3 3348.8 3396.2
============================================
Bryan (10/30) - Restructured thread migration.
Results obtained on k4.
3287.7 3348.8 3412.3 3428.6 3380.3 3412.3 3412.3 3396.2
3364.5 3364.5 3428.6 3396.2 3412.3 3478.3 3302.8 3380.3
3318.0 3318.0 3364.5 3364.5 3257.9 3461.5 3412.3 3428.6
3302.8 3380.3 3445.0 3428.6 3333.3 3380.3 3364.5 3428.6
3272.7 3333.3 3412.3 3396.2 3412.3 3428.6 3396.2 3428.6
3257.9 3412.3 3302.8 3333.3 3478.3 3380.3 3364.5 3333.3
============================================
Marc (11/1) - changed treatment of shared sements and cache sync
Results on k8
3287.7 3012.6 3318.0 3364.5 3302.8 3348.8 3348.8
3171.8 3348.8 3348.8 3302.8 3333.3 3333.3 3318.0
3333.3 3318.0 3333.3 3318.0 3348.8 3302.8 3318.0
3185.8 3287.7 3318.0 3200.0 3228.7 3396.2 3380.3
3287.7 3333.3 3333.3 3364.5 3348.8 3348.8 3333.3
3272.7 3348.8 3333.3 3348.8 3257.9 3318.0 3214.3
============================================
Marc (11/4) - replace copy on write kludge with copy on write support
Results on k8
3257.9 3272.7 3272.7 3287.7 3257.9 3302.8 3318.0 3257.9
3287.7 3302.8 3302.8 3333.3 3157.9 3318.0 3287.7 3396.2
3272.7 3243.2 3116.9 3272.7 3318.0 3318.0 3318.0 3214.3
3228.7 3318.0 3000.0 3302.8 3302.8 3272.7 3272.7 3287.7
3157.9 3200.0 3396.2 3396.2 3287.7 3171.8 3287.7 3257.9
3243.2 3287.7 3318.0 3302.8 3130.4 3243.2 3272.7 3287.7
With no multi-rep fcms
2868.5 3144.1 3116.9 3025.2 3050.8 3103.4 3130.4 3103.4
============================================
Michal (11/24) - dynamic linking increases overall workload, results go down
Results on k8
2236.0 2285.7 2293.0 2278.5 2236.0 2264.2 2236.0 2243.0 2222.2 2222.2
2264.2 2257.1 2257.1 2293.0 2243.0 2300.3 2271.3 2271.3 2271.3 2293.0
2285.7 2285.7 1967.2 2057.1 2149.3 2208.6 2074.9 2149.3 2278.5 2278.5
2271.3 2271.3 2222.2 2264.2 2271.3 2271.3 2264.2 2250.0 2264.2 2257.1
============================================
Michal (11/24) - new exec procedure (fast exec doesn't creat new process)
Results on k8
2457.3 2424.2 2376.2 2440.7 2457.3 2416.1 2278.5 2432.4 2376.2 2352.9
2337.7 2105.3 2074.9 2271.3 2264.2 2337.7 2352.9 2307.7 2337.7 2360.7
2368.4 2368.4 2360.7 2408.0 2376.2 2352.9 2440.7 2330.1 2352.9 2408.0
2345.3 2322.6 2352.9 2440.7 2236.0 2074.9 2117.6 2175.2 2400.0 2408.0
Note: 1-way numbers are approximately 1400, so we're losing 350 scripts/hour
by going 2-way.
============================================
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Change the way we count - numbers look better. Elapsed time is now
from when the driver signals the workers to start until the driver
detects that the last worker is finished. Also, machines have been
upgraded - so all comparison bets are off.
K8 4 way
5223.5 5859.6 5859.2 5819.4 5869.4 5822.6 5837.4 5830.2
5796.2 5822.7 5810.5 5765.2 5799.4 5818.3 5824.2 5821.7
K8 1 way
1408.3 1587.8 1588.0 1591.3 1584.9 1581.9 1593.9
============================================
Bryan (2/7/03) - New baseline numbers, obtained with the knightly/noDeb
build of 2/6/03. It seems that k4 is faster than k8.
K4 - 4-way
6217.8 6307.2 6302.2 6315.0 6314.2 6320.0 6296.8 6325.7
6253.6 6374.3 6346.5 6365.7 6356.5 6367.2 6360.3 6322.4
6192.2 6359.6 6347.7 6355.9 6360.5 6353.1 6308.6 6344.2
K4 - 1-way
1705.8 1705.6 1698.6 1703.2 1705.9 1691.5 1696.2 1706.0
K8 - 4-way
5829.2 5929.3 5925.0 5941.6 5902.4 5905.4 5860.7 5887.6