Strategic Double Clicking

[ewmayer]

Of my latest batch of 8 strategic DCs on Ryzen, the following 3 mismatched the first-test result:
58569487
58651457
58517029
Two of the 3 mismatches had first-tests run by 'Judge Hale', 2 other first-tests by same user in same batch matched.

[GP2]

Originally Posted by ewmayer

Of my latest batch of 8 strategic DCs on Ryzen, the following 3 mismatched the first-test result:
58569487
58651457
58517029
Two of the 3 mismatches had first-tests run by 'Judge Hale', 2 other first-tests by same user in same batch matched.

Any indications of errors in your log files? If so, I'll take them and check interim results like last time, but if your runs were routine, then I'll leave them to Madpoo. He usually picks up triple checks within 24 hours (unless the anonymous churners get them first, but these exponents are far outside of the churn zone).

I have to be a little more parsimonious these days, because spot prices for c4.large have been edging up steadily from around 1.3 cents an hour a month ago to 1.7 cents now. Can't wait for the c5's to finally arrive.

The first-test machines in question have only a limited track record of results verified good or bad, but they might end up being in a grey zone where the error rate is way above average but still too low to justify systematically running DCs on everything. Those are the most frustrating kind.

[ewmayer]

Originally Posted by GP2

Any indications of errors in your log files? If so, I'll take them and check interim results like last time, but if your runs were routine, then I'll leave them to Madpoo. He usually picks up triple checks within 24 hours (unless the anonymous churners get them first, but these exponents are far outside of the churn zone).
-iner
I have to be a little more parsimonious these days, because spot prices for c4.large have been edging up steadily from around 1.3 cents an hour a month ago to 1.7 cents now. Can't wait for the c5's to finally arrive.

The first-test machines in question have only a limited track record of results verified good or bad, but they might end up being in a grey zone where the error rate is way above average but still too low to justify systematically running DCs on everything. Those are the most frustrating kind.

Good point - yes, all 3 of the above runs had a mix of expected (because close to the 3072K exponent limit) of ROEs up to 0.4375, plus at least a dozen of the catastrophic-ROE-followed-by-successful-interval-retry of the kind I spotted in my first runs on the Ryzen. Far fewer of the latter since David installed a water cooler on the system, and note the matching DCs in the same latest-batch had similar occurrences.

[ewmayer]

p.s. @Gord: If you do queue any of the 3 up, please make sure res64-printing is enabled for some multiple of 10Kiters, if possible - we can cross-check the every-1M-Res64s to check for divergences. Here are my 1Miter Res64 excerpt files:

[Madpoo]

Here's a list of top 100 "unknowns"... by that, I mean it's the smallest available exponent for any machine with no bad or good results yet, and I ordered the list in descending order of the # of unknown tests they've submitted.

I already skimmed the top 96 from the list for myself and these are the next 100.

It goes without saying (but I will anyway) that these exponents are very likely to match, but there's just no way of knowing until we do a DC for each CPU. Out of this list of 100 (and the 96 I took) I expect there'll be 6-10 mismatches, but then that gives us an opportunity to more closely examine those particular CPUs and the rest of their unknown results.

Enjoy.

Exponent	Unk	worktodo
51348503	49	DoubleCheck=51348503,73,1
52306921	49	DoubleCheck=52306921,73,1
47440243	48	DoubleCheck=47440243,72,1
48577201	48	DoubleCheck=48577201,72,1
51734191	48	DoubleCheck=51734191,73,1
51772601	48	DoubleCheck=51772601,73,1
55301227	48	DoubleCheck=55301227,73,1
55310261	48	DoubleCheck=55310261,73,1
59697343	48	DoubleCheck=59697343,73,1
48149501	47	DoubleCheck=48149501,72,1
48274627	47	DoubleCheck=48274627,72,1
51396391	47	DoubleCheck=51396391,73,1
52318787	47	DoubleCheck=52318787,73,1
59346977	47	DoubleCheck=59346977,73,1
59716109	47	DoubleCheck=59716109,73,1
47225333	46	DoubleCheck=47225333,72,1
53281387	46	DoubleCheck=53281387,73,1
57341029	46	DoubleCheck=57341029,73,1
59152091	46	DoubleCheck=59152091,73,1
59335007	46	DoubleCheck=59335007,73,1
59450137	46	DoubleCheck=59450137,73,1
47778431	45	DoubleCheck=47778431,72,1
52268147	45	DoubleCheck=52268147,73,1
59987717	45	DoubleCheck=59987717,73,1
50034679	44	DoubleCheck=50034679,73,1
50802509	44	DoubleCheck=50802509,73,1
51480197	44	DoubleCheck=51480197,73,1
51896479	44	DoubleCheck=51896479,73,1
52096019	44	DoubleCheck=52096019,73,1
54752591	44	DoubleCheck=54752591,73,1
57466309	44	DoubleCheck=57466309,73,1
57747979	44	DoubleCheck=57747979,73,1
51192539	43	DoubleCheck=51192539,73,1
48828289	42	DoubleCheck=48828289,72,1
51809033	42	DoubleCheck=51809033,73,1
52595617	42	DoubleCheck=52595617,73,1
55348021	42	DoubleCheck=55348021,73,1
57898807	42	DoubleCheck=57898807,73,1
58483847	42	DoubleCheck=58483847,73,1
59516917	42	DoubleCheck=59516917,73,1
47124983	41	DoubleCheck=47124983,72,1
48440479	41	DoubleCheck=48440479,72,1
51452213	41	DoubleCheck=51452213,73,1
52007789	41	DoubleCheck=52007789,73,1
56963989	41	DoubleCheck=56963989,74,1
59246713	41	DoubleCheck=59246713,73,1
47445667	40	DoubleCheck=47445667,72,1
48191909	40	DoubleCheck=48191909,72,1
48649201	40	DoubleCheck=48649201,72,1
51264649	40	DoubleCheck=51264649,73,1
51531119	40	DoubleCheck=51531119,73,1
51592679	40	DoubleCheck=51592679,73,1
51720037	40	DoubleCheck=51720037,73,1
51800569	40	DoubleCheck=51800569,73,1
52077679	40	DoubleCheck=52077679,73,1
47489479	39	DoubleCheck=47489479,72,1
51495679	39	DoubleCheck=51495679,73,1
51850237	39	DoubleCheck=51850237,73,1
52013141	39	DoubleCheck=52013141,73,1
52596121	39	DoubleCheck=52596121,73,1
52728727	39	DoubleCheck=52728727,73,1
48285961	38	DoubleCheck=48285961,72,1
51607687	38	DoubleCheck=51607687,73,1
51622553	38	DoubleCheck=51622553,73,1
51835547	38	DoubleCheck=51835547,73,1
51984901	38	DoubleCheck=51984901,73,1
52005763	38	DoubleCheck=52005763,73,1
52112639	38	DoubleCheck=52112639,73,1
52129087	38	DoubleCheck=52129087,73,1
52409297	38	DoubleCheck=52409297,73,1
53157737	38	DoubleCheck=53157737,73,1
53968877	38	DoubleCheck=53968877,73,1
48234181	37	DoubleCheck=48234181,72,1
48540889	37	DoubleCheck=48540889,72,1
49524823	37	DoubleCheck=49524823,72,1
49910737	37	DoubleCheck=49910737,72,1
51230551	37	DoubleCheck=51230551,73,1
51598891	37	DoubleCheck=51598891,73,1
51891487	37	DoubleCheck=51891487,73,1
51980053	37	DoubleCheck=51980053,73,1
52273597	37	DoubleCheck=52273597,73,1
52326199	37	DoubleCheck=52326199,73,1
52483889	37	DoubleCheck=52483889,73,1
52806571	37	DoubleCheck=52806571,73,1
52861967	37	DoubleCheck=52861967,73,1
52966877	37	DoubleCheck=52966877,73,1
54084379	37	DoubleCheck=54084379,73,1
55621201	37	DoubleCheck=55621201,73,1
48246743	36	DoubleCheck=48246743,72,1
48652831	36	DoubleCheck=48652831,72,1
51116383	36	DoubleCheck=51116383,73,1
51595897	36	DoubleCheck=51595897,73,1
51823129	36	DoubleCheck=51823129,73,1
51892961	36	DoubleCheck=51892961,73,1
51929609	36	DoubleCheck=51929609,73,1
51977287	36	DoubleCheck=51977287,73,1
52281431	36	DoubleCheck=52281431,73,1
53930671	36	DoubleCheck=53930671,73,1
53953309	36	DoubleCheck=53953309,73,1
54711067	36	DoubleCheck=54711067,73,1

[storm5510]

It seems like I saw this same list a while back. Is this a revision of some type?

[Madpoo]

Originally Posted by storm5510

It seems like I saw this same list a while back. Is this a revision of some type?

Probably just a refresh. I guess people weren't interested. I also prefer doing the double checks that are more likely to have been done wrong the first time... this list, not so much, but it's the best way of making sure these unknown system were okay or not.

I did come up with a slightly different approach for dealing with CPUs with no real track record, and that is to look at the track record of the user. If they have multiple computers then the track record of one won't really indicate how the other is doing (Curtis Cooper is a great example...some are rock solid, some are fluffy marshmallows). But there are times when it's the same computer but they reinstalled the OS or upgraded or whatever which in the data makes it look like a new machine.

If a user has generated a bunch of bad results and here's another of their systems with no record, that's a better candidate for checking a few out than just a random sample, I'm guessing.

I need to refine that process a bit though and then I'll create some lists using that method.

[storm5510]

Originally Posted by Madpoo

...If a user has generated a bunch of bad results and here's another of their systems with no record, that's a better candidate for checking a few out than just a random sample, I'm guessing.

I need to refine that process a bit though and then I'll create some lists using that method.

I took the last one in your list. Just to ease your mind, I just completed a different DC and the residue matched exactly.

I seem to remember something here about a different work type number and the criteria for anyone being able to do that work. Perhaps you could go that way. Just a thought.

[GP2]

Originally Posted by ewmayer

Of my latest batch of 8 strategic DCs on Ryzen, the following 3 mismatched the first-test result:
58569487
58651457
58517029
Two of the 3 mismatches had first-tests run by 'Judge Hale', 2 other first-tests by same user in same batch matched.

58569487 Judge Hale had the correct result, yours was wrong
58651457 Needs a quadruple check, which Madpoo has already queued up
58517029 You had the correct result, Judge Hale's was wrong

[ewmayer]

Originally Posted by GP2

58569487 Judge Hale had the correct result, yours was wrong
58651457 Needs a quadruple check, which Madpoo has already queued up
58517029 You had the correct result, Judge Hale's was wrong

Thanks, Gord - I note that none of my current batch of 8 DCs on the Ryzen has emitted any of the worrisome instant-fatal-ROE-followed-by-iteration-interval-retrys that afflicted my 3 runs above - hopefully that also implies that the system flakiness that seems to have hosed my run of at least the first of the above trio is no longer occurring. 4 of the current 8 ongoing runs will finish in the coming week, will alert if any mismatch the previous submission(s) for those.

[Madpoo]

Here's a list of "unknowns"

For this, I looked at systems with no bad or good, but the user has more bad than good as an aggregate across all their systems. There weren't many like that, just 45 systems.

These are the smallest available exponents from each of those along with the total # of unknowns they have so far. Even if there's only a single unknown (and this exponent is it), I figure there's at least a so-so chance it'll be done wrong if the user has a pretty poor track record anyway.

Enjoy!

Exponent	Unk	worktodo
41676091	1	DoubleCheck=41676091,72,1
42142421	1	DoubleCheck=42142421,72,1
43835563	1	DoubleCheck=43835563,72,1
43837823	2	DoubleCheck=43837823,72,1
44504917	3	DoubleCheck=44504917,72,1
45512063	2	DoubleCheck=45512063,72,1
48566747	1	DoubleCheck=48566747,72,1
48647167	1	DoubleCheck=48647167,72,1
49214119	29	DoubleCheck=49214119,72,1
49233293	28	DoubleCheck=49233293,72,1
51752761	8	DoubleCheck=51752761,73,1
53633119	2	DoubleCheck=53633119,73,1
53823769	2	DoubleCheck=53823769,73,1
53845861	3	DoubleCheck=53845861,73,1
53913649	2	DoubleCheck=53913649,73,1
54229709	4	DoubleCheck=54229709,73,1
55322647	3	DoubleCheck=55322647,73,1
56853239	4	DoubleCheck=56853239,73,1
57124861	4	DoubleCheck=57124861,73,1
57321079	1	DoubleCheck=57321079,73,1
57432247	4	DoubleCheck=57432247,73,1
57920869	6	DoubleCheck=57920869,73,1
60169433	2	DoubleCheck=60169433,74,1
60452207	1	DoubleCheck=60452207,74,1
60595879	3	DoubleCheck=60595879,74,1
60959519	6	DoubleCheck=60959519,74,1
61401101	1	DoubleCheck=61401101,74,1
61452109	12	DoubleCheck=61452109,74,1
61706009	29	DoubleCheck=61706009,74,1
61724309	26	DoubleCheck=61724309,74,1
61823791	2	DoubleCheck=61823791,74,1
63738713	4	DoubleCheck=63738713,74,1
65434531	2	DoubleCheck=65434531,74,1
65891149	2	DoubleCheck=65891149,74,1
68223187	16	DoubleCheck=68223187,75,1
69486343	3	DoubleCheck=69486343,75,1
69490037	1	DoubleCheck=69490037,74,1
70581211	12	DoubleCheck=70581211,75,1
72113731	4	DoubleCheck=72113731,75,1
72411901	1	DoubleCheck=72411901,75,1
72443809	1	DoubleCheck=72443809,75,1
73141063	1	DoubleCheck=73141063,75,1
73146581	4	DoubleCheck=73146581,75,1
78261119	1	DoubleCheck=78261119,75,1
80387807	1	DoubleCheck=80387807,75,1