papnikol Posted May 4, 2018 Share Posted May 4, 2018 One of my arrays (12 data disks & 1 parity disk) is showing very slow and strange patterns of parity check. The setup is quite old and slow: Mobo: MSI RD-480 CPU: AMD Athlon 64 3700+ RAM: 3GB 1st SASLP (PCIe 8x) has 4 disks 2nd SASLP (PCIe 8x) has 6 disks Mobo controller has 3 disks The mobo has 2 16xPCIe slots but they are v1. Also, the mobo contollers support only SATA1. Still, the behavior I see when doing parity check seems strange. The array had 2x4TB disks and 11x2TB disks. I decided to replace most of them with larger 8TB disks and started by replacing the 4TB parity disk with an 8TB disk. The speed was slow but when the array reached the 2GB point (which means that only the 8TB & 4TB disks are reading/writing) I was expecting it to pick up. And yet, it is only ~45MB/s. The same was happening when I did a parity check with the 4TB disk as parity. The problem should not be the low RAM (according to the pic I attached). Nor should it be the bandwidth of the SASLP card (both drives are on an SASLP). It seems that somehow each disk's bandwidth cannot exceed the 50MB/s limit. This is the write speed for the parity drive during the whole parity check up to now: Can someone help me pinpoint the bottleneck? PS: sorry for the long post but I am trying to be precise. Link to comment
JorgeB Posted May 4, 2018 Share Posted May 4, 2018 5 minutes ago, papnikol said: MD Athlon 64 3700+ It's CPU limited, look at the CPU usage in the end, it's @ 100% Link to comment
papnikol Posted May 4, 2018 Author Share Posted May 4, 2018 6 minutes ago, johnnie.black said: It's CPU limited, look at the CPU usage in the end, it's @ 100% I though of it, but it seems strange. The CPU usage is 100% when there are only 2 disks involved in the parity check operation and it is lower when all 13 are reading/writing (reaching a collective read speed >500Mb/s. And it is a single core CPU but it seems abnormal for it to be limited at 50MB/s speed per disk. Link to comment
JorgeB Posted May 4, 2018 Share Posted May 4, 2018 . The CPU usage is 100% when there are only 2 disks involved in the parity check operation and it is lower when all 13 are reading/writing (reaching a collective read speed >500Mb/s. Yes, because it's going faster per disk, but it's very close (or already) CPU limited before. If the board supports it upgrading to an X2 AMD should provide a decent bump, but a dual core Sandy Bridge or newer would be much better. Link to comment
papnikol Posted May 4, 2018 Author Share Posted May 4, 2018 I see. Thank you for your input. I happen to have an Athlon X4 840 available (quad core, 3.1Mhz) and a mobo that supports it (Asrock FM2A88X Pro+, probably an unraiders dream since it has 8 SATA3 ports) . I was hoping to avoid the whole process but it seems I will have to give those a try.... Link to comment
Vr2Io Posted May 4, 2018 Share Posted May 4, 2018 If just parity check/sync, the memory usage shouldn't have any change. So pls check does some task run in same time. May be simple try perform parity job in safe/maintenance mode, 50MB/s speed for 8TB need 45hrs. Link to comment
papnikol Posted May 4, 2018 Author Share Posted May 4, 2018 16 minutes ago, Benson said: If just parity check/sync, the memory usage shouldn't have any change. So pls check does some task run in same time. May be simple try perform parity job in safe/maintenance mode, 50MB/s speed for 8TB need 45hrs. I think no other task is consuming significant power (I think mdrecoveryd and unraidd should be running at parity check): And, yes, it seems parity check will last 2 days :( Link to comment
EvilSpice Posted May 8, 2018 Share Posted May 8, 2018 @papnikol what are you using to generate these usage graphs? Link to comment
papnikol Posted May 8, 2018 Author Share Posted May 8, 2018 6 hours ago, EvilSpice said: @papnikol what are you using to generate these usage graphs? This plugin: Link to comment
Recommended Posts
Archived
This topic is now archived and is closed to further replies.