5 ms·
two cents basing on similar experience: 1. those disk might be SMR. there's a high chance they are. that might make performance worse. 2. those disk support s
by znpy 5y ago
two cents basing on similar experience:
1. those disk might be SMR. there's a high chance they are. that might make performance worse.
2. those disk support sata 3 @ 6.0 gbps, but the network interface on the odroid (as well as most other SBCs) is a 1gbps. not surprisingly, the graph for throughput kinda reflects that: the best performance for network-based sequential read (jiva) tops at about 1/6 of local physical volum device. iops for sequential reads are similar.
Uh, i've said this many times, i'll say it again... Single board computers are very nice for small embedded project or very small server needs.
But if you want anything performant, they're pretty much all crap.
The author of the article doesn't seem to be aware of this kind of stuff either:
> I haven’t tried to tune the performance of Jiva and cStor pools, so most probably I could get more performance out of those disks after playing with some parameters.
According to this seagate spec sheet (https://www.seagate.com/www-content/datasheets/pdfs/3-5-barracudaDS1900-11-1806GB-en_GB.pdf https://www.seagate.com/www-content/datasheets/pdfs/3-5-barr...) the disks they are using (ST2000DM008) tops at 220 MB/s.
BUT on the physical volume benchmark, they only get 130 MB/s at most.
So... Yeah, mostly wasted money. Nice hobby I guess, but a waste of money, performance-wise.
- trojan295 5y ago> 2. those disk support sata 3 @ 6.0 gbps, but the network interface on the odroid (as well as most other SBCs) is a 1gbps. not surprisingly, the graph for throughput kinda reflects that: the best performance for network-based sequential read (jiva) tops at about 1/6 of local physical volum device. iops for sequential reads are similar. Jiva for sequential reads was 20MB/s, so the 1gbps network port wasn't saturated at that point.
- znpy 5y agoIndeed, indeed. What I wanted to say is that even if Jiva had no performance penalty, it would only deliver at most 125 out of the 220 MBps that the drive is supposed to be capable of sustaining.
- trojan295 5y agoOh yeah, I agree here. That's why I think it would be nice to try the Mayastor engine, which, based on some benchmarks, provides much better performance. I'm curious, if still the OpenEBS would be the bottleneck, or maybe I hit the SBC limitations. Unfortunately, I'm not sure if they are planning official ARM64 support...