Thread: New switch

Right need some help, i'm sure you lot will be more helpful than the WHT people

I need a different 48 port switch for a rack. Any recommendations? We are currently using an old Cisco 2948G which has about 30 of the ports being used and pushes around 130mb at maximum peak through its ge port.

Its got nothing special running on there other than SNMP (if you can call that special) so guess that is my only basic requirement other than that is its an managed switch. Reason for change? Found a nice Cacti script to monitor the switch's CPU and we found its hitting 1.0 solid at peak periods in the evening which looks (well to me anyway) as 100% CPU usage which can't be a good thing.

Any thoughts is appreciated

-Chris

I don't think a 2948G doing only basic L2 switching is going to be at 100% CPU with only 30 ports at 130MB peak.

That 1.0 is probably more like 1%.

To be honest, we not sure but its too close to 1.0 everytime at the same day to be 1% in my mind but been wrong many times before. Graph attached, make of it as you will.

.... that looks regular enough to be backup traffic?

I don't think actual port thoughput has anything to do with CPU utilisation on the 2948G as its done in hardware - its more likely to be something like MAC learning, STP changes or snmp\telnet\http activity or similar

Originally Posted by markcastle

.... that looks regular enough to be backup traffic?

People get home. Play games. Go to bed. Repeat.

People get home. Use Ventrilo/Teamspeak. Go to bed. Repeat.

OK, this might sound really obvious but if it's happening at the same time each day then can you lot log in at that time and do "show proc cpu" and see what the CPU load is and if it's high, which process is killing it? Maybe I'm missing something here...

the 1 will be 100%, and no, thats not good for your service...

you want to look at something like a 6504 to keep the density and pps or split it across multiple switches aggregated to a core switch capable of handling all the traffic

3*2960
connected to a 3750G
connected to your upstream
for example.

gaming is hard on switches if you want to keep to low/no packetloss etc, and if your graphing it you want to watch both the unicast and non-unicast traffic as well.

Yup, it's the PPS rather than the absolute traffic, and gaming is a high PPS usage type compared to normal web traffic. I assume you've got Cacti graphing PPS?

Still seems high to me. Have had 2950's do similar under gaming/voice and didn't see a problem.

2950s are significantly newer than 2948G from memory though (someone correct me if I'm wrong, I'm no Cisco follower)

What does "sh proc cpu" show at peak times? I can't believe you're hitting 100% CPU with layer 2 traffic.

The 2948G (the layer 2 version) was a bit of a weird one, despite being part of the 29xx series of fixed switches, it ran CatOS like the higher end 4x00/5x00/6x00 modular series switches. It was EOLed several years ago whereas the 2950 was only EOLed recently to be replaced by the 2960.
Is yours the layer 2 (CatOS) or layer 3 (IOS) version? The straight 2948G is the CatOS/layer 2 version IIRC whereas the 2948G-L3 is the IOS/layer 3 version.
I'd take a look at what's hogging CPU time on that 2948G as I've got one (and the related 2980G) and there is no way that should be maxing out under normal use.

Layer 2, CatOS version.

All we are using atm the is the memory usage, cpu and general bandwidth. Due to the fact its CatOS all I could monitor with the scripts available was the memory, CPU and network usage.

While we do have some 2950 in other racks doing fine, please do remember that this is 130mb/s of pure gaming and ventrilo/teamspeak traffic with the majority being voice servers. I tried to do show proc cpu last night but had an issue else where to attend too.

-Chris