Right now I’m doing all my farming from a single farmer on my primary desktop machine with 12 drives attached (mix of internal and external). I’m looking to offload this to a dedicated hardware setup to free up some USB slots on my main machine along with giving the flexibility to deploy my farmer somewhere else with lower energy costs. I’ve scored some really good deals on some 2nd hand gear to potentially accommodate this but not sure which deployment option would be better. Here’s my inventory
Synology DS2413+ - 12 Bay Nas - Intel Atom D2700 - 4GB RAM
QNAP TS-453A - 4 Bay Nas - Intel N3160 1.6 GHz quad-core processor - 8GB RAM
Orico 4-Bay USB HardDrive Enclosure
6 External Harddrives, currently Unshucked
10 Internal Drives (4 unplotted thus far, came with the QNAP)
The big question in my mind is whether the DS2413+ has enough processing power long term to be my primary farmer (in which case I could resell the QNAP) as that would simplify the setup with everything running off that single unit.
If the DS2413 is likely to struggle then I guess the next most logical setup would be to run the QNAP as the primary farmer and then either run the DS2413 as a harvester or network share all the drives to the QNAP.
If you plan to use a NAS or similar appliance as a farmer (full node), here are potentially disqualifying factors:
need to host blockchain DB on a good dedicated SSD
no way to run GUI, so you need some kind of root access or docker runtime
If you plan to run the farmer on another machine and only run harvester on the NAS, that’s easier to set up, but then you have to be running a separate computer on the network, potentially negating any energy savings.
I personally ended up building a dedicated rig that runs full node and has disks directly attached through SAS interface. I have another desktop machine that’s plotting and running a harvester microservice while at it.
On the full node, the processing power is needed during initial syncing. It could be a significant bottleneck.
Also need to make sure NAS does not put HDDs to sleep, and does not use “energy savings” of some sorts on the network adapter (it would cause a lot of stales)
I would consider:
QNAP as full node.
RAID-1 system volume on SSD’s (2x500GB would be nice).
Install Virtualization Station and setup a Ubuntu Desktop VM with 6GB memory and 180 GB virtual disk on the SSD system volume.
Nice to have a full graphical OS with Chia GUI and other (OS) tooling for the full node, you can control the VM from your workstation with Windows Remote Desktop. Important to have the blockchain db on SSD, lots of issues for people having it running from HDD with syncing the db. Suppose you could backup the db from your desktop and restore on the VM somehow.
Stop and backup/snapshot the VM before any changes. In case of problems easy restore to the working situation including db.
Maybe a docker container on the QNAP harvesting on two single volumes created on HDD’s in the remaining 2 bays.
Synology with Docker, running a container harvester on the twelf basic (single) volumes holding plots.
Not sure how to test but I’d make sure that the ext4 formatted disk on the synology are at least readable from a bare metal linux install. Would be sad if the Synology died on you and left you with otherwise unreadable disks.
Raspberry Pi with Ubuntu Server and the Orica 4-bay as room to grow another 4 disks to harvest. You can use the Raspberry Pi to check if a drive formatted on the Synology and containing some plots are readable in the Orico unit. If not my plan goes down the drain I guess…
Use the 4 unplotted disks in the new setup first, copy the plots from your desktop connected drives, transfer thus emptied drive to the new setup and so on.
On this subject, I use a HP Microserver Gen8 as primary farmer for some 3700 plots on USB connected drive enclosures and a couple of harvesters. CPU is Celeron G1610T, 2.3 GHZ dual core so comparable to the NAS. No problems so far, load between ~25 to 50% farming. Close to 100% when syncing the db and somewhere between those when for instance chia plots check command.
Just think of all the people that are farming with a PI-4.
The PI-4 chip is quad-core compared to dual-core on the Atom and is 8 years newer. Wasn’t there also some issues during dust storm with weaker nodes not being able to win blocks?
need to host blockchain DB on a good dedicated SSD
Is an SSD required or is it just a nice to have for performance? I’ll copy the initial sync of the DB from my existing setup so ongoing syncing load should be minimal
no way to run GUI, so you need some kind of root access or docker runtime
That’s fine, I intend to run it via docker (should be more performant than spinning up a full ubuntu VM). I’m comfortable enough on the command line that it’s not a big issue
On this subject, I use a HP Microserver Gen8 as primary farmer for some 3700 plots on USB connected drive enclosures and a couple of harvesters. CPU is Celeron G1610T, 2.3 GHZ dual core so comparable to the NAS. No problems so far, load between ~25 to 50% farming. Close to 100% when syncing the db and somewhere between those when for instance chia plots check command.
Thanks for that info, really useful to see a comparable setup. The G1610T does appear to be a tad more performant than the atom but in a similar ballpark.
Not sure how to test but I’d make sure that the ext4 formatted disk on the synology are at least readable from a bare metal linux install.
Will definitely test the Synology EXT4 formatted disks on another system before completely migrating everything across. Definitely don’t want to end up in a situation with 12 disks worth of plots I can’t access.
Turn off the upnp setting and stop tcp8444 port-forwarding .They could cause a fullnode’s high cpu-usage .
Maybe you can try motherboard with alot of sata port and pcie2sata card.
Disk Stations have a internal boot chip to boot and download/install the main OS (DSM). DSM is installed in a (small) partition on each and every drive in the NAS so as long as any (one) drive is still in the NAS the full OS is available. Also user/config data are in that partition. Outside of the NAS this partition can be ignored, you’d just have to mount the partition the plots are on. But no harm in testing this!
Also I believe a special bit is set in the bootsector of each drive, designating it ‘part of a linux RAID’. This stops some OS’s from partitioning/formatting the drive. You have to dd/overwrite a first few sectors to get rid of this. But then, formatting the drive is exactly what’s unwanted
Just for fun decided to test it, had some time to spare;-)
Slapped a fresh drive in my DS118+, downloaded and installed DSM7, set up a share (plots) and copied a few .plot files to it.
Out of the DS118+ and connected by USB to my Ubuntu Desktop this is what I got (below).
I guess no real issue to mount and use SINGLE disks from the synology on a linux box.
Just a few extra synology specific files and folders on the data partition but the plots directory showed up no problem and can be added to the farmer/harvester plot directories.
A lot of random i/o and the database size is constantly growing. A HDD would struggle with it, so as would a QLC flash. For long term and better stability you’d need a TLC flash drive, nothing fancy - even a SATA would be fine (for example Intel S4610, S4620…) or if you have an M.2 slot, something like SK hynix Gold P31
Hello, some of my drives automatically hide from the farming machine. I have put the power option so that they should not sleep. Anyone know how I can take them back without unplugging and replugging them and how I can prevent them from being hidden?
Not sure what you mean by ‘hide’? What Operating System are you using on that machine?
What is your configuration, only a farmer or with remote harvesters?
