Xmit errors on Cat 4006 FE ports

[mmedwid]

In the tech note below - an explanation for Xmit error is when say a higher bandwidth pipe like gig comes to a switch and then traffic needs to egress over say a fast ethernet link. I have a situation where there are 16 servers on an FE blade blade and they communicate with each other. But even when snmp bandwidth graphing does not show traffic approaching 100Mbps - I am seeing xmit errors. e.g. the top talker and top listener in this group might hit 30Mbps.

I expect to see xmit errors when bandwidth consumption is approaching 100Mbps on a port. But I'm not understanding what would cause xmit errors in the lower bandwidth scenarios. My only other thought was that perhaps higher bandwidths are being reached in between the sampling of counter periodicity of the graphing tool which is four minutes. So I need to go back and manually do some octet diffing in narrower time bands to see if perhaps that's what's causing the mystery. But meantime - I wondered if anyone on the forum might have insight into what else could cause xmit errors on FE ports when the traffic is all between 16 hosts on a FE blade as described.

http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note09186a008015bfd6.shtml

[gpulos]

there are a few things that could cause xmit errors:

cabling (run and patch; connectors; damage)

cpu too busy

queues full

speed/duplex settings

30mbps is not to bad but you should remember you will really never seee 100mbps throughput as there is network overhead, tcp overhead, etc.

an ethernet network 60%-70% utilized is said to be saturated and will begin seeing drops if increased and possibly sustained. (this is especially so as the packet sizes decrease; the larger the packet size, the more utilization you can obtain. the smaller the packet size, less utilization can be obtained)

see these links for more info:

http://www.ethermanage.com/ethernet/pdf/molle-report.pdf

http://support.3com.com/infodeli/tools/netmgt/tncsunix/product/091500/c8bandut.htm#14315

[mmedwid]

Thanks for the thoughts. CPU stayed at normal, speed and duplex settings are all matched at 100Mbps FD. There are no collision errors or runs or other - just the xmits. The odds that just these 16 hosts all have cabling problems is pretty unlikes. I have in fact graphed traffic bps to 100Mbps in some cases - possible as the connection is full duplex to a switch so you don't have back-off to contend with.

The queues full is certainly what's going on here. But what is not clear to me is how and why the queues fill up when there is less than say 50Mbps of traffic heading out the port. Like I say - it could be that the actual amount of traffic to the port is bursty and not caught within the 4 minute smoothing of the graphing tool. But perhaps there's something else in the details of how the queue (transmit buffer) works. ?

[glen.grant]

I believe the switching asics are grouped into groups of either 4 or 8 ports a piece and if you have high traffic densities within that group it could be why you are seeing the xmit errors as the queues for that particualr group are being overwhelmed at times...