Re: service-routing mDNS creates constant CPU usage around 10%, even when no queries

bcoverstone · ‎09-20-2018

Given the following mDNS configuration to share printers via mDNS between two subnets on a C3560 with version 15.2(4)E5:

service-list mdns-sd permit-printers permit 10
 match message-type announcement
 match service-type _pdl-datastream._tcp.local
service-list mdns-sd permit-printers permit 20
 match message-type query

service-list mdns-sd permit-all permit 10

service-routing mdns-sd
 designated-gateway enable

interface Vlan10
 service-routing mdns-sd
  service-policy permit-printers IN
  service-policy permit-all OUT
interface Vlan11
 service-routing mdns-sd
  service-policy permit-printers IN
  service-policy permit-all OUT

If I enter debug mdns packet, it floods out the following localhost_prl request from 127.0.0.1 every 50-100 milliseconds:

Sep 20 20:58:32.454: mDNS packet dump
0 0 84 0 0 0 0 4 0 0 0 0 D 6C 6F 63 61 6C 68 6F 73 74 5F 70 72 6C 0 0 1 0 1 0 0 0 0 0 4 7F 0 0 1 C0 C 0 1 0 1 0 0 0 0 0
4 7F 0 0 1 C0 C 0 1 0 1 0 0 0 0 0 4 7F 0 0 1 C0 C 0 1 0 1 0 0 0 78 0 4 7F 0 0 1
Sep 20 20:58:32.462:
Sep 20 20:58:32.462: DOM: id=0, response, opcode=0, aa=1, tc=0, rd=0, ra=0
Sep 20 20:58:32.462:      rcode=0, qdcount=0, ancount=4, nscount=0, arcount=0
Sep 20 20:58:32.462: Answer section:
Sep 20 20:58:32.462:    Name='localhost_prl'
Sep 20 20:58:32.462:    RR type=1, class=1, ttl=0, data length=4
Sep 20 20:58:32.471:      IP=127.0.0.1
Sep 20 20:58:32.471:    Name='localhost_prl'
Sep 20 20:58:32.471:    RR type=1, class=1, ttl=0, data length=4
Sep 20 20:58:32.471:      IP=127.0.0.1
Sep 20 20:58:32.471:    Name='localhost_prl'
Sep 20 20:58:32.471:    RR type=1, class=1, ttl=0, data length=4
Sep 20 20:58:32.471:      IP=127.0.0.1
Sep 20 20:58:32.471:    Name='localhost_prl'
Sep 20 20:58:32.471:    RR type=1, class=1, ttl=120, data length=4
Sep 20 20:58:32.471:      IP=127.0.0.1
Sep 20 20:58:32.471: Authority section:
Sep 20 20:58:32.471: Additional record section:

This looks like the mDNS process querying itself, but there's not a lot of information on this process. Could this be a bug?

Georg Pauwen · ‎09-21-2018

Hello,

127.0.0.1 could be a loopback interface, do you have those configured ?

You could try and rate limit the mDNS packets:

Router(config-mdns)#rate-limit in 50

bcoverstone · ‎09-21-2018

Hi Georg,

No loopbacks are configured. There are only around 15-20pps, which doesn't seem that it should be causing almost 10% CPU usage.

Strange thing, after an hour of setting up mDNS, the unknown loopback queries stopped and mDNS is now taking about .1% CPU.

Besides the high CPU, the debug packets seem vague on what and where they come from. Googling anything about it seems to lead to a dead end.