Re: [PATCH net-next] pkt_sched: fq: Fair Queue packet scheduler

On Sat, Aug 24, 2013 at 08:08:02PM -0700, Eric Dumazet wrote:
[...] 

Attempts have been made to add TCP pacing in TCP stack, but this
seems to add complex code to an already complex stack.

TCP pacing is welcomed for flows having idle times, as the cwnd
permits TCP stack to queue a possibly large number of packets.

This removes the 'slow start after idle' choice, hitting badly
large BDP flows.

Nicely spaced packets : here interface is 10Gbit, but flow bottleneck is
~100Mbit

This is great. I just gave this a try in a real-world scenario where
TCP pacing (implemented in the TCP stack) has provided a significant
performance improvement.

# netperf -v 2 -H 10.162.184.110 
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.162.184.110 (10.162.184.110) port 0 AF_INET
Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    10^6bits/sec  

 87380  16384  16384    10.20      85.25   

Alignment      Offset         Bytes    Bytes       Sends   Bytes    Recvs
Local  Remote  Local  Remote  Xfered   Per                 Per
Send   Recv    Send   Recv             Send (avg)          Recv (avg)
    8       8      0       0 1.087e+08  16385.34      6635   14583.07   7455

Maximum
Segment
Size (bytes)
  1424

# tc qdisc add dev eth0 root fq
# netperf -v 2 -H 10.162.184.110
TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.162.184.110 (10.162.184.110) port 0 AF_INET
Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    10^6bits/sec  

 87380  16384  16384    10.12     354.79   

Alignment      Offset         Bytes    Bytes       Sends   Bytes    Recvs
Local  Remote  Local  Remote  Xfered   Per                 Per
Send   Recv    Send   Recv             Send (avg)          Recv (avg)
    8       8      0       0 4.488e+08  16384.08     27394   15526.00  28908

Maximum
Segment
Size (bytes)
  1424

Is there an iproute2 patch? I don't think I've seen one yet.

As far as this patch is concerned, I like what I see so far from
performance. Are there concerns about boundary crossing through
teaching the scheduler about things like the rate limit, special
handling of TCP retransmits, etc?

--msw