Heavy Arp-flooding can make IncludeOS unresponsive

[alfreb]

I've experienced the VM becoming unresponsive after running this command: sudo arping -w 100 -I include0 -c 100000 10.0.0.42

Not sure why yet. It's a pretty heavy and unrealistic load, but we should handle it.

[fwsGonzo]

Maybe related to bogus buffer in virtio tx dequeue

[alfreb]

Possibly. The system I'm seeing this on:

• 48 x AMD Opteron(TM) Processor 6234
• Ubuntu 14.04, running on bare metal
• ARPing v2.11

No such symptoms on

• 4 x Intel Core Processor (Haswell, no TSX)
• Ubuntu 16.04, running with nested hw-virtualization in OpenStack
• ARPing v2.14

[YanVugenfirer]

It might also be related to RX processing. Put a threshold on an amount of the processed packets here https://github.com/hioa-cs/IncludeOS/blob/master/src/drivers/virtionet.cpp#L221 otherwise you might hit a condition where the host always has a packet in a queue.

[fwsGonzo]

@YanVugenfirer good idea, I'll try a max 128 receives for starters