How to optimise RAM in 1.29.1?

Since the 1.29 Hard fork, nodes that run on 8gb tend to choke every now and then and shutdown. With nodes that are being run on AWS the next logical upgrade is to 16gb. For nodes run on bare metal servers this is a one off upgrade that doesn’t cost too much. However this upgrade on AWS doubles the cost to run them! For a smaller pool this is terrifying. Is there anything that can be done to optimise the RAM usage on these nodes other than turning trace mempool off? Does anyone know if adding SWAP helps? Is it even possible to run nodes on 8gb anymore?

With K=500 soon to be K=1000 in the future, do you think IOHK envisioned that smaller pools wouldn’t be able to survive and the 2800 pools now would decrease down to these numbers? Cardano has always been “by design”.

Side question about topology configuration. How many relays do other SPO’s have? and are they used as back ups or are they also connected to their BP? Right now I have 4 relays connected to my BP that are dispersed in different AWS regions, but I am suspicious that the high latency might be causing the BP to miss slots every now and then.

IMO 16 Gb RAM is the minimum requirement for 1.29.

Indeed, turning off mempooltrace can help.

You can also set topologyUpdater to connect a maximum 5-6 peers.

You can disable monitoring tools (prometheus).

You can add 2-4 Gb swap, but your CPU will go crazy.

Side anwser, i only connect 2 relays on BP and i have a spare node not connected on BP. The spare node can be use as relay or BP.

Did u checked the contabo prices?

U have to many relays, in my opinion 2 are more than enough :wink: I made blocks with 1 single relay (the 2nd one it should be for redundancy in case one of relays is down )

So u can cancel 2 relays and upgrade to other relays

The reason for running 4 was to experiment with block propagation times, that being if I were to have one in America and Germany then the blocks I mint in Australia should propagate through the network faster and minimise slot height battles. However I’m not actually too sure if that has helped.

Check the peers from your relays… should be from anywhere around the world

Just a side note, I wouldn’t want to use 1.29.1 on mainnet yet as it’s not an official stable release. 1.29.0 is the latest.

Regarding relays, in case you have many, do not register them all, keep some private as a prevention for DDOS.

I wondered the same thing. Experimentation showed more local, or very close, relays are good. Remote (far away) are pretty much useless.

That having been demonstrated, we run with 2 local and one remote within “yellow” distance away.

Better to connect to nearest relays and let the neighbors do the work.

Some details in my blog (see profile).

1 Like