Best crypto for severely underused ESXi boxes

[dwright1542]

I've got a mess of ESXi boxes with little CPU usage, so I might as well cryptomine, although I realized mining under ESXi isn't efficient. So a few questions:

1. Which crypto (figured monero / aero)
2. one giant machine with let's say 20CPU? Or a few smaller to keep NUMA pinned? Is it that much more efficient?

Power is a zero cost to me, but time is valuable, so I want quick and dirty. I don't want to "tinker" for hours.

 

[Patrick]

You are right on the efficiency. What I would do:

• VMs NUMA node pinned to all but one of the NUMA nodes. Keep one NUMA node clear for day jobs. You are likely in a situation where you have like 256GB of RAM, 80% used and 3% CPU utilization. That is normal.
• Put Ubuntu on the VMs
• Make a Docker swarm and run Portainer on one of the nodes.
• Tag each node in Docker with the CPU type and machine name.

From there, you can run XMR or AEON. Portainer will give you an easy GUI and that will let you manage your VMs easily. You can power up/ down the VMs individually. You can push new miners to all or some of the cluster.

It will take you an extra 5-10 minutes to setup the swarm, Portainer, and tagging (we have guides on STH), but from a long-term maintenance perspective when you want to switch from XMR to AEON or something new, it will save you lots of time.

 

[dwright1542]

You are right on the efficiency. What I would do:

• VMs NUMA node pinned to all but one of the NUMA nodes. Keep one NUMA node clear for day jobs. You are likely in a situation where you have like 256GB of RAM, 80% used and 3% CPU utilization. That is normal.
• Put Ubuntu on the VMs
• Make a Docker swarm and run Portainer on one of the nodes.
• Tag each node in Docker with the CPU type and machine name.

From there, you can run XMR or AEON. Portainer will give you an easy GUI and that will let you manage your VMs easily. You can power up/ down the VMs individually. You can push new miners to all or some of the cluster.

It will take you an extra 5-10 minutes to setup the swarm, Portainer, and tagging (we have guides on STH), but from a long-term maintenance perspective when you want to switch from XMR to AEON or something new, it will save you lots of time.

Ok that's a good start, here's some followups:

1. The machines are not in the same location / cluster. They are spread out quite a bit, so I'm not sure docker / portainer is applicable? Takes me 30 seconds to roll a VM out, so that's easy.
2. Take a Dual E5-2680's. Dual 14 core CPU's. I'd keep 4 cores just for doing Veeam-ish stuff(even though these are backup machines and have ZERO CPU usage). So should I Do 2x12 core VM's OR just one big 24CPU VM? WIll it really make much of a difference?
3. I've already noticed that if I don't use a 1:1 Thread to CPU ratio, the CPU's don't go 100%. I tried the L3 cache/2 method, but I was getting very low hash rates and low CPU usage.

 

[Patrick]

Maybe not, docker v. bare metal or VM differential is very small.

I might actually do 14 and 10 and NUMA pin. Let the Veeam sit on one CPU with the storage controller(s) and NICs.

Yea VM H/s is much harder. A virtual core does not necessarily equal a physical core. With Docker (not in a VM), a core is a core. In the environment, you may also have more sitting in L3 cache so you may not have full L3 cache available.

 

[dwright1542]

Maybe not, docker v. bare metal or VM differential is very small.

I might actually do 14 and 10 and NUMA pin. Let the Veeam sit on one CPU with the storage controller(s) and NICs.

Yea VM H/s is much harder. A virtual core does not necessarily equal a physical core. With Docker (not in a VM), a core is a core. In the environment, you may also have more sitting in L3 cache so you may not have full L3 cache available.

I'll give the NUMA a shot purely for interest sake when I have time.

I'm not being too picky about this....the machines I setup this morning in 15 mins are doing 5KH/s, which is 5KH/s more than yesterday, so that's a win.

Really appreciate your input.

 

[dwright1542]

Wow. Well NUMA affinity made a HUGE difference.

Single VM Non NUMA E5-2640's, 396H/sec.
Dual VM, pinned to each NUMA node, 310H/sec EACH.

That's a big deal.

 

[Patrick]

@dwright1542 Too much mining happening here