SG-2100 packetloss in internal 5 port switch

keyser

Hi All

I have discovered a hardware issue with the builtin 5 port switch in the SG2100 that I can reliably reproduce.

Setup: Multiple tagged client VLANs on mvneta1 (switched through the internal 5 port switch) to an external Aruba 6100 Switch.
It makes no difference whether I

Disable 802.1q mode and just uplink one port to a trunk port on Aruba
Enable 802.1q mode and assign client VLANs untagged to different ports on the 2100 and uplink them to the VLAN accessports on the Aruba Switch.

Problem: Copying large files or moving large amounts of data at speed between 2 devices on different VLANs (traffic passes pfSense), will cause the internal switch to drop packets in increasing "chunks" until the session eventually breaks down and the TCP connection is reset between the clients.
The copy speed is initially the 55'ish MB/s possible with 2100 doing routing and PF'ing. But it will degrade slightly until after about 20 - 30 seconds where the session will abruptly get disconnected and stop copying.

I'm sure It's the internal 5 port switch that causes the issue because if I instead assign the VLANs to mvneta0 and directly link that to a trunkport on the Aruba switch, then there is no pocketless and disconnection. I have analyzed the loss with a packet capture on clients, and the packet loss comes in "pulses" until it is finally great enough that TCP breaks down.

There is no port- or packet errors on the Aruba switch at any time during this, and I suspect the issue is actually a speed/flowcontrol negotiation issue on the 5 port switch side. I have tried fixing that to 1000FDX but it makes no difference. Unfortunately I do not have another external switch to test with, so I was wondering if anyone else have experienced the same issue?

I have found only one on the internet seeing the same thing, but unfortunately he uses the EXACT same hardware (2100 and Aruba 6100), so that could suggest it's a link issue between those two devices specifically.
https://www.reddit.com/r/PFSENSE/comments/twswig/weirdest_issue_ever_experts_needed_smb_hangs_over/

luckman212

@keyser Have you tested the 25.07-RC on it? Funny, that reddit post you linked to has a reply from me 3 years ago, so I guess I encountered this same issue too (I'd forgotten).

keyser

@luckman212 No, and that’s obviously the next step.
I don’t have any hopes of improvement though.

Do you know if there is any tweaks to be done with software on the built in Marvell switch? Buffers, see packetdrops and such

luckman212

@keyser Sorry, I don't, and I only have a few 2100's left in service. I assume you've already experimented with disabling LRO/TSO/hardware offload, and MSIX etc. I can't find any tunables specifically for the Marvell 6000 switch. Good luck.

stephenw10

Did you try enabling dot1q mode and trunking the VLANs through the internal switch on one port? Hard to see why that would be any different but....

Did you try a different external switch?

keyser

@stephenw10 said in SG-2100 packetloss in internal 5 port switch:

Did you try enabling dot1q mode and trunking the VLANs through the internal switch on one port? Hard to see why that would be any different but....

Did you try a different external switch?

Yes, I tried all combinations, and the issue remains as long as the builtin switch is the connection to the Internal aruba Switch.