diamondsw

joined 11 months ago
[–] diamondsw@alien.top 1 points 9 months ago

Fastest "normal" way would be write it to a USB 3 drive on machine 1, read from drive later on machine 2. An SSD would be best. Even with having to do the write and read in two phases, I'd still expect it to be faster than a 1G network.

(Less normal ways would be things like Thundrerbolt SSDs, 10G or 40G ethernet for direct transfer, etc. But if you have that kind of tech already in place, then you're not asking. ;)

[–] diamondsw@alien.top 1 points 9 months ago

Mine is sadly over an order of magnitude higher than this...

[–] diamondsw@alien.top 1 points 9 months ago

Running out of space will cause problems with anything. I just went to edit a config file on my Docker VM, and not only would it not write the changes, but it somehow deleted the original file at the same time! Yep, root volume was full (I really need better monitoring/alerting...), but after cleaning out some unneeded Docker build cache I was able to restore the file from yesterday's backup and all was well.

Moral of the story - cloud isn't the problem, it's lack of storage. It's always lack of storage. ;)

[–] diamondsw@alien.top 1 points 9 months ago

I am really struggling to parse this. What?

[–] diamondsw@alien.top 1 points 10 months ago

Hard drives are for capacity. SSD's are for performance. This has been settled for a number of years, and is why you see multiple levels of caching in front of any modern enterprise storage system.

[–] diamondsw@alien.top 1 points 10 months ago (2 children)

If you need IOPS, you need SSDs. The days of getting IOPS from multiple hard disks have been over for a decade.

[–] diamondsw@alien.top 1 points 10 months ago

Shelves use almost no power; everything is in the drives themselves. I have a 60-bay JBOD that uses ~40W empty. That's for massive 1400W 208V-only PSUs, backplanes, etc. That said, the ~15 drives I have installed use quite a bit more, so I leave it off 95% of the time. Just fire it up, run a backup to it, shut it down.

[–] diamondsw@alien.top 1 points 10 months ago

Costco had 14TB for $150. Sit tight through the weekend and see what pops.

[–] diamondsw@alien.top 1 points 10 months ago

As others have implied, you need to have something run on startup that will execute that IPMI tweak again.

In my case ESXi auto-starts my VMs, one of which is a docker VM and it auto starts a container that does the IPMI tweak (and monitors temps and adjusts and all that).

You don’t need to layer it this deep, but fundamentally you need things to auto-start on boot, including this.

[–] diamondsw@alien.top 1 points 10 months ago

It's not Thunderbolt - it's just a USB disk enclosure with the cheapest possible 2.5" hard drive - yes, a spinning rust hard drive. The only thing you get over something $100 less is it stacks nicely with the Mini.

I'd say hard pass. What might work better is an external powered enclosure for the M.2's, as that would reduce the power budget your Mini is trying to satisfy.

[–] diamondsw@alien.top 2 points 10 months ago

That's essentially what I do. Bunch of drives, each formatted individually, then pooled with MergerFS. Get a new drive, add to the pool. If a drive fails, it only loses that much of my backup; rest of the disks are unaffected. All the advantages of storage pooling without the redundancy (which here is useful).

[–] diamondsw@alien.top 0 points 10 months ago (2 children)

Best you’re probably going to be looking at is a Tesla card of some variant, and in a 1U without external power support it’s not going to be beefy. You’re not getting anywhere close to an RTX anything in that server.

view more: next ›