VPS farming problems

have 8/30/1.2 vps. Tried plotting different sizes, changing ulimit, monitored node and server. 8 weeks no rewards at all.
I'd like to look deeper into what exactly is preventing farming on VPS. 

No matter how shared it’s resources may be, it must work. And there are cases someone receives rewards on a VPS, even including on Contabo!
But there are interesting article on their site and i cite em, "Plotting is the initial phase in proof-of-storage approach…bla-blah… The process is so intensive that the disks run off warranty and often even break completely after 1-2 months of use. Breaking an SSD used as shared storage could result in a disruption for other users on a given host system, that is why plotting is strictly prohibited on VPS or Virtual Dedicated Servers… "
Obviously they consider any PoSt bc as Chia, without even looking at the protocol, which says, “…unlike in Chia, plotting does not destroy SSD or HDD and is up to 10x faster…”. But I don wanna ask them what they are doing to (allegedly) protect their SSDs , before you…
Could someone share where the catch is?

Subspace is not behaving the same way Chia does when it comes to SSD writes, quite the opposite actually.

Chia derives special proof of space tables during plotting. Ideally, you’d use RAM for them, but since it is A LOT of RAM that is required, many farmers used SSDs as a replacement for that, which meant a lot of writes were done to SSD, eventually destroying them in some cases.

Subspace does use Chia proof of space as an internal component, but for a different purpose. Due to smaller value of K (17 on Gemini 3f and 20 on upcoming network it fits in RAM completely. For context Chia uses 32 and even increase by 1 approximately doubles the amount of space/RAM required.

Plotting in Subspace happens in sectors, each sector ~1G in size. Sectors are built in-memory and then written to disk as a whole. So instead of many random reads/writes/overrides, plotting is just writing sequential sectors, 1G at a time one after another. Like if you have copied a big movie filling the whole disk start to end.

Once plotting is done there is replotting that kicks in periodically, it’ll override select sectors with new sectors after expiration. In this case it is the same deal: old 1G sector is overridden sequentially with new 1G sector. ~50% of sectors expire every time history doubles, which means it both happens infrequently and less frequently over time as blockchain history increases.

So as far as farmer is concerned there is minimal write amplification due to the write approach and not as many writes generally as in Chia due to completely different protocol.

Eventually during duration of any individual network you’ll override the whole plot a few times, which even on the worst SSDs with low endurance (like 1000 overrides during lifetime) should not cause any issues whatsoever. Moreover, this is happening in SSD-friendly way, where random big sectors are overridden and not the same ones over and over again, which helps SSD to balance wear of cells well over time.

And farming is just random reads, which SSDs are very well optimized for.

I’d say Subspace farming is designed to be very friendly and gentle with SSDs, if you observe different behavior in practice, let as us know because it must be a bug somewhere.

Thank you for the detailed and very informative reply. But what would you advise to do to make it possible to make money on VPS Contabo for farming in Subspace?
I could try to petition them to reconsider their attitude towards our network

We never recommended to buy hardware to farm Subspace, it will likely result in negative ROI.

It is up to you to communicate with VPS provider if you choose to go that route regardless, I’m just providing information about how protocol works, which may or may not be useful in communication with Contabo or whoever else it might be.

I apologize, I failed to explain my question:
What do you think could be the reasons why rewards might not be coming while farmig with VPS’?

(aside of limitations that can be done on the hosting side)

If they limit IOPS it will impact farming. You can follow Oct. 3 windows release not signing rewards - #6 by nazar-pc and see if your audit times are slow as well (read that thread for details, there are some others as well).

I once asked GPT about BTRFS w/o data CoW vs XFS. Among other things, I got an answer,

" If your workload involves a lot of small random writes or requires high write performance, disabling CoW in Btrfs may provide some performance benefits. However, if your workload primarily involves large sequential reads or requires high read performance, XFS may be a better choice.

While I don’t have a direct source to cite for that specific statement, it is a general understanding based on the characteristics of the file systems and the impact of CoW on write performance…"

It’d be great if someone could make any comments about it

We have a note about CoW in official docs already. CoW is useless for Subspace, but Subspace farmer is not doing a lot of random writes anyway (node does). Subspace farmer is doing a lot of random reads, but I’m not sure if there is any SSD that would suffer on any file system to be honest.

Can you run programs like ASS SSD on your VPS to test if your IOPS are up to standard?

I think you need at least 2000 IOPS