topic Dropped packets, Frame errors, rxbds_empty, rx_discards in Operating System - Linux

Dropped packets, Frame errors, rxbds_empty, rx_discards

Jack Kidwell — Mon, 17 Aug 2009 20:11:59 GMT

Details

machine: HP PROLIANT SE1210
NIC: Tigon3 rev a200 PCI Express 10/100/1000Base-T Ethernet
driver: tg3
version: 3.93
firmware-version: 5722-v3.07, ASFIPMI v6.02
network config: 1000Mb/s full duplex, autoneg on

Issue

We are experiencing high TCP retransmit volume when received network traffic approaches 200Mb/s. When retransmits occur, the ifconfig drop and frame counters increment rapidly, as do the ethtool rxbds_emtpy and rx_discards statistics.

We tried enabling hardware flow control using ethtool -a option, and see pause traffic, but the drops, frames, rxbds_empty and rx_discards continue.

This condition is crippling throughput.

Any ideas?

Re: Dropped packets, Frame errors, rxbds_empty, rx_discards

Vitaly Karasik_1 — Tue, 18 Aug 2009 06:50:37 GMT

200Mb/s is pretty high traffic.
I suggest you to check:
- how many packets/sec you have. In fact, network throughput mostly depends on packets/sec, not bytes/sec
- what is your linux version/patchlevel?
- system utilization

Vitaly

Re: Dropped packets, Frame errors, rxbds_empty, rx_discards

Jack Kidwell — Tue, 18 Aug 2009 13:19:27 GMT

When working, the box handles 70,000 packets/s with throughput of 151 Mb/s, and exhibits errors when packets/s peaks at 118,000 and throughput is 202 Mb/s.

uname -a:
Linux hostname 2.6.18-128.1.16.el5 #1 SMP Tue Jun 30 06:07:26 EDT 2009 x86_64 x86_64 x86_64 GNU/Linux

/etc/redhat-release:
CentOS release 5.3 (Final)

uptime:
09:29:26 up 3 days, 19:21, 3 users, load average: 5.18, 5.01, 5.04

CPUs:
eight (8) Intel(R) Xeon(R) E5430 @ 2.66GHz

MemTotal:
16,443,824 kB

Seems like a capable machine. But the NIC is a choke point.

Do you know what rxbds_empty is? Looking at the source code, it appears to be a circular list of block descriptors. Could this be a resource depletion issue?