Bits per second on a 1Gbps interface?

[wardy]

Hi,

We're monitoring ports on a Cisco 3500 switch. All the ports are running at 1Gbps, as too as the linux servers that connect to them.

However, the graphs for the switch are only showing around 60 megatbits per second (according to the axis on the graph).

I'm running 64bit counters so is it just that the axis is wrong? It actually says "bits per second" with the values having a "M".

Throughput is definitely higher, I wrote a large file to an NFS mount:

[root@sscdevadm01 restore]# dd if=/dev/zero of=/restore/monster-file bs=1024 count=30000000
30000000+0 records in
30000000+0 records out
30720000000 bytes (31 GB) copied, 722.557 seconds, 42.5 MB/s


Thanks.

[gandalf]

It _may_ happen, that the test was too short. If possible, double the duration. In general, you can't expect that a GB interface is always running at that speed.

As you confirmed, you're running 64 bit graphs (that's a crucial point, you know). You may note the COUNTER values at start and at end of the test, derived from a manual snmpwalk against that interface to confirm (write a small shell script that walks the interface, runs the test and again walks the interface).

R.

[wardy]

Thanks for the reply, I wrote a big file to try and get around the problem of the test duration not being long enough. That took 12 minutes so should've been picked up by cacti (We're running 1 minute polling too)

Good idea with the counters, I'll give that a try tomorrow.

Originally I was on 32bit counters but researching this problem I found about 64bit counters so I've deleted all my graphs and data sources for this switch and recreated the graphs with 64bit - I presume this is all that's necessary?

[gandalf]

Yep. Find the OIDs used now at "System Utilities -> View Poller Cache" for confirmation.
R.