TL-SX3016F > CPU RISING THRESHOLD (figured out the cause and a workaround)
TL-SX3016F > CPU RISING THRESHOLD (figured out the cause and a workaround)
I got this switch about a week ago and have notice the CPU usage is always 87-89%. That's true whether at idle or running iperf3 tests on 2 machines saturating 4 of the ports.
The setup is nothing special. I have a few VLAN's, RTSP, and a LAG configured.
A log entry is made every 3 minutes on the nose.
Any thoughts as to why this might be and how to fix it? I did disable SNMP and restarted the switch but that didn't seem to change anything.
I ordered another one of these switches as well so getting this figured out would be great as these are my first TP-Link switches but probably not my last if I can get this single issue figured out.
- Copy Link
- Subscribe
- Bookmark
- Report Inappropriate Content
Hi @jlficken
Thanks for posting in our business forum.
jlficken wrote
jlficken wrote
I figured out what causes the CPU usage to go to 88-90%.
If I have 1 x SFP+ MMF transceiver installed the CPU sits at 0-1%.
If I have 2 x SFP+ MMF transceivers installed the CPU sits at 73-76%.
If I have 3 x SFP+ (or more) transceivers installed the CPU sits at 88-90%.
My other TL-SX3016F arrived today and I tested it in both units with different transceivers (all Mellanox MFM1T02A-SR) and it happens with both switches.
I did also try some UBNT transceivers and they show the same symptom...with a twist. If I just have them installed but not in use then the utilization stays at 0-1%, however, if I move the fiber cable from Port 1 to Port 2 or Port 3 (UBNT #1 and #2) then the CPU usage goes to 73-76%. I then move the fiber cable back to Port 1 and CPU usage goes back down to 0-1%.
I'm now more confused than ever.
This is expected to be normal. Since this was brought up on the forum by someone else again, here's the explanation
When the module is plugged in but at idle(no fiber is connected to it), the CPU usage would keep a high level. It is normal due to the chipset design.
If you don't use the SFP port, you can unplug the module from the switch to avoid high CPU usage. When the fiber and module are used at the same time on the switch, the CPU usage would fall back to a normal level.
- Copy Link
- Report Inappropriate Content
What information is shown in the log on the switch?
It would be clearer to show a screenshot.
- Copy Link
- Report Inappropriate Content
- Copy Link
- Report Inappropriate Content
I figured out what causes the CPU usage to go to 88-90%.
If I have 1 x SFP+ MMF transceiver installed the CPU sits at 0-1%.
If I have 2 x SFP+ MMF transceivers installed the CPU sits at 73-76%.
If I have 3 x SFP+ (or more) transceivers installed the CPU sits at 88-90%.
My other TL-SX3016F arrived today and I tested it in both units with different transceivers (all Mellanox MFM1T02A-SR) and it happens with both switches.
I did also try some UBNT transceivers and they show the same symptom...with a twist. If I just have them installed but not in use then the utilization stays at 0-1%, however, if I move the fiber cable from Port 1 to Port 2 or Port 3 (UBNT #1 and #2) then the CPU usage goes to 73-76%. I then move the fiber cable back to Port 1 and CPU usage goes back down to 0-1%.
I'm now more confused than ever.
- Copy Link
- Report Inappropriate Content
Dear @jlficken ,
jlficken wrote
I figured out what causes the CPU usage to go to 88-90%.
If I have 1 x SFP+ MMF transceiver installed the CPU sits at 0-1%.If I have 2 x SFP+ MMF transceivers installed the CPU sits at 73-76%.
If I have 3 x SFP+ (or more) transceivers installed the CPU sits at 88-90%.
My other TL-SX3016F arrived today and I tested it in both units with different transceivers (all Mellanox MFM1T02A-SR) and it happens with both switches.
I did also try some UBNT transceivers and they show the same symptom...with a twist. If I just have them installed but not in use then the utilization stays at 0-1%, however, if I move the fiber cable from Port 1 to Port 2 or Port 3 (UBNT #1 and #2) then the CPU usage goes to 73-76%. I then move the fiber cable back to Port 1 and CPU usage goes back down to 0-1%.
Thank you very much for reporting this issue in the community.
1. Just to confirm, when you mentioned "A log entry is made every 3 minutes on the nose." does it mean that the CPU goes up to 87-89% every 3 minutes and then drops back down?
Or does the CPU always stay at 87-89%, but the log is prompted every 3 minutes?
2. You mentioned the CPU spike caused by inserting the transceiver, but after confirming with the support team, it may be because the optical module was not linked when you tested it and the switch mechanism caused the CPU spike.
However, once you have actually inserted the fibre into the transciever and ensured that both sides are linked, the CPU will immediately drop back to normal values.
Best Regards!
- Copy Link
- Report Inappropriate Content
Hank21 wrote
Thank you very much for reporting this issue in the community.
1. Just to confirm, when you mentioned "A log entry is made every 3 minutes on the nose." does it mean that the CPU goes up to 87-89% every 3 minutes and then drops back down?
Or does the CPU always stay at 87-89%, but the log is prompted every 3 minutes?
2. You mentioned the CPU spike caused by inserting the transceiver, but after confirming with the support team, it may be because the optical module was not linked when you tested it and the switch mechanism caused the CPU spike.
However, once you have actually inserted the fibre into the transciever and ensured that both sides are linked, the CPU will immediately drop back to normal values.
Best Regards!
1) The CPU stays at 87-89% at all times. The log entry is made every 3 minutes.
2) That appears to be the case but only with certain transceivers. I have 2 machines that aren't always running, however, they do have the fiber installed into the transceiver on both ends. Since they're off though the Link is "down".
I discovered that with Mellanox MFM1T02A-SR modules appear to cause the CPU to stay high if there's no active link even if the fiber is plugged in on both ends. The only way the CPU usage stays low is if the link is "up".
I grabbed a couple of UBNT UF-MM-10G modules from my EdgeSwitch and tested them as well and here's what I found:
1) Just plugging them in doesn't cause the CPU usage to rise.
2) Turning the machines on so the links come up doesn't cause the CPU usage to rise.
3) Turning the machines off so the link goes down doesn't cause the CPU usage to rise.
4) Rebooting the switch with the machines off doesn't cause the CPU usage to rise even when those machines are off.
For now I'll leave the UBNT modules installed for those 2 machines (and any others that will be off sometimes in the future) since it seems to fix the CPU usage issue.
- Copy Link
- Report Inappropriate Content
jlficken wrote
I figured out what causes the CPU usage to go to 88-90%.
If I have 1 x SFP+ MMF transceiver installed the CPU sits at 0-1%.
If I have 2 x SFP+ MMF transceivers installed the CPU sits at 73-76%.
If I have 3 x SFP+ (or more) transceivers installed the CPU sits at 88-90%.
My other TL-SX3016F arrived today and I tested it in both units with different transceivers (all Mellanox MFM1T02A-SR) and it happens with both switches.
I did also try some UBNT transceivers and they show the same symptom...with a twist. If I just have them installed but not in use then the utilization stays at 0-1%, however, if I move the fiber cable from Port 1 to Port 2 or Port 3 (UBNT #1 and #2) then the CPU usage goes to 73-76%. I then move the fiber cable back to Port 1 and CPU usage goes back down to 0-1%.
I'm now more confused than ever.
- Copy Link
- Report Inappropriate Content
- Copy Link
- Report Inappropriate Content
@jlficken hi
- Copy Link
- Report Inappropriate Content
- Copy Link
- Report Inappropriate Content
jlficken wrote
I figured out what causes the CPU usage to go to 88-90%.
If I have 1 x SFP+ MMF transceiver installed the CPU sits at 0-1%.
If I have 2 x SFP+ MMF transceivers installed the CPU sits at 73-76%.
If I have 3 x SFP+ (or more) transceivers installed the CPU sits at 88-90%.
My other TL-SX3016F arrived today and I tested it in both units with different transceivers (all Mellanox MFM1T02A-SR) and it happens with both switches.
I did also try some UBNT transceivers and they show the same symptom...with a twist. If I just have them installed but not in use then the utilization stays at 0-1%, however, if I move the fiber cable from Port 1 to Port 2 or Port 3 (UBNT #1 and #2) then the CPU usage goes to 73-76%. I then move the fiber cable back to Port 1 and CPU usage goes back down to 0-1%.
I'm now more confused than ever.
- Copy Link
- Report Inappropriate Content
Information
Helpful: 0
Views: 2544
Replies: 11
Voters 0
No one has voted for it yet.