Gigahorse farming weird pooling pattern

Chia Farming & Harvesting
1

Flexfarmer has got weird disk access bursts every 10s when harvester checks out plots.

I was told that vanilla Gigahorse node/farmer behaves like Chia official with silent disk access without raging disk heads finding something on disk.

It may also be caused by stripped JBOD because each part of the plot may be stripped on a different disk.

There is no change FlexFarmer or Gigahorse…every 10s raging disk access.

But there is an interesting pattern with pool points while running Gigahorse node :thinking:

2023-08-05_C5_diff20_gh
2023-08-05_C5_diff20_gh2374×896 237 KB

FlexPool diff 20, wild swings. chia plotnft show shows more submitted points. There is a warning not mess with diff. Sadly, payments didn’t reflect that either.

FlexPool diff 1, also weird in comparison to FlexFarmer bridge.

2023-08-06_C5_diff1_gh
2023-08-06_C5_diff1_gh2374×1172 240 KB

Also weird pattern. No idea what is going. FlexFarmer has got nice point distribution on same farm/plots.

Any suggestions @madMAx43v3r , please?

I have followed Gigahorse guide, no errors in log, all seems fine :thinking:

2

Why are your disks striped?

3

  1. in case of 20 TB, 4 disks JBOD is optimal use of space. 18TB is 8x JBOD.

  2. linear JBOD always fills 1,2,3,4,5,6,7,8 disk. Writing speed is always one disk.

  3. Stripped 4 disk JBOD = 500-1000MBs If your plotter spits one plot per 3 mins, you have an issue with linear JBOD

4

Bladebit/chia.
Runs without problems. And also free of charge.

I’ve been wondering for a while why gigahorse is preferred over bladebit(?).

5

One reason is that Chia compressed plots are still in beta, another is C7 only gives 30% more space, whilst GH C8 is 42% more space IIRC which easily off sets the fee. Another is that GH came months before Chia compressed plots.

I’m personally using Chia, replotted about a month ago to C7

6

Gigahorse is already here almost a half year. It is faster.

Only weird thing I experience is weird raging disk access bursts every 10s when harvester checks proofs.

7

Do you farm c7 with cpu or gpu?

What I understand mmx c5 is ~c7 Chia

I can run mmx c5 efficiently, it makes me wonder if Chia c7 will be same

8

it is pool inherent behaviour, no?

9

GPU 3060, it’s not a TI

10

its not what really matters, memory size does

11

My bet is more on compression side. Also I use stripped JBOD. Missing tables must be “decompressed”, which requires more disk access I guess.

What I understand, harvester checked tables before. Now, it needs to reconstruct tables first.

I have a weird feeling this compression stuff will kill HDDs prematurely. Copying data to my file server doesn’t make such a wild head movements

Perhaps in near future when ssd drops to same level as hdd…and we can buy 30tb sata ssd for same price as hdd now

12

Someone asked this question on the Chia Discord: (not sure why they linked to my post though)

Does compressed plots farming wears HDD more quickly? Are there more disk reads required due to decompression? I’m asking after reading this post: Gigahorse farming weird pooling pattern - #9 by Ronski

And got this answer, although not from Chia

Logically not, as it’s really not a compressed file, but bits are missing that have to be reconstructed within memory (either computer RAM if no GPU or in GPU VRam) to get back to what was stored before the days of compressed plots. So there’s actually less reads from the hard drive (less data stored, hence smaller files), but lots more recalculation (a portion of what old plot generation did before storing), hence why you need more CPU power or a GPU do do this re-generation of the missing pieces on the fly within memory.

13

Interesting. In this case, it has something to do with GH plots. My Chia farm is almost silent, while mmx plots drive HDD heads crazy every 10s