From: Willy Tarreau
> Sent: 10 August 2020 12:47
> On Sun, Aug 09, 2020 at 06:30:17PM +0000, George Spelvin wrote:
> > Even something simple like buffering 8 TSC samples, and adding them
> > at 32-bit offsets across the state every 8th call, would make a huge
> > difference.
>
> Doing testing on real hardware showed that retrieving the TSC on every
> call had a non negligible cost, causing a loss of 2.5% on the accept()
> rate and 4% on packet rate when using iptables -m statistics. However
> I reused your idea of accumulating old TSCs to increase the uncertainty
> about their exact value, except that I retrieve it only on 1/8 calls
> and use the previous noise in this case. With this I observe the same
> performance as plain 5.8. Below are the connection rates accepted on
> a single core :
>
> 5.8 5.8+patch 5.8+patch+tsc
> 192900-197900 188800->192200 194500-197500 (conn/s)
>
> This was on a core i7-8700K. I looked at the asm code for the function
> and it remains reasonably light, in the same order of complexity as the
> original one, so I think we could go with that.
>
> My proposed change is below, in case you have any improvements to suggest.
>
> Regards,
> Willy
>
>
> diff --git a/lib/random32.c b/lib/random32.c
> index 2b048e2ea99f..a12d63028106 100644
> --- a/lib/random32.c
> +++ b/lib/random32.c
> @@ -317,6 +317,8 @@ static void __init prandom_state_selftest(void)
>
> struct siprand_state {
> unsigned long v[4];
> + unsigned long noise;
> + unsigned long count;
> };
>
> static DEFINE_PER_CPU(struct siprand_state, net_rand_state) __latent_entropy;
> @@ -334,7 +336,7 @@ static DEFINE_PER_CPU(struct siprand_state,
> net_rand_state) __latent_entropy;
> #define K0 (0x736f6d6570736575 ^ 0x6c7967656e657261 )
> #define K1 (0x646f72616e646f6d ^ 0x7465646279746573 )
>
> -#elif BITS_PER_LONG == 23
> +#elif BITS_PER_LONG == 32
> /*
> * On 32-bit machines, we use HSipHash, a reduced-width version of SipHash.
> * This is weaker, but 32-bit machines are not used for high-traffic
> @@ -375,6 +377,12 @@ static u32 siprand_u32(struct siprand_state *s)
> {
> unsigned long v0 = s->v[0], v1 = s->v[1], v2 = s->v[2], v3 = s->v[3];
>
> + if (++s->count >= 8) {
> + v3 ^= s->noise;
> + s->noise += random_get_entropy();
> + s->count = 0;
> + }
> +
Using:
if (s->count-- <= 0) {
...
s->count = 8;
}
probably generates better code.
Although you may want to use a 'signed int' instead of 'unsigned long'.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT,
UK
Registration No: 1397386 (Wales)
