HP5308xl tx Drops

[davidle75]

Hi wonder if anyone can help, I'm new to HP Procurve, but been racking my brain on a 5308 switch. I been seeing alot of Tx Drops on the one of th interfaces:
FCS Rx : 0 Drops Tx : 147,883,275
Alignment Rx : 0 Collisions Tx : 0
Runts Rx : 0 Late Colln Tx : 0
Giants Rx : 0 Excessive Colln : 0
Total Rx Errors : 0 Deferred Tx : 0

The link is a 100Mb line going to our DR, the module is HP J4907A XL Gig-T/GBIC module, burst traffic replications are sent down the line. Is it a known problem? or is there a setting that I'm missing? Or is it because the 5308 & module can not handle the high bust? Please help...

[Matt Hobbs]

It is highly possible that those drops are due to the module being used, the J4907A is oversubcribed by nature. From memory, ports 15-16 may give you better performance on that module.

If you have a 4-port Gigabit module, that would be a better choice to use for servers or switch-to-switch links.

[davidle75]

Thank Matt for the advice, I would like to use Gig line, but find out that the link is 100Mb LES to our DR, ports 15 and 16 are being used already on the module, are there any settings on the switch that might elivate the Drop Tx? or another module that can be recommended?

[Richard Brodie_1]

Looking at the utilization on the 100MB line averaged over short periods would probably be a good idea.

If your traffic bursts saturate the link for any length of time then there's likely not much you are going to be able to do tweaking hardware.

Does it have a measureable impact on your application performance?

[davidle75]

Thanks richard, its replication traffic going over to the DR, its just annoyeing that the Drop tx are occuring, I like to see clean lines with no drops or errors.

[André Beck]

Hi,

> its replication traffic going over to the
> DR, its just annoyeing that the Drop tx are
> occuring, I like to see clean lines with no
> drops or errors.

With a 100Mbps egress port in a switch that's likely to be uplinked with multiple Gbps, transmit drops are completely natural. Queues can only grow so much, then it's drop time inevitably. I dunno what a "DR" is, if it's a host, maybe activating flow control could help a little. In practice it's mostly useless as soon as the traffic is coming into the switch on an ISL. Only when the traffic comes in from another host connected to the same switch and that host has flow control negotiated as well there might be a chance of it helping.

So in short, transmit drops are business as usual at bandwidth bottlenecks, ignore them unless they overshadow anything else and become a packet loss burden the applications actually feel. Like with collisions in the elder days.

HTH,
Andre.