DS415+ slow performance with QNAP QNA-UC5G1T USB 3.2 Gen 1 to 5GbE #48
Comments
|
Or would it simply be better / more stable to go with a 2.5gbit adapter? |
|
First of all, you should use the native version of iperf. The docker version consumes a lot of CPU for network processing, so it is not suitable for measuring the throughput of USB-based ethernet devices which is CPU-hungry. You can get the native iperf from here. Unfortunately, there is no package for DSM 7.0, but you can use it without any problem by extracting the binary inside. Next, to get throughput in SMB close to iperf, you need to set the SMB version to v2 with LargeMTU or v3. See here for more details. Lastly, assigning a different subnet to each NIC is recommended to prevent communication through unintended NICs. In this case, it is a good idea to assign multiple IP addresses to the PC as aliases. |
|
I think I reeeaaaally screwed up this time. I extracted the So I thought, ok, then I'll copy over the /lib directory as well. The second I did this my ssh connection broke and now my NAS is completely unreachable.... shit... |
|
Due to loads of issues I've just reinstalled my NAS. How should I go about installing iperf3 natively without bricking my NAS again 🙂 ? |
|
I got a hint from someone else. Simply run this command: Once my Synology is completely running again I'll try to see what the performance is with native iperf3. |
|
I did some more tests today with the 5gbit adapter and these are the results: It seems that MTU matters A LOT for the 5gbe adapter. Even though it makes no difference at all for the 1gbit integrated network ports. Also difference for this adapter on my specific NAS between native iperf3 and docker iperf3 seems to be minimal. |
|
I also did some tests with 10 threads in iperf3 (iperf3 docker): And 10 threads in iperf3 native: |
|
To do some more testing I also ordered the CLUB 3D CAC-1420 (Type-A, confirmed working). Here's some testing results for that device. (I'll also add them to the respective github):
Notes
Additional comments I did a few file transfer tests. The transfer went quite okay at around 150+ megabytes per second. However after a few second I could hear my NAS starting to have a lot of disk activity (and then also slow the transfer down) which is presumably because it only has like 100gb of free space. So I'm most likely running into disk fragmentation issues here.
Doing read tests is quite cool to be able to get 2 separate files at full speed though:
|
|
Sorry I didn't give you more details about iperf. I'm relieved to hear that you were able to recover. Could you try to connect the devices via Y-cable or self-powered hub with an AC adapter? It's unlikely this is the cause, but. insufficient bus power has been known to cause performance or stability issues. |
|
I've decided to return the AQC module and just focus on the 2.5gbit. I hope that's understandable 😄 |
|
Since there is no way to further investigate this symptom, this issue is closed. To other people: if you encounter a similar problem, please try increasing
|
|
My observation is that USB network adapters can interact really badly with some USB Host Controllers and hubs. I have an AMD motherboard with native CPU USB ports and expanded USB ports provided by the A320 chipset (which is basically a custom ASMedia IP core). The native CPU port has no problems at all. But if a USB NIC is plugged into the ASMedia port, sending a lot of traffic from another machine to that NIC can cause an immediate hardware lockup for RTL8156B (hardware buffer overrun?) adapters, or limited to a slow 300 Mbps for AQC111 adapters. Adding a USB 3 Hub in the middle can improve it somewhat, no more lockups or throttling, but it still can't go faster than 1 Gbps on the USB side. The ASMedia port itself has no problems handling USB HDD writes at 400 MB/s. Thus, I suspect performance problems are related to how USB transactions are handled in different host controllers, with subtle interoperability problems that can only be understood by USB stack gurus... |
Description of the problem
Hi all, I just ordered the QNAP QNA-UC5G1T USB 3.2 Gen 1 to 5GbE adapter. Before testing I first hooked it up to a windows machine and did an iperf3 from that machine to my own PC. That worked fine with 1 thread steadily transferring at 3.5gbit / sec.
After that I put the QNAP adapter in one of the rear USB ports of my NAS. Since I'm running DSM 7.0 I couldn't find a version of iperf3 to run natively though so I just tested with the docker version.
Running a single thread on my 1gbit line results in the following performance (as expected):
Running a single thread on my 5gbit line through the QNAP adapter results in the following performance (slower then 1gbit line):
Running a 10 thread test on my 5gbit line through the QNAP adapter results in the following performance (A bit faster but still not getting the expected speed):
Also Windows SMB transfers are significantly slower using the QNAP adapter in comparison to the native 1gbit port:
1gbit port
5gbit QNAP
So it seems that in my case the QNAP adapter works perfectly fine from windows (getting 3.5 gbit) but not from my DS415+. Any suggestions? (On windows I did disable
Low Power 5Gbpsas stated on their website: https://www.qnap.com/en/product/qna-uc5g1t)Description of your products
Output
uname -a:Linux Devenology 3.10.108 #41890 SMP Fri Jun 25 02:40:49 CST 2021 x86_64 GNU/Linux synology_avoton_415+DSM version: 7.0
Ethernet Dongle: QNAP QNA-UC5G1T USB 3.2 Gen 1 to 5GbE
Description of your environment
Output of
dmesgcommandTruncated
Output of
lsusbcommandOutput of
ifconfig -acommandThe text was updated successfully, but these errors were encountered: