Topic: OFFICIAL CGMINER mining software thread for linux/win/osx/mips/arm/r-pi 4.11.0 - page 433. (Read 5805983 times)
November 28, 2012, 02:06:05 AM
Where does the name 'cgminer' come from? Are you French ckolivas?
November 27, 2012, 07:53:27 PM
Was this with getwork, gbt or stratum? One thing about share submission for non-stratum on cgminer - it will keep retrying to submit if for some reason the submission failed (server responded with empty reply, busy etc.) until it has deemed there is no chance it is still valid. It retries sending every 5 seconds. Perhaps that is what's going on here? cgminer certainly does throw out all work on a longpoll equivalent, and checks if it should on every "cycle" through the work in GPUs at least.
November 27, 2012, 07:40:59 PM
US2 is having some trouble, I'm looking into it now. US3 and US1 are fine though, it's kind of strange. Definitely good to see things get sorted out.
Can you comment on that by chance?
Quote
This should get things working in the new block case of cgminer, but the fact that cgminer was still working on an old job after 24 seconds (at 800+ Mh/s) means it isn't switching to the new job when giving devices more work right away, so there's still a (cgminer) bug there (althrough the reduced 55 second update period will likely avoid that being as common).
Can you comment on that by chance?
Dropping scantime to 80 and expiry to 100 seconds did not fix it on Stratum. Since I should have a 120 second window to submit shares and the oldest shares I could be trying to submit (short of an unintended bug) are 100 seconds old I see no way to actually have run afoul of this reject overnight. My stales seemed just as high this morning and I noted some of the same invalid work type stales as I previously named. The rate despite only working on a work unit for (it appears) the higher of server or scan time and expiry set low should have fixed it if the issue was cgminer. Well unless somehow I was actually getting work results far in excess of the time alloted but even there I would have 2-3 results that met difficulty and 1 would be rejected sometimes the last sometimes the first usually the middle. Timing is either because its an old submission (expiry should have dropped it) or its being
rejected out of hand.
With the upgrade (on Eclipse) I am restarting cgminer to get my stats in line with only the new setup.
EDIT: added On Stratum to 1st sentance for clarity.
November 27, 2012, 06:11:18 PM
US2 is having some trouble, I'm looking into it now. US3 and US1 are fine though, it's kind of strange. Definitely good to see things get sorted out.
Can you comment on that by chance?
Quote
This should get things working in the new block case of cgminer, but the fact that cgminer was still working on an old job after 24 seconds (at 800+ Mh/s) means it isn't switching to the new job when giving devices more work right away, so there's still a (cgminer) bug there (althrough the reduced 55 second update period will likely avoid that being as common).
Can you comment on that by chance?
November 27, 2012, 06:08:38 PM
All servers are updated on EMC. I'll be curious to see how it works.
Great!Trying it out now on stratum. Will report back.
Accepted shares: 723
Rejected shares: 1
I watched diff go from .99 to 2 and fluctuate frequently during that time and a handful of block changes went by as well.
I happened to see the rejected share as well, and it was a diff 1/2 high hash reject.
This means diff had risen, and cgminer knew about the rise, but being conservative as per a scenario A described here: https://bitcointalksearch.org/topic/m.1357099 cgminer still submitted it.
All in all, great stuff. I'm glad to see this issue sorted out for both the pool and the miners.
November 27, 2012, 05:26:42 PM
All servers are updated on EMC. I'll be curious to see how it works.
November 27, 2012, 02:34:26 PM
Not sure this is intended or not : using cgminer 2.9.5 under windows7 (64), downloaded from http://ck.kolivas.org/apps/cgminer/cgminer-2.9.5-win32.7z while connected to a GBT-enabled pool, the "Q" value reported by cgminer stays at "3" and won't change, even after hours of running.
This gives giant efficiency values (which may alter the various load-balance algorithms ?)
Part of my conf file:
EDIT : downgrading to 2.9.4 fixes that "issue" ; "Q" indicator increases again.
This gives giant efficiency values (which may alter the various load-balance algorithms ?)
Part of my conf file:
Code:
"failover-only" : true,
"submit-stale" : false,
"intensity" : "d",
"gpu-threads" : "1",
"gpu-dyninterval" : "7",
"expiry" : "110",
"scan-time" : "60",
"log" : "1",
"queue" : "1",
EDIT : downgrading to 2.9.4 fixes that "issue" ; "Q" indicator increases again.
November 27, 2012, 12:40:26 PM
Well, if you can't roll ntime (sounds like a major flaw to me)
Not a flaw - just not needed.
November 27, 2012, 08:42:42 AM
With the clarifications from slush, I've fixed 2 bugs in eloipool's stratum server:
This should get things working in the new block case of cgminer, but the fact that cgminer was still working on an old job after 24 seconds (at 800+ Mh/s) means it isn't switching to the new job when giving devices more work right away, so there's still a (cgminer) bug there (althrough the reduced 55 second update period will likely avoid that being as common).
Note that these changes won't affect EMC until Inaba updates it there.
- clean_jobs is now sent true when previous jobs are immediately invalid
- idle job updates occur every 55 seconds instead of 96 (stratum expects a 60 second grace window on old jobs)
This should get things working in the new block case of cgminer, but the fact that cgminer was still working on an old job after 24 seconds (at 800+ Mh/s) means it isn't switching to the new job when giving devices more work right away, so there's still a (cgminer) bug there (althrough the reduced 55 second update period will likely avoid that being as common).
Note that these changes won't affect EMC until Inaba updates it there.
November 27, 2012, 08:10:49 AM
Looks like there's discrepancy between your and mine expectations about the ideal protocol. I've been focused to minimalistic design, and the lack of any extended features is benefit for me. I'm open to designing some (optional) calls like "report me your stale ratio", which will do the same job, it allows handy graphics on pool website, but it doesn't make the protocol core more complicated.
November 27, 2012, 07:58:03 AM
How do pools say "previous job remain valid for N more seconds"?
Is there any use case for this feature? Except pools which "pay for stales" (=pay bad miners on behalf good ones).Quote
This was an important fix added to getwork longpolling
Is there any pool actually using it? As we're primary talking about bitcoin mining, there's no motivation to work with stale shares on pool side... November 27, 2012, 07:55:27 AM
How do pools say "previous job remain valid for N more seconds"?
Is there any use case for this feature? Except pools which "pay for stales" (=pay bad miners on behalf good ones).
Quote
This was an important fix added to getwork longpolling
Is there any pool actually using it? As we're primary talking about bitcoin mining, there's no economic motivation to work with stale shares on pool side...
Quote
avoid losing merged mining shares in some cases.
Well, this is valid point. We can discuss if losing less than 0.1% of merged mining shares is worth of complicating the mining protocol. I'm personally strictly for "keep it simple" and minimalistic design, otherwise implementations became monsters full of checking edge cases. Actually there's only one condition in whole Stratum protocol, and it is (I think necessary) "clean_jobs" flag. The rest is plain simple algorithm without exceptions.
November 27, 2012, 07:04:40 AM
Ya my first several shares are usually rejects. I am mining over wifi, with about a dozen+ other computers on the network, so DCs are a fact of life. Still, I can live with 0.3% for now. 
Don't go further. This is the issue.
I've read a certain number of posts from Kano speaking of these problems with Stratum now. If I believe Kano, it looks like the solution are somewhat trivial to implement, by just modifying the specifications we would be good to go. In 2 months from now on, Stratum will be the de facto standard protocol (because GBT has too much stales), now is a good time to correct forgotten cases/issues by Slush, not in 2 months.
Ask a pool to implement it fixed so they can call it something else (since it isn't what slush wants he has no claim over it) and then all the other pools can see it's better and switch over to it.
If it goes on for much longer I'll try to push this.
LOL at above post, he's getting lonely with his GBT and BarbieMiner that's people are slowly realising what's wrong with it and no one wants
November 27, 2012, 06:57:18 AM
Stratum jobs can be used infinitely, until the block is invalid or the pool expires the job. Unlike getwork, there is no 232 nonce limit.
True. But there is a limit to ntime. If a pool were to hash a stratum job ad nauseam and finally solve it, and then submit it, it would be rejected based on ntime (being some large time into the future). Stratum doesn't roll ntime and thus depends on the pool pushing new work requests.
Perhaps slush could clarify the current meaning
I don't understand what is unclear here.a) clean_jobs=False means that previous jobs are still valid.
b) clean_jobs=True means that previous jobs become invalid.
c) If new work is received, miner automatically should use this job. There's no reason why to add a new flag to specify this, it is "by default". However miners can do whatever they want, because with or without a special flag, previous jobs are still valid.
KISS, please.
How do pools say "previous jobs shouldn't be used, but please submit them anyway"? This was an important fix added to getwork longpolling to correct pool-side statistics as well as avoid losing merged mining shares in some cases.
I've read a certain number of posts from Kano speaking of these problems with Stratum now.
Kano's problem is just one of many...If I believe Kano,
Sounds like a pretty bad idea.(because GBT has too much stales)
This is either a myth or a cgminer bug. November 27, 2012, 06:32:38 AM
now is a good time to correct forgotten cases/issues by Slush, not in 2 months.
Actually I was thinking about reconnection pretty well, unfortunately I didn't find any proper solution yet. Resuming mining on another TCP connection goes against the KISS of Stratum and it requires advanced methods of load balancing.
That say, when you open two connections to my pool, you will most likely talk to completely different backends. Synchronizing them online so the second one will be able to accept and validate shares generated from another connection is really non-trivial issue...
November 27, 2012, 06:21:51 AM
Ya my first several shares are usually rejects. I am mining over wifi, with about a dozen+ other computers on the network, so DCs are a fact of life. Still, I can live with 0.3% for now.
Don't go further. This is the issue.I've read a certain number of posts from Kano speaking of these problems with Stratum now. If I believe Kano, it looks like the solution are somewhat trivial to implement, by just modifying the specifications we would be good to go. In 2 months from now on, Stratum will be the de facto standard protocol (because GBT has too much stales), now is a good time to correct forgotten cases/issues by Slush, not in 2 months.
November 27, 2012, 05:56:25 AM
Just adding my 0.0016 BTC worth here..
Since p2pool has gone berserk, I have half my miners pointing at EMC, and half at Oz.
Oz is using stratum.
EMC is using GBT.
After about 12 hours, there were noticeably more rejects on EMC than on Oz. That's with using GBT, not stratum, on Oz.
Not sure if it means anything.
M
Since p2pool has gone berserk, I have half my miners pointing at EMC, and half at Oz.
Oz is using stratum.
EMC is using GBT.
After about 12 hours, there were noticeably more rejects on EMC than on Oz. That's with using GBT, not stratum, on Oz.
Not sure if it means anything.
M
To give some numbers. After about 10 hours of mining, 2 identical rigs (3x7970 ~ 2GH/s):
oz:
Code:
cgminer version 2.9.5 - Started: [2012-11-26 19:03:48]
--------------------------------------------------------------------------------
(5s):2.002G (avg):1.997Gh/s | Q:1398 A:18114 R:14 HW:0 E:1296% U:28.0/m
TQ: 7 ST: 0 SS: 0 DW: 2574 NB: 75 LW: 20980 GF: 0 RF: 0 WU: 28.0
Connected to stratum.ozco.in with stratum as user
Block: 048858edfdf0a45c5fe39cb2... Started: [05:49:39] Best share: 8.65K
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0: 73.0C 2787RPM | 666.9M/667.3Mh/s | A:6035 R:4 HW:0 U: 9.32/m I: 9
GPU 1: 73.0C 2908RPM | 666.6M/667.5Mh/s | A:6040 R:5 HW:0 U: 9.32/m I: 9
GPU 2: 73.0C 3197RPM | 662.3M/662.7Mh/s | A:6041 R:5 HW:0 U: 9.33/m I: 9
--------------------------------------------------------------------------------
eclipse:
Code:
cgminer version 2.9.5 - Started: [2012-11-26 19:04:25]
--------------------------------------------------------------------------------
(5s):1.958G (avg):1.956Gh/s | Q:1782 A:13009 R:53 HW:0 E:730% U:20.1/m
TQ: 0 ST: 11 SS: 0 DW: 3946 NB: 75 LW: 21963 GF: 0 RF: 0 WU: 27.5
Connected to us1.eclipsemc.com with GBT as user
Block: 048858edfdf0a45c5fe39cb2... Started: [05:49:41] Best share: 572K
--------------------------------------------------------------------------------
[P]ool management [G]PU management [S]ettings [D]isplay options [Q]uit
GPU 0: 73.0C 2789RPM | 653.0M/655.3Mh/s | A:4348 R:21 HW:0 U: 6.71/m I: 9
GPU 1: 73.0C 2606RPM | 654.5M/654.4Mh/s | A:4320 R:10 HW:0 U: 6.67/m I: 9
GPU 2: 73.0C 2406RPM | 645.0M/646.7Mh/s | A:4341 R:22 HW:0 U: 6.70/m I: 9
--------------------------------------------------------------------------------
That's 0.07% rejects on Oz and 0.4% on Eclipse. Note that's fewer blocks submitted on eclipse because it is serving me somewhere between 1 and 2 difficulty shares, whereas Oz is 1 (even though I have it on var diff, 2gh must not be enough for auto to raise it).
M
November 27, 2012, 04:53:28 AM
Perhaps slush could clarify the current meaning
I don't understand what is unclear here.
a) clean_jobs=False means that previous jobs are still valid.
b) clean_jobs=True means that previous jobs become invalid.
c) If new work is received, miner automatically should use this job. There's no reason why to add a new flag to specify this, it is "by default". However miners can do whatever they want, because with or without a special flag, previous jobs are still valid.
KISS, please.
November 27, 2012, 02:28:04 AM
Ya my first several shares are usually rejects. I am mining over wifi, with about a dozen+ other computers on the network, so DCs are a fact of life. Still, I can live with 0.3% for now.
Don't go further. This is the issue. November 27, 2012, 02:10:30 AM
Stratum jobs can be used infinitely, until the block is invalid or the pool expires the job. Unlike getwork, there is no 232 nonce limit.
True. But there is a limit to ntime. If a pool were to hash a stratum job ad nauseam and finally solve it, and then submit it, it would be rejected based on ntime (being some large time into the future). Stratum doesn't roll ntime and thus depends on the pool pushing new work requests.
Jump to: