HBA bandwidth required?

Chia Farming & Harvesting
1

Hello all,
I currently have 24,700 C7 plots, solo farming with an RTX3080 and 2x Intel 4-port SAS HBAs on Ubuntu. All 8 of those SAS ports are connected to several SAS expanders which then connect to 236 HDDs. My current average harvester latency is under 5 seconds.

What I’d like to do is downgrade my MB/CPU to a board with one PCIe X16 Gen4 for the GPU, and a PCIe X8 Gen4 for just one SAS HBA. I’ll have to connect additional SAS expanders as necessary.

Question is, will one SAS HBA cause my harvester latency to degrade beyond 8 seconds? I am unsure how much bandwidth is being used by both HBAs during farming.

2

I have 2 x HBA SAS2 + 2x expanders for 48 drives.

I had recently collision with smartd + chia + whatever runs in background. It completely killed IO, and crashed server.

if I recall the usual SAS2 HBAs have like 1000 IOPS, divided by so many drives. I would check out iotop, observe iowait and you will know bottlenecks.

I am runing juct 16k C4…used to be good on old Xeon, after filter halving I am getting too long proof times. Installed RTX 3060Ti, back below <1s with some +20Wh estimated penalty.

3

BTW what the hell do you have 1TB drives?

It should fit 1PiB and same amount of drives I have or even less with C7 compression

4

I have some 1TB, but the majority are 8TB and 12TB HDDs. The biggest is 24TB, the smallest is 1TB.
My SAS HBAs are RS3P4GF016J. HBAs do not have IOPS ratings, more like bandwidth based on total PCIe lane bandwidth.

5

yes, they usually hide the most important stuff :wink:

HDD is about 20-100 IOPS. SAS2 8i = 8*6gb = 48Gb theoretical

48Gb/236= ~40MBs I am guessing that would be not problem, but iops may be issue. If you are in Linux, you can check out what is happening now. Win should have some iops tools as well I guess

6

Doesn’t matter this PCIe x8 gen 4, cause majority of older HBAs are PCIe Gen 2.

For farming, GPU can be even on x1
I have almost 100 disks on old SAS1 HBA, and works OK.