This should be below 5 seconds to minimize risk of losing rewards

I learned from a YT video that you need to keep an eye on debug.log to look `for these warnings:

2021-05-30T17:49:58.741 harvester chia.harvester.harvester: WARNING  Looking up qualities on /mnt/<snip>/plot-k32-2021-05-29-09-03-<snip>.plot took: 16.567266702651978. This should be below 5 seconds to minimize risk of losing rewards.

What I noticed is when I had the plotter wound out to 30 plots/day, I was getting these pretty regularly.

Now I’ve backed my Swar config back to 8 sim / 4 stage1 (from 12/6).

Hope things still stay in the 25/day+ range but need to run this for a few days to see.

So far in the last 6 hours I’m seeing a lot less of those warnings.

I can def see the point of having a separate farmer from the plotter :frowning:

I’m afraid I didn’t understand what you meant. I have not mind any Chia yet, so if that’s what you meant you are correct.

There’s a 30 second time limit to farm a block. This prevents folks from plotting after the fact, which would ruin the network but is currently impossible because it takes hours to plot which would fail this limit.

I usually get these warnings if I am copying plots onto a drive I’m already farming on, because the lookups are slow due to the copy load.

Unless your hard drive is having significant problems you should see these only rarely and it should never get close to 30 seconds.

Haven’t seen a value over 15 secs so far but I did read the 5 sec cutoff is something to really strive for.

I believe that this is going to become less of a problem in 1.1.7 - there is some parallelization being put into place that significantly reduces the search time.

2 Likes

Is that the reason? I read some Keybase responses that indicated it was to prevent people from using tape drive or very slow forms of mass storage?

1 Like

I doubt the answer for why 30 sec is this simple. Preventing “plotting on the fly” is certainly a goal but this could have been done with a much larger value than 30 sec and still covered future improvements in hardware. Was explicitly preventing tape drives a goal? 30 sec would not be needed to do this. The value needed to exclude tape drives is probably more like 5-10 minutes or higher. Seek times on tape are slow (secs or 10’s of sec). They are not intended to be used as random access devices. Seeking a tape repeatedly will wear it out. I would imagine it would only take a few days of continuous seeks to ruin a tape. Imagine loosing 200TB+ of plots when that tape dies.
My guess is that 30 sec is a result of the number of payouts per day that Bram was targeting in combination with how the consensus algorithm works.

1 Like

Yeah, it’s a curious choice of number, and 30 seconds is a bit aggressive given the variability of spinny rust hard drives… especially as farms get very large! On a giant farm the harvester might be checking, say, 30 passing plots across 30 different hard drives…

But I heard they are making improvements to parallelize the proof checks in 1.1.7 so hopefully that will help.

1 Like