Has anybody tried to use raid of fast 15k sas hdd to plot(as temp dir)?
Theoretically it looks good, but i cant find any comments, maybe I didn’t search well.
I can buy used lsi 9260 and 8 seagate savvio 15k 300gb for 220usd.
This array will have ~1400-1600 mb/s write speed, 4300 iops, and much more lifetime then any nvme.
Is it reasonable, or easier to find some good nvme? Thanks, and sorry if my eanglish is bad))
Brother, if this has a consistent yield over the speeds you are mentioning (considering there is no degradation in speed due to heat), you cannot find a better bang for your buck… I would say it is worth the venture, if it doesn’t perform to satisfaction then you can either trade it in or use it as plot storage - win win win…is how I look at at it… Although there are other more tech savvy people out here who should also weigh in…
Performance wise you’d have to test it, but my gut says latency might do you in if you are looking for plot speed (which means you bought into the narrative of ‘racing’ Chia which I think is a past relic but that’s another discussion).
Apart from that, 15K disk arrays typically need lots of cooling, and are noisy AF. And keep in mind you will need to replace some once in a while, so you;ld need a good cheap source for that.
I took the exact same approach. A used JBOD enclosure with a MegaRAID SAS 9260-4i (or 8i?) RAID Controller. I currently have 18x 2TB SATA Disks attached to it (couldnt get others at reasonable prices, and they will be converted to farmer disks when plotting is done). Each disk is presented as RAID0 to the OS (JBOD). Will start plotting this WE so I can then maybe share some numbers. Eventually will try it with RAID0 across 2x disks.
Try to run parallel 4-8 plots (1) all on single raid 0 and (2) on separate disks. This will show difference.
I bet on next config: 4 plots on separate RAID0 with four disks each and RAID0 of two disks as secondary temp )))
I think this is the best solution to plot in the long run… nvme turn to trash to soon and normal hdd are a bit too slow, for plotting is super ok imo, for farming way less because heat/noise and power usage
With slow sata drives, i think, it will works only if number of disks in raid0 is two times bigger then number of running plots.
And one of my nvme died in first day of plotting)) because of this, I started thinking about more durable solutions.
Big difference between sas raid and nvme is delay time, about 60ms raid and only 3ms nvme, and iops, 4k raid, and 300+k nvme.This matters with small data blocks. What data block size is used in chia?
My advice would be to use the disks but not use RAID. Just parallel mount points.
If a disk goes offline w JBOD and (on Linux at least) xfs or ext it’ll just hang the processes on that fs. Raid goes blooie, you lose everything. And if you lose a disk and it starts resyncing, the whole thing goes to hell throughput and IOPS wise.
And there’s no actual content so if a disk is lost you can just re-plot. In my first tests, RAID alone for plotting still could lag enough to be an issue for farming. I didn’t test exhaustively though and really try to tune.