Low Throughput Between FastE and GigE Ports

[Rontec]

The switch in question is a 2510-24 J9019A on release #Q.11.57 serving as an edge switch in a data center.

Port 25 of the J9019A is linked in auto 1000FDx to a Cisco WS-C3560G-48TS.

I also tried connecting port 26 of the J9019A to a Cisco Catalyst 4507 with the same, or worse, results.

The interface on the Cisco core switch is configured in access mode with all default settings.

The ProCurve switch has mostly default settings as well for the switch and interfaces.

When data is sent as such:

Juniper router --> WS-C3560G-48TS (GigE) --> (GigE port 25) J9019A (GigE port 26) --> (GigE) MyComputerA

then performance is very good at over 60 Mbps in a single TCP thread when downloading from Microsoft

However, in this scenario:

Juniper router --> WS-C3560G-48TS (GigE) --> (GigE port 25) J9019A (FastE port 1 to 24) --> MyComputerA

then per thread TCP throughout maxes out at 20 Mbps.

In either case, when data is sent in the opposite direction (from FastE to GigE), I easily hit 90 Mbps in a single thread so this is a one way problem.

Things I have tried:

1) Change GigE ports on the WS-C3560G-48TS. Same result.

2) Connect MyComputerA to WS-C3560G-48TS directly. Works perfect.

3) Disconnect interface 25 and connect interface 26 to Cisco Catalyst 4507. Even worse performance at only 10 Mbps.

4) Set ports to manual 1000 full and 100 full. No improvement.

5) Update ProCurve firmware and reboot the switch. No improvement.

6) I have tested performance between FastE ports. Works perfectly fine.

7) Pulling my hair out. No improvement.

What am I missing here?

It doesn't look like a software problem and I can't believe that all 2510-24 switches have such a defect.

[paulgear]

---

@Rontec wrote:
...

3) Disconnect interface 25 and connect interface 26 to Cisco Catalyst 4507. Even worse performance at only 10 Mbps.

...

6) I have tested performance between FastE ports. Works perfectly fine.

...

What am I missing here?

It doesn't look like a software problem and I can't believe that all 2510-24 switches have such a defect.

---

Was the test at 6) two 100 Mbps ports on the same switch?  I would suggest testing throughput from the same 100 Mbps port to one of the Gigabit ports on the same switch.

If you can't reproduce the problem that way, the issue is likely cabling.  Get the uplink cabling tested, and look for errors on the uplink interface stats.

If you CAN reproduce the problem that way, the issue is likely due to buffering.  My guess is that you're getting packet drops due to an insufficiently large buffer between the 100 Mbps and 1000 Mbps ports.  I can't find any specs on it, but i'd imagine that a cheaper model like the 2510 has pretty small buffers.  Others might be able to offer more expertise on possible workarounds, such as experimenting with flow control combinations, but it has been a long since i've had to touch that sort of thing.

Incidentally, if you load up the switch with more clients, it might perform a bit better.  If the uplink is fully utilised by 10 or 15 different 100 Mbps clients, it won't be trying to hammer packets out that one port so fast... ;-)

I'm assuming in all of this that you don't have another identical switch that you can swap in for it.

Good luck!

[Rontec]

Paul, thanks for your response.

I replaced the copper cat6 cable, which was plugged into port 25, with a SM fiber optic cable in a mini GBIC in port 26. Same performance problem - so the issue is certainly not a cable.

I can't see the switch buffers being overloaded with utilization on the GigE port being about 9% and total traffic on the switch, all directions and ports combined, is under 500 Mbps. There is nothing at all in the logs or diagnostics to suggest insufficient buffers or packet drops.

I seem to recall previously having a 2524, which I upgraded to the 2510, with the exact same problem.

As such, I don't think my switch is broken.