GEP Gold Certified Farming Specs

Approve Specs Gold Certified Farming

Dear community we would like to ask your approval for the definition of gold certified farming.

In order to grow utilization on the grid, we need to be known for being as reliable as any of the major cloud service providers.

In order to do that, we need to attract and retain the highest quality hardware and keep it connected with an SLA that matches AWS/Azure.

Gold Farming Qualification requirements:

  • Hardware purchased from a recognized vendor (approved by DAO - initially HPE & ThreeFold itself)
  • Minimum 5 IPv4 addresses per server
  • Two power supplies or PDU with 2 power feeds which automatically fails over
  • Two routers per rack (connections to internet)
  • At least Two Internet Service Provider connections
  • Tier 3 or 4 data center certified to ISO 27001
  • Uptime 99.8%
  • Network connection at least 1GBit/sec per TBD nr of CU/SU
  • Geographic decentralization - no more than one full datacenter rack per TBD Region size unless and until utilization in that rack is > 50%.
  • The 3Nodes need to be certified (can be done by TFTech or other actors on blockchain)

Farming reward simulator.

See HERE

The gold certified farming is updated in this calculator (50% more farming reward and better SU calculation price).

Please note that each Certified farm will have to be approved by the TFGRID DAO.

TODO: link to GEP to approve specs of new TFGRID DAO improvements

VOTING

  • OK with definition & specs of certified gold farming
  • Not OK

0 voters

Once poll approved the wiki and simulator on https://simulator.grid.tf/ will be adjusted.

2 Likes

That looks very good.

About this:

Network connection at least 1GBit/sec

Shouldn’t it be based on a given quantity of resources, in order to scale, instead of having one absolute number?

Say farmer A has a certified farm with 10 servers providing each X available resources, with 1GBit/sec.
Then comes farmer B who has 5 servers having each X available resources, shouldn’t 0.5GBit/sec be sufficient in this case?

Or the opposite, say farmer C has 20 servers providing each X available resources, shouldn’t farmer C then provide 2GBit/sec?

I hope I’m being clear.

Thanks!

Wow, amazing update. As someone who as spent the last weeks prepping for a move to a colo, I’m obviously very much on favor of it. I like the limit of one rack per town to avoid centralization. I’d be in favor of a higher connection speed, but a 10g connection is $4000 per month at my colo, so 1g is good.

1 Like

Detail: is it 2 power supplies per rack ? as in two routers per rack.

Thanks!

I don’t like this idea as limiting one rack per town is unfair to other farmers who join threefold late and have no chance of becoming gold certified. We want it to be fair to all farmers even the ones who join in late.

2 Likes

I interpreted it as one rack per farmer. But I may be wrong.

1 Like

2 power supplies per server, with 2 independent phases powering the rack.

2 Likes

I think you are right, @kristof can you please confirm?

1 Like

Amazing!! That’s exactly what we are looking for.

We are running a full rack in a DC and we allready fullfill all criterias except redundant router/ISP (we will do it immediately). Apart from ISO27001 the DC is also certified for EN50600. In addition the DC is completly powered by 100% renewable energy ressources (also certified).

Where can I sign up for this? :wink:

4 Likes

I do have one comment. 5 ip’s is going to be hard on 1U servers. They have 4 ports embedded and 2 pcie slots which are frequently occupied with SSD’s. I’ve typically installed NVME over SATA, needing pcie slots, to make the grid better.

2 Likes

Its one IP per MAC address, so one per port, not NIC.

1 Like

Though I’m recalling the recommended structure was to have one port for just boot download on the server. Which would only leave 3 ports, basically requiring another card anyways.

lets keep the comments coming please, I will try to adjust the proposal accordingly, not sure how to do with vote then? Hmmm how do we normally do something like this?

  • about 1 full rack, I suggest to change to certain amount of CU/SU, can anyone suggest amount please
  • please note, we can do much more than 1 rack but we need utilization of +50%, I think this is fair, because otherwise its not good for the planet. @noretreat is this ok?
    Also farmers can ask exceptions thats where the TFChain DAO comes in, so farmers who already got a lot of capacity just need to ask to DAO.
  • @FLnelson is a tricky one, per farmer means its very easy to fake, hmmm, not sure how to do this, anyone?
  • indeed agree with @Mik, should prob have link between CU/SU and required bandwidth, can anyone make suggestion please?
  • @weynandkuijpers maybe having 2 power feeds per rack is enough, but if servers only have 1 power supply they need to use power loadbalancing unit (like intellegent PDU) on top which can use both feeds. Ok?
  • @FLnelson it is not independent NIC’s, its just having enough public ip addresses, 5 per server is prob even at low side, maybe this needs to be more more, anyone?
3 Likes

indeed a NIC can have thousands of IP addresses (lucky)

I was under the incorrect impression that I could only assigned one public IP per MAC, but now that I think about it, that makes no sense since a server can host thousands of websites. Its not like every website out there has its own Ethernet port.

1 Like

Town is somewhat of a useless geographical term. Most towns are just vague unincorporated areas and vary between 1 and 10,000,000 people. “Turkey Town”, a real town near me, doesn’t need a rack, while Manhattan may need thousands of them. Some metric based on population density over a multi sq mile hex would be more useful. Tracking location is a another hoop to jump since as far as I know the IP only reports to the local ISP, which can be far away.

2 Likes

That’s not good enough since the PSU can fail. Only 2 PSU switches. If you are using a 1 PSU router, there should be two setup in HA mode with each power cord going to opposite power rail. No one has built up a large stash of 1 PSU rack mount servers yet so I don’t think that will get any pushback.

2 Likes

Here is what our setup looks like (don’t know if this is representative for what you’re looking for):

  • 1 rack = 46U
  • 8U needed for switches (3x), routers (2x), DC gateway (1x), cabel management (2x)
  • 38U left for 3nodes

server specs:

  • 1U rack space
  • 40 threads (2x CPU)
  • 320 GB memory
  • 4 TB SSD storage (1x drive)
  • 7 bays left for additional storage (X)

theoretical total capacity (38 nodes): 1520 vCPUs, 12160 GB RAM, 152 TB SSD storage + (X)

outdated!
In my opinion the certain amount of CU should not be set to high because there migth be other limiting conditions (apart from rack space) to be considered. For example in our case we can only use 33 nodes at the moment because of limited power supply and cooling capacity per rack. So actually we are running at 1312 vCPUs, 10496GB RAM and 138,8TB SSD storage.
outdated!

Edit: I forgot that we fixed that in the meantime by monitoring power consumption and heat emission via LOM and report this to the DC guys. The rack is fully piled up now running 38x 3nodes with 1512 vCPUs, 12096GB RAM and 158,8 TB SSD.

According to the result of the farming calculator the CU for this setup is approx. 3000.

Kristof is right that 1 per user would be easy to cheat and in reality its about fairness to the grid, not the provider. But since 1 per town should be revised to one per some ‘yet to be determined population density/area unit’ that would provide ample opportunities for additional racks. Otherwise someone will just call dibs on Chicago.

3 Likes

So, the racks Dany and I have each put together would not be eligible then?