Notifications
Clear all
Functionality Issues
18
Posts
2
Users
3
Reactions
1,095
Views
Topic starter
I just ran a certify on a ThunderBlade x8 16TB (8x2TB), starting on all blades simultaneously. The certify passed, but two of the blades (disk8 and disk11) finished around 75 minutes earlier than the other six. Any explanation?
2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk4, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk5, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk6, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk7, SoftRAID ID: XXX, PCI bus 0, id 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk8, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk9, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk10, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 10:17:07 - SoftRAID Application: Certifying the disk disk11, SoftRAID ID: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt). with 3 passes and 15 minutes of random access testing. During each pass, every sector on the disk is filled with a pattern. Then the pattern is read back and verified. 2024.09.07 - 13:04:19 - SoftRAID Monitor: Starting SMART test on all disks which support SMART. 2024.09.07 - 13:04:19 - SoftRAID Monitor: Finished SMART test on all disks. No disks failed the SMART test. 2024.09.07 - 17:27:33 - SoftRAID Application: The certify disk command for disk disk11, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 17:27:50 - SoftRAID Application: The certify disk command for disk disk8, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:43:50 - SoftRAID Application: The certify disk command for disk disk6, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:44:10 - SoftRAID Application: The certify disk command for disk disk4, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:44:27 - SoftRAID Application: The certify disk command for disk disk5, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:44:33 - SoftRAID Application: The certify disk command for disk disk7, SN: XXX, PCI bus 0, id 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:44:37 - SoftRAID Application: The certify disk command for disk disk9, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully. 2024.09.07 - 18:44:52 - SoftRAID Application: The certify disk command for disk disk10, SN: XXX, PCI bus 0, id 0, lun 0 (Thunderbolt) completed successfully.
Posted : 07/09/2024 8:15 pm
No real explanation, it can happen. Certify is just a read/write process, and will go as fast as the blades can individually perform reads/writes.
Posted : 08/09/2024 6:49 am
Topic starter
No real explanation, it can happen. Certify is just a read/write process, and will go as fast as the blades can individually perform reads/writes.
Is it necessarily the case that two of my disks are much faster than the others (with those 2 disks having remarkably similar results, and the other 6 disks also having remarkably similar results)? I notice the same speed delta between them and the others in Verify operations as well. My suspicion was some kind of bus priority thing.
Perhaps worth noting: the two fast disks also happen to be those mapped to LED 1 and LED 2.
This post was modified 2 years ago by astrostl
Posted : 08/09/2024 10:10 am
Only if your enclosure is the Flex8 or something where the first two slots are faster to accommodate U2 devices. The Thunderbay 8, all slots are the same speed.
Posted : 08/09/2024 1:13 pm
Topic starter
If I verify all 8 at once, drives 1 and 2 are far ahead (offsets ~50M as opposed to offsets ~30M in the same elapsed time). If I verify just 1 and 8, they're nearly the same (offsets ~30M). I think that eliminates drive speed as an explanation and can point only to internal bus bandwidth with the earlier devices getting clear priority when saturated?
Posted : 08/09/2024 2:25 pm
Sorry, you did say that. I can inquire about this, there should be no difference between different PCI lanes.
Posted : 08/09/2024 10:32 pm
Topic starter
Sorry, you did say that. I can inquire about this, there should be no difference between different PCI lanes.
No apology needed! My hypothesis is not that there are consistently faster lanes (this seems disproven by testing), but that there can be bus congestion and prioritization when whole device is jamming. I'll be interested to hear whatever explains the results I'm seeing, but it doesn't make the device unusable or problematic.
Posted : 08/09/2024 10:52 pm
There is actually an answer. It has to do with how IO's are queued with MacOS. If two complete at the same time, disk1 gets priority over disk2, so this reduces some latency on the first two drives.
Posted : 09/09/2024 7:07 pm
Topic starter
There is actually an answer. It has to do with how IO's are queued with MacOS. If two complete at the same time, disk1 gets priority over disk2, so this reduces some latency on the first two drives.
I'm not exactly sure how either of those follow. I always get 100% performance out of disk1 and disk2 in particular no matter how many disks are active, and disks 3-8 drop down to 50% performance when the enclosure is saturated. So I couldn't say it's a disk1 > disk2 thing, or that the difference is merely "some" latency.
Posted : 09/09/2024 9:07 pm
This is how it was explained to me and something we have seen since Firewire.
If you certify 6 disks only, I believe you will see the same behavior. (3-8)
Posted : 09/09/2024 11:57 pm
Topic starter
Okay, here's a grand experiment. I did a verify of drives 1-4, 1-5, 1-6, 1-7, 1-8, and 3-8. Three performance profiles emerge (color highlights are mine): green == a verify around 50m, yellow == a verify around 70m, and red == a verify around 100m.
It's clear that fewer disks results in higher performance, but there are themes and variations: disks 1 & 2 are always green no matter how many drives are verifying, and 1-6 (six disks) gets better results than 3-8 (also six disks!).
Posted : 10/09/2024 7:58 pm
Topic starter
Still present on 8.3 FYI.
Posted : 17/09/2024 9:06 am
Page 1 / 2
Next
