Increase TX_QUEUE_SIZE

[PlagueCZ]

Testing this is a bit complicated, because VM is the weakest link usually, so it requires many many VMs, or alternatively, one can send the packets out 10 times. This of course breaks communication, however in local testing works as intended.

I was able to reach 200-300 kp/s and then the error messages start appearing.

Increasing the size of TX queue by a factor of 4 makes my local tests start failing at 1 Mp/s, and in our real-life scenario makes all problems go away, only the occasional VM unable to cope with the traffic.

In real-life deployment, factor of 4 was still occasionally producing an error, factor of 8 did not. The memory allocation is not that big, so I would recommend using the larger factor for future-proofing. (130-200 MB increase in alloc_size, but still fits into 1GiB total heap size, OSC reserves 2GiB for dpservice anyway).

Increasing more is not really possible because we run into some DPDK limit, fortunately this seems to be enough.

EAL log message, eal_msg: mlx5_net: port 0 Tx WQEBB count (65536) exceeds the limit (32768), try smaller queue size

Fixes #748

[guvenc]

LGTM