I think its ONLY a matter of time…
You will see…
I dont have anything agaist Gigahorse. I loved it.
Just, NOSSD is better. - my personal opinion.
At the current rate I have no doubt that NoSSD will exceed GH, but what happens when NoSSD controls more that 50% of net space? Do you think that’s a good thing for Chia, one farmer signing more than half the blocks, because its not?
I agree, its a far better compression method, but really each farmer should be signing the blocks, not NoSSD. They could have done this, but choose not to, why?
I have various mix from C12 to C15; I have 6.671 PiB left to plot.
1 Like
Nothing special, I think.
Based on that doc I linked previously, the threshold to be able to start a majority attack is at 33%. Although, the higher the percentage, the lower extra requirements to start such attack (less timelords, slower timelords).
Also, that 33% is not really a real borderline, rather below that percentage the requirements on timelords are quite heavy making it much more difficult to do such attack.
Well as long as NoSSD just continue farming as normal nothing will happen, BUT doesn’t the use case for Chia depend on it being non-centralised and therefore secure, so even if NoSSD just carried on as normal, because the possibility exists then Chia will look less attractive to end users of the actual blockchain.
The less attractive the Chia blockchain is, the less XCH will potentially be worth, so the more people using NoSSD could potentially devalue XCH.
I probably haven’t explained it very well but I’m sure people will get the idea.
I would really put “less attractive” / “decentralized” arguments to rest. What NoSSD does is they play according to the protocol that chia has, so they don’t really break anything to make it happen (e.g., two of us can partner, and run just 1 farmer and of course separate harvesters (say I am too dumb to run a full node), will that change anything? chia’s protocol enables that).
It is rather hard to imagine (as was already stated by others) that farmers should care much about how it is being done, rather than how much such farm is making and how easy the package is. If this is a concern on chia’s side, a hard fork with a change of protocol will be needed, and nothing we (farmers) can do about it. (If neither GH nor NoSSD will support CHIP22, a hard fork may be the only way to solve it.)
For me, I invested my money in all my H/W, and I am paying for the electricity, so I look at the bottom line, but if it goes nowhere, well life is tough.
On the other hand, I think (don’t know stock distribution) 2 people own most of the prefarm, and that is potentially worth north of $1b. Also, those people got around $70m to make it happen. So, they don’t really spend their money to make all this work and at the same time are looking forward for that prefarm to go sky-high. I will let them worry about their share first.
3 Likes
So are you saying you loooove nossd? Is that what you are trying to say ? Enquiring minds want to know !
1 Like
Actually, this is the tricky part.
Chia didn’t put much effort to provide a working payment solution in case some would like to provide a “better” module (whether node that doesn’t suck during dust storms or harvester that can handle non-standard plots). Therefore, when NoSSD came out, they had basically only two choices: 1. do what they did, 2. do what GH does. Their implementation is much simpler that GH, as all the farmer / node code is on their side, thus they don’t care much about chia updates. On the other hand, Max bought for himself a headache to update his GH basically for every / every other release of chia’s software.
What also needs to be considered is that NoSSD didn’t expect / want to be purchased by chia, thus the obvious model was exactly what they did. On the other hand, Max was betting that chia will purchase his solution, thus he needed to have basically a complete solution that would be ready day one (if purchased), so that dictated his choice.
Chia only started to work on CHIP22 when NoSSD became one of the top pools (again, that proposal should be out the day after NoSSD went live). As both NoSSD and Max got burned out during the initial negotiating phases, I kind of doubt that they will be willing to incorporate this proposal, making it kind of too little and too late to change the status quo. (Being in their shoes, I would not incorporate it.)
As I mentioned before, all this can still be easily fixed basically today by chia purchasing either GH or NoSSD and opening the code. If they would go for GH, they could easily have v2.1.5 that would include GH code (as it can farm non-compressed plots, BB plots, as well as GH plots). If they go for NoSSD, the integration will take a bit more resources, but should not be that difficult.
1 Like
There is a third option, NoSSD farmer signs blocks, but is still locked to NoSSD pool, farmer gets the 0.25, NoSSD distrubute the pools 1.75 less their fee. IIRC Max discounted this option as he didn’t want to run a pool.
2 Likes
Yes, that is GH FlexFarmer option. If you recall, chia initially bashed like hell on Flex having that solution, but finally quietly backed off. All three options are basically trying to solve the same problem - how to collect money for the service they provide.
From technology point of view, both GH and NoSSD need to do whatever is the fastest route for them (thus no point incorporating CHIP22). NoSSD was the pioneer, so they were betting on people using it, thus didn’t want to spend much time on a “proper” solution. Nothing really to blame them for.
A lot of people that use NoSSD are doing it, because there is no node needed to run it. Having GH FlexFarmer would be a really nice addition and potentially some of NoSSD farmers would jump the ship for it. From what I recall, Flex offered Max FF code to take over and as you said he didn’t want to go that route. I understand that Max doesn’t want to run the pool (or rather be long term involved with chia), but for instance Space is already running one, so having them taking over GH FF is kind of no brainer. Otherwise, they will be beaten to the pulp. The same goes for your FoxyPool (or any other pool), not sure why they are not interested in this solution especially as they already have a half-baked their own FP Farmer (maybe would give them a shot at Space pool).
2 Likes
@Anthony @Dawson
Ok guys, wheres the remote compute?
Cmon plz I have limited bandwidth on my home internet plan and burning 1000GB/month on upload over nfs isnt fun.
2000 plots c14 do about 200GB a month, so each plot does about 100MB/month
What if you have a 30k plot farm?
Thats more bandwidth than the HIGHEST PLAN THEY SELL in my area. I live in a first world country (USA) and in a city. 3TB is the highest plan available and that assumes I use literally zero internet for anything else!
Why are you uploading the plot files over internet? What is your set up?
Remote is a good idea since GPUs cannot fit into all the server cases and I hope to have it but not sure what you are trying to achieve by uploading plots over internet
Does the top miner have an 1 exabyte farm? This is an industrial scale deployment which would require custom software to manage and a datacenter to run.
At that level operating costs are very significant since you have to pay for the labor. How is this working for them?
The major cost is the capex (expense of the hardware) and then the opex (datacenter/power cost). You’d be surprised how little systems admin time is required if you use automation properly.
NoSSD isn’t complex at all and uses very little bandwidth.
I’ve been in the datacenter business for a very long time and around large storage deployments at the exabyte level. Running NoSSD compared to filecoin is a walk in the park.
Plotting at scale was the most “complex” part of scaling a NoSSD deployment and automating moving plots around to keep things in balance.
It really is a set it and forget it once you have everything running, and once a week the DC tech will swap out bad HDD"s as needed.
1 Like
A lot of data centre’s and cloud storage facilities etcetera have a lot of spare capacity, might as well fill it with plots and earn money until the storage is required for other purposes.
As @ExFilecoin states once the initial setup up is done its quite straightforward.
What are the automation software you would use at that scale? Would you rent the capacity (my rough calculation, you would need 100 rack and 400MW for 1EB) from an existing datacenter(s) or create your own datacenter?
I still see lots of labor after initial deployment. We have been replotting pretty much every 6 months. First, I had to replot all my farm to join pools, then for basic compression and now advanced compression. The system is hardly self running. Initially, plotting was heavy CPU usage and we were scrambling to find CPU/memory/NVMe but now it is primarily GPU driven so you need to bring GPU capacity to datacenter at scale. This installation must have 100 GPU; these won’t get installed without significant labor.
Btw, can someone confirm for 17 961 240 shares, do we need 1EB?
I am not uploading plots over the internet
My old eth miner (10 gpus) is perfect for mass compute so I mounted the storage server and kept getting out of traffic notices from my isp. I want to grow to a very large farm. 100k plots+
Each plot sends 100MB/month on average so no point in mounting over nfs when the data transfer is insanely high.
I tested MM’s gigahorse and that uses KILOBYTES for compute per plot per sinage, not MEGABYTES. Its more efficient by a factor of 10+.
oh never thought about that, this will get worse after doubling. I will start monitoring my network usage as well. Thanks for taking time to clarify it for a noob!
nossd is a lot better on pcie bandwidth than gigahorse is, by a lot.
You can no joke hook up a 4090 on pcie 2.0 1x and it wouldnt have FARMING performance effected (but PLOTTING performance would be horrible).
So you can buy one of these bad boys
And then some of the ewaste CMP gpus
And you have yourself something that can blast partials.
3x 30hx’s can farm about the same as a a4000 yet cost 1/3 to 1/4!
so I ask again, Dawson and Anthony, can you PLEASE release remote compute.
do like --rc 8989 where it listens on 0.0.0.0:8989 for compute and sends the partials your way or let us specifiy address with --rc “127.0.0.1:8989” in case we use some vpn system between the computers.
A really good compute machine is a gaming laptop. Comes w/ a free UPS battery backup (its a laptop)
1 Like