Closed antoinedeschenes closed 6 years ago
Thank you for your interest in this project !
I grabbed a quick cpu profile of a running application :
What is taking the large majority of the cpu time is the packet capture with pcap. This definitely needs a more thorough investigation.
I see, I built a version using afpacket instead of libpcap earlier this week but didn't try it yet on a VLAN trunk. (At least I saw some logs going on my home network)
This should be fixed, as we updated gopacket
so we can benefit from https://github.com/google/gopacket/commit/fa2cf54d72acd452a072ef005e231b5d7a7b5ebd
If you still have any problem, please feel free to reopen the issue !
Hi, this seems to work really well, CPU usage has gone from 100% to <1%. My afpacket tests turned out to work on x86 but didn't work at all on the mips router
Hey there,
I'm running bonjour-reflector on a homeserver with constant network-load on a single interface. The server traffic is running on the interface with an untagged vlan and all wifi-vlans are transmitted tagged to the server.
An average of approx. 70MBit/s is running constantly through that interface, causing an average 20% load on the bonjour-reflector process.
Was anyone investigating on that issue further since last year?
Nevermind - https://github.com/Gandem/bonjour-reflector/pull/23 seems to resolve that issue, will use that until merging.
In my case the gopacket update on master was sufficient, but I was running bonjour-reflector on a separate interface passively listening for multicast traffic. Now of course BPF should be a lot more efficient.
CPU usage seems to be quite high, I compiled bonjour-reflector on MIPS architecture and I get 100% load (of course it's not a very powerful CPU to start with...). I'm wondering if packet processing could be threaded, source VLANs filtered earlier or something.