Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables

[PATCH] net/sock: move memory_allocated over to percpu_counter variables · Olof Johansson <hidden> · 2018-09-06
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Eric Dumazet <edumazet@google.com> · 2018-09-06
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Herbert Xu <herbert@gondor.apana.org.au> · 2018-09-07
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Olof Johansson <hidden> · 2018-09-07
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Eric Dumazet <edumazet@google.com> · 2018-09-07
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Eric Dumazet <edumazet@google.com> · 2018-09-07
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Olof Johansson <hidden> · 2018-09-08
Re: [PATCH] net/sock: move memory_allocated over to percpu_counter variables · Eric Dumazet <edumazet@google.com> · 2018-09-09
[LKP] [net/sock] b99259a614: netperf.Throughput_Mbps -6.6% regression · kernel test robot <hidden> · 2018-09-18

From: Eric Dumazet <edumazet@google.com>
Date: 2018-09-06 19:34:15
Also in: linux-crypto, linux-sctp, lkml

On Thu, Sep 6, 2018 at 12:21 PM Olof Johansson [off-list ref] wrote:

Today these are all global shared variables per protocol, and in
particular tcp_memory_allocated can get hot on a system with
large number of CPUs and a substantial number of connections.

Moving it over to a per-cpu variable makes it significantly cheaper,
and the added overhead when summing up the percpu copies is still smaller
than the cost of having a hot cacheline bouncing around.

I am curious. We never noticed contention on this variable, at least for TCP.

Please share some numbers with us.

`h`	back out one level
`j`	next message in thread
`k`	previous message in thread
`l`	drill in
`Esc`	close help / fold thread tree
`?`	toggle this help