Follow-up to that *slow* RAID setup from a few weeks ago...
-
I mentioned in a thread a few weeks ago I had acquired a Mediasonic HFR2-SU3S2 PRORAID 4-bay external enclosure, and was dismayed at the performance (2.5MB/s over USB3, 5MB/s over eSATA) with a RAID-5 configuration. There was mention that RAID-5 has significant overhead over other types of RAID. Someone suggested to try RAID-10. I'm "losing" an additional drive worth of space (compared to RAID-5), but with RAID-10, as I'm writing this, performance is currently holding steady at ~100+ MB/s (vs 5, at best). It blows my mind that performance would be that much different. There's overhead, and then there's 20x slower throughput. The system is otherwise identical - same drives, same enclosure, same cables, same system it's connected to. This is with full-disk VeraCrypt encryption. Without encryption, RAID-10 showed spikes of up to 160MB/s (but generally holding steady at 120-130MB/s). I'm okay with that. If there's something else to blame, I'm not seeing it. But right now, I have little choice but to say RAID-5 is a killer when it comes to write operations. Again, I knew you don't get any of that for free, there's some overhead, but I never expected it to reach that level. If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now. And I'm a much happier camper. This is usable.
I just found this site: [RAID Performance Calculator - WintelGuy.com](https://wintelguy.com/raidperf.pl) I can't comment on its accuracy, but it suggests that a RAID-5 made of 4 drives each doing 170 MB/S (A WD-Blue SATA 4TB drive is rated up to 185 MB/s) and 90% write (assuming you're doing a backup ...) you should be seeing 183 MB/s for RAID-5 and 357 (!) MB/s for RAID 10. Now eSATA should be able to handle up to 600 MB/s, so something is not right. Does the manufacturer have a help site you could consult? Maybe a user/support group out there somewhere. Maybe it's just a sucky RAID implementation, all round. I'd at least check the manufacturers Web site to see if there's a firmware update that might solve things.
"A little song, a little dance, a little seltzer down your pants" Chuckles the clown
-
I just found this site: [RAID Performance Calculator - WintelGuy.com](https://wintelguy.com/raidperf.pl) I can't comment on its accuracy, but it suggests that a RAID-5 made of 4 drives each doing 170 MB/S (A WD-Blue SATA 4TB drive is rated up to 185 MB/s) and 90% write (assuming you're doing a backup ...) you should be seeing 183 MB/s for RAID-5 and 357 (!) MB/s for RAID 10. Now eSATA should be able to handle up to 600 MB/s, so something is not right. Does the manufacturer have a help site you could consult? Maybe a user/support group out there somewhere. Maybe it's just a sucky RAID implementation, all round. I'd at least check the manufacturers Web site to see if there's a firmware update that might solve things.
"A little song, a little dance, a little seltzer down your pants" Chuckles the clown
-
k5054 wrote:
Maybe it's just a sucky RAID implementation, all round
That's what I suspect. I'll have a look at their web site, but I'm always a little nervous about firmware updates.
-
What you currently have is only slightly better than a brick, so it doesn't seem like you'd loose all that much ;)
"A little song, a little dance, a little seltzer down your pants" Chuckles the clown
-
I mentioned in a thread a few weeks ago I had acquired a Mediasonic HFR2-SU3S2 PRORAID 4-bay external enclosure, and was dismayed at the performance (2.5MB/s over USB3, 5MB/s over eSATA) with a RAID-5 configuration. There was mention that RAID-5 has significant overhead over other types of RAID. Someone suggested to try RAID-10. I'm "losing" an additional drive worth of space (compared to RAID-5), but with RAID-10, as I'm writing this, performance is currently holding steady at ~100+ MB/s (vs 5, at best). It blows my mind that performance would be that much different. There's overhead, and then there's 20x slower throughput. The system is otherwise identical - same drives, same enclosure, same cables, same system it's connected to. This is with full-disk VeraCrypt encryption. Without encryption, RAID-10 showed spikes of up to 160MB/s (but generally holding steady at 120-130MB/s). I'm okay with that. If there's something else to blame, I'm not seeing it. But right now, I have little choice but to say RAID-5 is a killer when it comes to write operations. Again, I knew you don't get any of that for free, there's some overhead, but I never expected it to reach that level. If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now. And I'm a much happier camper. This is usable.
-
I'm the person who suggested going to RAID-10 for both safety and speed. The fact that RAID-5 was so slow tells me that there is no hardware support on that enclosure for the RAID-5 calculations and it was being done entirely in software.
-
I mentioned in a thread a few weeks ago I had acquired a Mediasonic HFR2-SU3S2 PRORAID 4-bay external enclosure, and was dismayed at the performance (2.5MB/s over USB3, 5MB/s over eSATA) with a RAID-5 configuration. There was mention that RAID-5 has significant overhead over other types of RAID. Someone suggested to try RAID-10. I'm "losing" an additional drive worth of space (compared to RAID-5), but with RAID-10, as I'm writing this, performance is currently holding steady at ~100+ MB/s (vs 5, at best). It blows my mind that performance would be that much different. There's overhead, and then there's 20x slower throughput. The system is otherwise identical - same drives, same enclosure, same cables, same system it's connected to. This is with full-disk VeraCrypt encryption. Without encryption, RAID-10 showed spikes of up to 160MB/s (but generally holding steady at 120-130MB/s). I'm okay with that. If there's something else to blame, I'm not seeing it. But right now, I have little choice but to say RAID-5 is a killer when it comes to write operations. Again, I knew you don't get any of that for free, there's some overhead, but I never expected it to reach that level. If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now. And I'm a much happier camper. This is usable.
dandy72 wrote:
If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now.
While RAID 10 sounds like the way to go, gotta at least give some love to RAID 5. Due to using the parity data on RAID 5 (which is what makes those writes slow) you can still rebuild while claiming more space when compared to RAID 5. Reads are still ok on 5, it's just the write speeds that suck. More times than not, you'll never need 5 for home use, but on those instances where space is more important than speed (for backups rarely accessed, etc.) it's nice to know it's there.
Jeremy Falcon
-
dandy72 wrote:
If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now.
While RAID 10 sounds like the way to go, gotta at least give some love to RAID 5. Due to using the parity data on RAID 5 (which is what makes those writes slow) you can still rebuild while claiming more space when compared to RAID 5. Reads are still ok on 5, it's just the write speeds that suck. More times than not, you'll never need 5 for home use, but on those instances where space is more important than speed (for backups rarely accessed, etc.) it's nice to know it's there.
Jeremy Falcon
I agree completely. I mentioned I initially thought RAID-5 provided the better capacity:redundancy ratio so I went for that. If performance wasn't in issue, that'd still be my first choice. I'm only intending to use the RAID as an extra backup. In other words, I'd regularly replace bits that have changed and write whatever is new. If all went well, in theory, I'd never need to read anything back. The only time it'd be in use is for write operations. And this backup set includes VMs that I need to keep running most of the time. The performance I was seeing with RAID-5 made that impossible.
-
I agree completely. I mentioned I initially thought RAID-5 provided the better capacity:redundancy ratio so I went for that. If performance wasn't in issue, that'd still be my first choice. I'm only intending to use the RAID as an extra backup. In other words, I'd regularly replace bits that have changed and write whatever is new. If all went well, in theory, I'd never need to read anything back. The only time it'd be in use is for write operations. And this backup set includes VMs that I need to keep running most of the time. The performance I was seeing with RAID-5 made that impossible.
dandy72 wrote:
The performance I was seeing with RAID-5 made that impossible.
Fo sho, I read obermd's post after mine, but he made a great point.
Jeremy Falcon
-
I'm guessing it was a poor RAID-5 implementation. Do note also that I've installed zero software for this, so the parity computation still has to be done by something in the enclosure, and not the CPU of the system it's connected to...
But that enclosure must have some CPU (and RAM). And hopefully a hardware RAID controller. If it doesn't have a hardware RAID controller, all the work has to be done by the CPU (of the enclosure), with decent amount of RAM.And for RAID5, A LOT of calculation has to be done for the striping. RAID5 is aimed at use cases where the fault tolerance is of higher importance than performance, though read performance should suffer far less than write performance, as the later has the highest amount of computation and possibly data reordering... So if your enclosure is CPU+RAM only, with both a slow CPU and little RAM, then it is quite logical that that it will show a subpar performance.
-
But that enclosure must have some CPU (and RAM). And hopefully a hardware RAID controller. If it doesn't have a hardware RAID controller, all the work has to be done by the CPU (of the enclosure), with decent amount of RAM.And for RAID5, A LOT of calculation has to be done for the striping. RAID5 is aimed at use cases where the fault tolerance is of higher importance than performance, though read performance should suffer far less than write performance, as the later has the highest amount of computation and possibly data reordering... So if your enclosure is CPU+RAM only, with both a slow CPU and little RAM, then it is quite logical that that it will show a subpar performance.
Agreed. My original post included the full make/model of the enclosure. I haven't come across anything that documents its internals however - not that I've tried very hard. But, it's there if someone's curious... [Edit] I just asked ChatGPT to describe how that works. It first suggested the host CPU was being used to calculate parity and such. I then pointed out I installed no software, so what process, exactly, is performing the calculations...to which it said I was right and it was wrong, and proceeded to explain that's done by the RAID controller (its exact words: "the actual parity calculations and RAID management occur within the enclosure itself, independent of any proprietary software on the host system.") Yeah, I'm not worried about these glorified chatbots taking over just yet. Problem is, the majority of people interacting with them just don't question its responses and just go with them...
-
Agreed. My original post included the full make/model of the enclosure. I haven't come across anything that documents its internals however - not that I've tried very hard. But, it's there if someone's curious... [Edit] I just asked ChatGPT to describe how that works. It first suggested the host CPU was being used to calculate parity and such. I then pointed out I installed no software, so what process, exactly, is performing the calculations...to which it said I was right and it was wrong, and proceeded to explain that's done by the RAID controller (its exact words: "the actual parity calculations and RAID management occur within the enclosure itself, independent of any proprietary software on the host system.") Yeah, I'm not worried about these glorified chatbots taking over just yet. Problem is, the majority of people interacting with them just don't question its responses and just go with them...
Well, the "RAID controller" of this box is certainly software based, as in software running on the CPU and the RAM integral to that box. And with a $160 price tag, this is certainly NOT a real hardware controller, those cost just by themselves as much. Looked at their manual, and it doesn't even remotely reveal anything about the internals. Looks fancy though... ;)
-
Well, the "RAID controller" of this box is certainly software based, as in software running on the CPU and the RAM integral to that box. And with a $160 price tag, this is certainly NOT a real hardware controller, those cost just by themselves as much. Looked at their manual, and it doesn't even remotely reveal anything about the internals. Looks fancy though... ;)
Ralf Quint wrote:
Looks fancy though...
I've got to admit, it's a nice-looking box. I had bad experiences with RAID over a decade ago, figured it couldn't still be so bad, so I went with this one. I'm okay with the performance (now that I'm using RAID-10) so I'll stick with it. But I wouldn't exactly rush to buy another one, or recommend it to non-technical users.
-
I mentioned in a thread a few weeks ago I had acquired a Mediasonic HFR2-SU3S2 PRORAID 4-bay external enclosure, and was dismayed at the performance (2.5MB/s over USB3, 5MB/s over eSATA) with a RAID-5 configuration. There was mention that RAID-5 has significant overhead over other types of RAID. Someone suggested to try RAID-10. I'm "losing" an additional drive worth of space (compared to RAID-5), but with RAID-10, as I'm writing this, performance is currently holding steady at ~100+ MB/s (vs 5, at best). It blows my mind that performance would be that much different. There's overhead, and then there's 20x slower throughput. The system is otherwise identical - same drives, same enclosure, same cables, same system it's connected to. This is with full-disk VeraCrypt encryption. Without encryption, RAID-10 showed spikes of up to 160MB/s (but generally holding steady at 120-130MB/s). I'm okay with that. If there's something else to blame, I'm not seeing it. But right now, I have little choice but to say RAID-5 is a killer when it comes to write operations. Again, I knew you don't get any of that for free, there's some overhead, but I never expected it to reach that level. If I've reached the wrong conclusion, then I'm wrong. I just see nothing else to blame right now. And I'm a much happier camper. This is usable.
Of course RAID-5 has terrible write performance. That's practically its purpose. I've seen SQL Server installations that had one gigantic (by the standards of the day) RAID-5 volume to hold everything. I was able to improve performance by a significant amount (2x?) just by fixing that stupid configuration. This was 20+ years ago so some of the details are a bit hazy.
-
Of course RAID-5 has terrible write performance. That's practically its purpose. I've seen SQL Server installations that had one gigantic (by the standards of the day) RAID-5 volume to hold everything. I was able to improve performance by a significant amount (2x?) just by fixing that stupid configuration. This was 20+ years ago so some of the details are a bit hazy.