SF300 ARP Table Overflow

[carrickbradley]

I have a SF300 in layer 3 mode for interVLAN routing with four other SF300s in layer 2 mode.

In my syslog I noticed a lot of ARP Table Overflow errors. We have an enviroment with over 300 nodes which doesn't surprise me that I would see some ARP table overflow but not at the rate I'm actually receiving.

Should I adjust the aging time to agreesivly age the CAM table? Will tuning the time to a shorter duration cause a potential ARP storm?

[Tom Watts]

Hi Carrick, what software version is the switch running? Is there any negative impact with the way things are?

Can you disconnect all connecting devices then reconnect them and does the arp overflow persist?

I recommend against messing with the cam as it would add network overhead / noise.

The switch is spec'd to handle that many access hosts but in my experience I've seen adverse behavior when the 250+ marker is hit.

-Tom
Please mark answered for helpful posts

[carrickbradley]

Hi Tom,

We where runing 1.3.0.62. We had a lot of latency between VLANs. Transferring files from one VLAN to another was less than 200 kbps. Further troubleshooting lead to the core switch which is running MLS.

After running basic show commands on the core switch, the switch locked up. I reverted back to

1.2.7.76. This cleared up the issues. The ARP traffic is nominal again, latency between VLANs is back to sub millisecond,  transfer are back to nonimal rates and the show commands are not locking up the switch.

I'm not sure if the SMB firmware does a good job of cleaning up the config from an older version of firmware. I have a few spare switches to see what changes are applied and what is deprecated. In the past I found really odd random lines of garbage in the configs when comparing the config from one version to another with examdiff. Then again this was comparing an apple to a orange with firmware from a few years ago with 1.2.7.76. Though I am curious.

On a side note:

I was looking at changing the core switch with two SG500x switches stacked for redudancy. I can't budget at the moment for 3750's which I would prefer at the core/dist level for the size we are. Were an SMB that's growing rapidly and my gut is telling me to go with the 3750s.

[Tom Watts · Accepted solution]

Hi Carrick, your observations are accurate. A lot of the firmware updates don't carry over well and it has been generally a best practice to factory default the switch after the upgrades.

I have always felt the switches lost something after the 1.1.2.0 release as far as overall performance goes even with the excellent feature enhancements that came after.

The SX500X switches are probably close to the same price as a 3750 switch in some markets. Even if it were an extra $500 per switch the 3750's would be a better choice so long as you can also afford the contracts. The SB switches are growing tremendously and they are a great product, especially at the price point. The warranty is fabulous and the support is good.

I don't feel the SB switches are yet in to the position to be a core switch, certainly an access or a light distribution switch they are more than perfect in most scenarios.

Personally if you have the choice, I wouldn't be surprised if you downgraded to 1.1.2.0 and defaulted the switch most or all performance caveats you observe would disappear. It was personally my most favorite release.

-Tom
Please mark answered for helpful posts

[carrickbradley]

Hi Tom,

I appericiate your assistance and advise.

This makes perferct sense. Ever since we upgraded to newer versions I would run into random periods of latency for no reason. I could never pinpoint the issue: VLAN overload, broadcast storms, STP, QoS, etc. Heck the 1.2.7.76 caused massive issues when I had a faulty cable between from the core to the distribution causing GVRP to transmit over and over when the link went up and down. I disabled GVRP and manually added the VLANs to all switches.

My main reason for going with newer firmware was to unlock the full command line. I come from ye olde days where CLI ruled the land. I find the GUI to be slow and cumbersome at times. The sacrifice for performance is not worth the CLI. I'll be running the 1.1.2.0 in lab to fully test the features I would like to implement.

For the 3750s, I agree the smartnets drive me nuts espcially renewing for nearly half the cost of the original purchase of the gear and smartnets.