Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-08 Thread David Miller
From: Eric Dumazet Date: Wed, 07 Dec 2016 09:19:33 -0800 > From: Eric Dumazet > > Under UDP flood, many softirq producers try to add packets to > UDP receive queue, and one user thread is burning one cpu trying > to dequeue packets as fast as possible. > > Two parts of the per packet cost are

Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-08 Thread Eric Dumazet
On Thu, 2016-12-08 at 07:36 -0800, Rick Jones wrote: > On 12/08/2016 07:30 AM, Eric Dumazet wrote: > > On Thu, 2016-12-08 at 10:46 +0100, Jesper Dangaard Brouer wrote: > > > >> Hmmm... I'm not thrilled to have such heuristics, that change memory > >> behavior when half of the queue size (sk->sk_rcv

Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-08 Thread Rick Jones
On 12/08/2016 07:30 AM, Eric Dumazet wrote: On Thu, 2016-12-08 at 10:46 +0100, Jesper Dangaard Brouer wrote: Hmmm... I'm not thrilled to have such heuristics, that change memory behavior when half of the queue size (sk->sk_rcvbuf) is reached. Well, copybreak drivers do that unconditionally, e

Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-08 Thread Eric Dumazet
On Thu, 2016-12-08 at 10:46 +0100, Jesper Dangaard Brouer wrote: > Hmmm... I'm not thrilled to have such heuristics, that change memory > behavior when half of the queue size (sk->sk_rcvbuf) is reached. Well, copybreak drivers do that unconditionally, even under no stress at all, you really shoul

Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-08 Thread Jesper Dangaard Brouer
On Wed, 07 Dec 2016 09:19:33 -0800 Eric Dumazet wrote: > From: Eric Dumazet > > Under UDP flood, many softirq producers try to add packets to > UDP receive queue, and one user thread is burning one cpu trying > to dequeue packets as fast as possible. > > Two parts of the per packet cost are :

Re: [PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-07 Thread Eric Dumazet
On Wed, 2016-12-07 at 09:19 -0800, Eric Dumazet wrote: > From: Eric Dumazet > This patch gave me about 5 % increase in throughput in my tests. > BTW, this reduces the gap with GRO being on or off for me : mlx4 drivers has a copybreak feature when GRO is off : Packets below 256 bytes are simpl

[PATCH net-next] udp: under rx pressure, try to condense skbs

2016-12-07 Thread Eric Dumazet
From: Eric Dumazet Under UDP flood, many softirq producers try to add packets to UDP receive queue, and one user thread is burning one cpu trying to dequeue packets as fast as possible. Two parts of the per packet cost are : - copying payload from kernel space to user space, - freeing memory pie