pmacct-discussion  

[pmacct-discussion] problem with lost packets

Corin Langosch
Tue, 04 Mar 2008 01:43:40 -0800

Hello,

I'm running a lightweight used webserver (only lighttpd) and wanted to 
try pmacctd on it
to account the IP Traffic. I successfully installed pmacct 0.11.4-1 
using normal debian
packages. For the pmacct backend I use mysql. The server is a dual 
opteron, both
CPUs are around 98% idle, also when pmacct is running.

My problem is that pmacctd seems to miss a lot of packets. IPtraf shows 
only about 1Mbit/s incomming (470 packets/s), 1.5Mbit/s outgoing (400 
packets/s).

****
starting eth0 ip accounting...
OK ( default/core ): link type is: 1
OK: Exiting ...

3342 packets received by filter
724 packets dropped by kernel
****

My current config looks like:
****
daemonize: false
debug: false

interface: eth0
aggregate: src_host,dst_host
promisc: false
imt_path: /tmp/pmacct_eth0.pipe
pmacctd_flow_buffer_buckets: 65536
flow_handling_threads: 2

!ports_file: /etc/pmacct/ports.list

plugins: mysql

sql_host: db1
sql_user: pmacctd
sql_passwd: xxxxxx
sql_db: pmacctd
sql_table: pmacctd_eth0
!sql_table_schema: /etc/pmacct/acct_v1.schema
sql_table_version: 1
sql_refresh_time: 300
sql_optimize_clauses: true
sql_history_since_epoch: true
sql_history: 5m
sql_history_roundoff: mdh
sql_recovery_logfile: /var/lib/pmacct/sql_recovery_eth0
****

What do the dropped packets exactly mean?
How can I improve my config so that now packets are lost and the traffic 
is correctly accounted?
How much mbit/s and packets/s should pmacct be able to handle correctly 
on a modern system (dual opteron 2212)?

Best Regards,
Corin


_______________________________________________
pmacct-discussion mailing list
http://www.pmacct.net/#mailinglists
  • [pmacct-discussion] problem with lost packets Corin Langosch