If done correctly, you should be able to set up Windows (or Ubuntu) on the servers and network share the SAS drives on both. Some would recommend sharing all the drives individually and some would recommend sharing the drives as a RAID stripe. I use my Windows 2019 NAS much the same way but I have 3 drives in a RAID 0 (stripe) and a single drive. The single drive is a temp drive where Chia puts all the plots and the RAID 0 is where Chia is doing all the farming from. I suggest starting with 1 server, the 360p G8 1st and fill it up then the move to the other server then add or upgrade drives as you wish.
You could also just use the servers to plot on the hdd’s directly. It’s slower than using pc+nvme but you might as well put those cpu’s to work. One plot per hdd works fine.
The use the pc to plot some extra. You could setup shared network folders. I havn’t tried it but I don;t see why that should be a problem.
I would use each server to plot to 4-6 hdd’s and then use the pc to plot the rest and move the plots over the network.
Some people have had good results putting 2xhdd in raid for plotting. The only thing you really don’t want to do is have your farm drives with the plots on them in raid0. If one of the drives fails, you lose all plots.
To add to what I said originally, you can use the PC to plot away and store the plots over the network to the 1 server. The 2nd server can be plotting away 1 plot per HDD till the 1st server is full then switch and make the 2nd server a NAS. I envy you for having the ability to have those 2 servers vs what I have.
Let me try to understand one plot for each hdd.
In windows when try create one plot on chia blockchain software it ask me for one temp folder and on plot folder.
In first setup with only 1st server with 12 drives how you do it?
I can make raid 0 for first to drives 1,and 2 and leave 10 drives individually.
After I use the first drive as temp plot, but how i can have only one temp folder and add the the rest of 10 drives for final plot?
I just meant use the hdd as both temp and destination, just making a folder /plots/ or something on each hdd as the final directory.
That way you can use it for plotting until it is almost full, then just copy a few around and fill the last spaces…
But I don’t know if you have the option to divide the hdds over the two systems, but at least you can swap the cpu’s around to have the faster one in the system with more hdd’s.
My idea is:
360P: just run 4 plots parallel (with some stagger maybe), 1 to each hdd until they are full
380P:
use 6 drives in the same way as above (ideally after swapping the cpu’s).
share the other 6 drives as network drives.
PC, plot on the nvme and use the network drives on the 380p as destinations.
with Swar (plotman to i think) you can set it to fill the drives equally by alternating the destinations and skipping full drives.
I think that is a pretty good way to use everything you have together.
Edit: for the plots running on a hdd, you also want to set the second temp to the same folder as the destination. Otherwise it will still copy the file from one folder to the other, even though they are on the same drive.
Edit: for the plots running on a hdd, you also want to set the second temp to the same folder as the destination. Otherwise it will still copy the file from one folder to the other, even though they are on the same drive.
I can define the same folder for temp files and for final files?
temp; where the plotting happens with a whole bunch of temp files
second temp; in phase 3, the compressed file will be written to this destination, this is just one file close to being finished.
final destination; where the final plot will be copied to.
If second and final folder are the same, then there is no need to copy because it is already there, so it will save you that time for every plot (about 15-20 minutes)
You would really would not need to do anything like what you are doing here.
Drives c: , d: , e: , f: ,
What I would recommend is:
temp folder 1 c:\temp
temp folder 2 d:\temp
temp folder 3 e:\temp
temp folder 4 f:\temp
final destination for all folder c:\ final_all
In order for the c:\ drive to work though, you would need to stagger it out so like at least an hour or more starting your plots. If you do more than 4 drives, you are better off having a dedicated storage drive for your plots.
Or you can do a 5th hard drive g:\ final_all. You have the number of drives to do that.
If you are using your 380P 12 drive server as a NAS farm, then what you will want to do is periodically pull your plots off of the 360p and move them to the 380p. You can automate robocopy or powershell scripts to move your plots to your NAS. For me, I have the plots going on a timed queue but manually move the plots over the network because I don’t trust scripts (even though I use them at work). OR:
Better yet…
temp folder 1 c:\temp
temp folder 2 d:\temp
temp folder 3 e:\temp
temp folder 4 f:\temp
Final destination for ALL folder: Plot folder on NAS
But for config 380p as nas server I have 3 options raid 0 i have 24 tb space, but if one drive fails everything goes down.
Raid 10 secure but only 12tb
or maybe the best Raid 6 with 20tb available and 2 drive failure support
I live life on the edge and have a low end Marvell raid that only does raid 0, 1 and 10. So I am stuck doing raid 0 on my 4 bay HP Microserver G10 AMD.
Raid 6 is nice, but I see no need for 2 drive failure protection. Unless your drives will be pounded and your drives are really really old and have an obscene amount of read/write cycles on them. Raid 5 should be fine as in my experience, 2 drives never die at the same time. So… Raid 5 + hot swap is perfectly fine and allows for faster rebuilds and you regain that storage space back.
I have buy this servers used for my testing servers on my office and i dont know how many reads and writes this drives have.
Raid 5 support 22 tb. Is fine
You can use Crystal Disk Info to check disk health to get that info. Crystal Disk Mark will benchmark individual disks as well. My experience suggests Raid 5 is most common and every major company uses it exclusively nowadays unless Raid 10 or raid 1 is needed.
I would do Raid 5 on your NAS and go with that. Let us know how that turns out and your progress. I’m curious.
You have a ton of different option with these three machines, you’ll just have to choose what seems best or most practical to you.
My thinking is just to use as much of the plotting power available, and keeping the setup as simple (least point of failure) as possible.
I have to go to my office where servers are located, maybe during weekend for i can install windows and change raids 10 to 5 for have more storage available.
In the begining i was thinking Use temp and final folder on same drives for maximize the amount of space.
So let me know the process, first chia plots on temp folder and after moves the plot to destination folder. what happens to temp file are moved to destination right by chia software or I have to move it to nas by hand?
After more reading I think this way
PC with nvme and core i7 make the plots and move to NAS.
360p with 4 tb drives with 2 raid zero for faster write performance and ploting.
380p as nas server.
Good idea. Just make sure the NVME is at least 2TB, so you generate enough plots. Also make sure you have enough RAM to sustain the plotting otherwise you’ll kill your speed. Don’t do what I did and use 2x 512 NVMEs. I ended up having to RAID my NVMEs (benefit is I get a slight speed boost and the writes are split between both NVMEs but the downside is I have to use RAID 0).
On your NAS server, you can set that up any way you want. Per disk is probably the way to go for simplicity sake and skip RAID and just activate/share a few disks at a time, unless you want to do RAID 5 or something and you want that added layer of security.