Which parameters of the node should be adjusted to increase the drawing speed?

Support
1

When I was using 1 node and 1 farm, the drawing speed was very fast. However, when I increased the number of farms to 10, that is, 1 node with 10 farms, each farm being 7T, the drawing speed for all the farms seemed to decrease significantly. I suspect the bottleneck might be caused by the node. Can I change some connection quantity to increase the drawing speed? Here is my node configuration:

      [
        "--chain", "gemini-3f",
        "--base-path", "/var/subspace",
        "--execution", "wasm",
        "--blocks-pruning", "256",
        "--state-pruning", "archive",
        "--port", "30334",
        "--dsn-listen-on", "/ip4/0.0.0.0/tcp/30434",
        "--rpc-cors", "all",
        "--rpc-methods", "unsafe",
        "--rpc-external",
        "--rpc-port", "5846",
        "--rpc-max-connections", "2000",
        "--in-peers", "500",
        "--out-peers", "500",
        "--in-peers-light", "500",
        "--discover-local",
        "--dsn-out-connections", "700",
        "--dsn-pending-in-connections", "400",
        "--dsn-enable-private-ips",
        "--validator",
        "--name", "ooplay-node-2"
      ]
2

There are about 5 configurations related to the number of connections. I want to know which specific parameter should be adjusted to improve the drawing speed?

3

None of them should be related to farming speed. The numbers you put there are not very healthy and not necessary either. I’d suggest to remove everything related to connections that you customized. The only thing you may have to change is --rpc-max-connections, but only if farmer errors during start, which for 10 farms I believe shouldn’t be the case.

On the node side check per-thread CPU usage, unless you see one of the threads at 100% utilization all the time, it is not node that is the issue.

For farms make sure you use one farm per physical disk, don’t split one disk into multiple farms, it will only make things worse and can impact farming success.

Check what is your CPU usage on farmers and if it is not too high run farmer with RUST_LOG=info,subspace_farmer=debug for 2 minutes and send logs here (use pastebin.com or gist.github.com since files can’t be attached here).

4

I set up two 7T hard drives in a farm, will this have an impact?

5

I hope they are SSDs and not hard drives. Also I’m not sure what do you mean exactly by “set up two 7T hard drives in a farm”, can you provide an example of the command you use for to start a farmer?

6

Within a farm, two specific physical SSD drives have been designated.

7 
"path=/var/subspace1,size=1.71TiB",
        "path=/var/subspace2,size=1.71TiB",
        "path=/var/subspace3,size=1.71TiB",
        "path=/var/subspace4,size=1.71TiB",
        "path=/var/subspace5,size=1.71TiB",
        "path=/var/subspace6,size=1.71TiB",
        "path=/var/subspace7,size=1.71TiB",
        "path=/var/subspace8,size=1.71TiB"

The physical SSD is 2*7T

8

This is not the intended setup. If you have 7T SSD it should be one 7T farm. If you have two of those thentehre should be two 7T farms, not 8 small farms. RAID and such setups will likely increase CPU usage and reduce farming speed.

9

Is there no problem running two 7Ts ? in one application, or do I have to open two applications?

10

You should use one application, there should never be a need to run more than one application unless you have more than 256 farms 65TiB each. Two 7T farms should be no problem, but let me know if there are improvements to make.

11

You previously mentioned that plotting speed is related to nodes. If I have too many SSDs, does it mean I need to create multiple nodes on one computer?

12

Where did I mention that? It is not the case at all, one node should be sufficient.

13

But after a few days of testing, if it’s 1 node to 1 farm, then the plotting speed is very fast, like 10% in one day. However, if it’s 1 node to 7 farms, then the plotting speed is very slow, with each of the 7 farms only having about 2-3% per day.

14

I already explained above the correct setup, looks like you’re doing the opposite and wondering why it is slower that way :man_shrugging:

15

I am using an enterprise grade router and my node configuration has always been like this, during testing I only operated farm, my node has never been shut down.

16

My 7TB farm’s plotting progress is approximately 1% per day. This is extremely slow. If it continues at this pace, I will need 100 days to complete the plotting. I’m uncertain if this is due to a CPU problem or because of the network conditions in China. Here are the debug data I collected. Is there a problem with the CPU?

17

I see nothing wrong with logs, but plotting is clearly not fast. There aren’t many things here, so if CPU usage is high (should be 100% with 3 farms) it is CPU-bound, otherwise it is having a hard time retrieving data from the network.

18

EPYC 7702 CPU, only 7% used

19

7% of that CPU is till a lot, check per-core CPU usage

20

https://raw.githubusercontent.com/qingyuhuaming/Su-b/main/farm.log

This is a new log, and I want to know exactly where the problem originated. Under completely identical configurations, my other CPU, which is a 7003, can plot 7% in one day for mining. However, this 7002 can only plot 2% per day. The performance gap between these two CPU models should not be this large, so I’m wondering if there are other issues.