Copy process to array takes ages, high CPU load. Why my unraid is that slow?

[richwood110]

Hi!

Unraid is new for me, I'm still on trail license. I setup Unraid 6.12.13 on the following system:

Gigabyte Z97X-Gaming 3, Intel i5-4690K, 16 GB DDR3 RAM, 4x HDD, 1 GBit LAN.

 

Currently I'm only running and array with 3HDD, 1x parity HDD and no cache drive. Everything is fresh installed and the drives are empty. Currently I only want to use Unraid as NAS network share.

 

When I copy data to the share (via Windows 11 client samba share) the transfer rate starts at ~100 MB/sec. The CPU load is around 20% (according the dashboard). Around 2mins later the transfer rate drops significant down to below 10MB/sec and it even comes to stop. The CPU load on 1-3 CPU cores reaches 100% (it varies). Finally the complete system stucks, samba share disconnects, Dashboard reponds slow etc.

 

If I stop the copy process, the CPU load is back to idle (~7%).

 

Questions:

 

1. Is the used i5-4690K to slow for this purpose?

2. How to fix this slowness issues?

 

What I already tried, all with no help:

- rebooting the unraid

- disabling the (not used) docker service

- restarted my Windows 11 client

 

Currently I only want to use unraid as a NAS. This system worked fine for me as a homeserver with Windows 10 + Drive Bender. But due to lost support of Drive Bender, I wanted to give Unraid a chance to replace this but currently the system is not useable as it is now, which is very sad.

 

Please find attached some screenshots showing the transfer rate drop and also the diagnostics.

Any help greatly appreciated!

Thanks!

Richwood

 

 

 

media-diagnostics-20240919-1505.zip

[dlandon]

Turn off Jumbo frames on your NICs.  It is not necessary these days and offers no speed improvements.

[richwood110]

I just checked, on my Windows 11 client, this option was not active. How to check if this is active on my homeserver?

[dlandon]

This is in your log:

Sep 19 08:50:08 Media kernel: r8169 0000:05:00.0 eth1: RTL8125B, 00:e0:4c:01:2a:19, XID 641, IRQ 33
Sep 19 08:50:08 Media kernel: r8169 0000:05:00.0 eth1: jumbo features [frames: 9194 bytes, tx checksumming: ko]

I don't have any Realtek NICs, so I can't tell you how to fix this.

[richwood110]

Ok, tested the following:

- re-checked Jumbo package setting on the Realtek LAN on my Windows 11 client (activated and deactivated again)

- removed the 2nd Realtek LAN card (2.5GBit) from server (currently not used)

- rebooted the server & windows 11 client

- started copy process again (freefilesync on Win11 client)

 

First 13min looked promising (~20MB/sec) but then shortly after, it droped again

See screenshots and new diagnosys...

 

media-diagnostics-20240919-1655.zip

[JonathanM]

Install this and run the benchmarks, both drive and controller.

 

[richwood110]

The result looks normal for me. Find it attached. Then benchmark finished without any issue. I also made a screenshot of the CPU load during the benchmark.

Thank you for any tipps.

[richwood110]

Hi! Just a short update:
Today the copy process works normal, without high CPU load and without interupts. No idea why, but when it stays like this, I'm happy 🙂

 

 

Nothing changed on my setup compared to yesterday...

[richwood110]

Sadly this is still a mystery for me. I still have this issue! The copy speed constantly drops when the CPU load hits 100%. I started copying 3,7 TB from external HDD which is attached to my unraid server. There is no other operation on the server (it just runs the docker Cloud Commander for copying). The server currently runs 24h/7d to finish this job (I'm still trying to fill my array with my data). If I'm lucky I get transfered 200 GB per day!

 

If anybody has any further advise how to check what exactly is causing this high CPU load, I would be very happy!

 

Thanks in advance!

 

media-diagnostics-20240924-1708.zip

[JorgeB]

CPU load is likely due to i/o wait, I do see you have SMR disks, and that is likely the reason for the slowdown during writes.

[richwood110]

2 hours ago, JorgeB said:

CPU load is likely due to i/o wait, I do see you have SMR disks, and that is likely the reason for the slowdown during writes.

But I never had this issue with my server using Windows 10 + Drive Bender software raid. The transfer rate was always >20Mb/sec (except for very small files). This is the exact same system, no change in hardware. With unraid the copy starts fast (around 100 MB/sec) and after a certain time it slows down more and more. CPU load increases until the transfer rate drops to alomst 0. It stays down for some minutes and then it comes up to ~40 MB/sec until the next stop. This repeats with different time intervals... As I have written, currently it takes around 24h to transfer 200 GB this way!

 

I thought my HDDs are commonly used and my CPU should be able to process such copy process without any issue...

Or am I missing any specific driver for Unraid?

[primeval_god]

SMR disks perform poorly in RAID arrays (including the unRAID array). If i remember correctly Drive Bender is not actually RAID, it used disk pooling and some sort of file level redundancy.

[richwood110]

9 hours ago, primeval_god said:

SMR disks perform poorly in RAID arrays (including the unRAID array). If i remember correctly Drive Bender is not actually RAID, it used disk pooling and some sort of file level redundancy.

True, Drive Bender is no RAID. Just informed myself about SMR disks and yes, this might indeed be the issue here. Ok, I might have to think about an replacement then...

Thank you for notice!

[richwood110]

I exchanged my drives from SMR to CMR (Seagate Ironwolf NAS) and I can confirm, that the issue with high CPU load and blocked copy process is gone now! Thank you for this hint!

So, I can confirm that SMR drives indeed performs very poor in Unraid!