[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <48dab68e-2c10-444c-a219-0e1c05091573@gmail.com>
Date: Mon, 18 Mar 2024 03:20:33 +0000
From: Pavel Begunkov <asml.silence@...il.com>
To: Jason Xing <kerneljasonxing@...il.com>
Cc: netdev@...r.kernel.org, edumazet@...gle.com, davem@...emloft.net,
dsahern@...nel.org, pabeni@...hat.com, kuba@...nel.org
Subject: Re: [PATCH net-next v2] net: cache for same cpu
skb_attempt_defer_free
On 3/18/24 02:44, Jason Xing wrote:
> On Mon, Mar 18, 2024 at 8:46 AM Pavel Begunkov <asml.silence@...il.com> wrote:
>>
>> Optimise skb_attempt_defer_free() when run by the same CPU the skb was
>> allocated on. Instead of __kfree_skb() -> kmem_cache_free() we can
>> disable softirqs and put the buffer into cpu local caches.
>>
>> CPU bound TCP ping pong style benchmarking (i.e. netbench) showed a 1%
>> throughput increase (392.2 -> 396.4 Krps). Cross checking with profiles,
>> the total CPU share of skb_attempt_defer_free() dropped by 0.6%. Note,
>
> I suspect that we can stably gain this improvement. The reason why I
> ask is because it might be caused by some factor of chance.
I guess it might be the kernel is even booting only by
some factor of chance, but no, the testing was quite stable :)
>> I'd expect the win doubled with rx only benchmarks, as the optimisation
>> is for the receive path, but the test spends >55% of CPU doing writes.
>
> I wonder how you did this test? Could you tell us more, please.
Well, the way I did it: choose a NIC, redirect all its IRQs
to a single CPU of your choice, taskset your rx side to that
CPU. You might want to make sure there is no much traffic
apart from the test program, but I was lucky to have 2 NICs
in the system. Instead of IRQs you can also try to configure
steering.
I used netbench [1] for both rx and tx, but that shouldn't be
important as long as you drive enough traffic, you can try
iperf or something else.
[1] https://github.com/DylanZA/netbench
>> Signed-off-by: Pavel Begunkov <asml.silence@...il.com>
>> ---
>>
>> v2: pass @napi_safe=true by using __napi_kfree_skb()
>>
>> net/core/skbuff.c | 15 ++++++++++++++-
>> 1 file changed, 14 insertions(+), 1 deletion(-)
>>
>> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
>> index b99127712e67..35d37ae70a3d 100644
>> --- a/net/core/skbuff.c
>> +++ b/net/core/skbuff.c
>> @@ -6995,6 +6995,19 @@ void __skb_ext_put(struct skb_ext *ext)
>> EXPORT_SYMBOL(__skb_ext_put);
>> #endif /* CONFIG_SKB_EXTENSIONS */
>>
>> +static void kfree_skb_napi_cache(struct sk_buff *skb)
>> +{
>> + /* if SKB is a clone, don't handle this case */
>> + if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
>> + __kfree_skb(skb);
>> + return;
>> + }
>> +
>> + local_bh_disable();
>> + __napi_kfree_skb(skb, SKB_DROP_REASON_NOT_SPECIFIED);
>
> __napi_kfree_skb() doesn't care much about why we drop in the rx path,
> I think. How about replacing it with SKB_CONSUMED like
> napi_skb_finish() does?
I'm sticking here with the current behaviour, __kfree_skb(),
which it replaces, passes SKB_DROP_REASON_NOT_SPECIFIED.
I can make the change if maintainers ack it, but maybe
it should better be done in a separate patch with a proper
explanation.
--
Pavel Begunkov
Powered by blists - more mailing lists