Messages in this thread | | | Date | Sun, 22 Mar 1998 11:40:53 -0800 | From | "David S. Miller" <> | Subject | Re: bug in 2.1.89 include/net/sock.h? |
| |
From: "A.N.Kuznetsov" <kuznet@ms2.inr.ac.ru> Date: Sun, 22 Mar 1998 22:19:57 +0300 (MSK)
Just create sock_wfree_donotwakeup and skb_set_owner_[somewhat]
I hacked up something quick like this, it was relatively straight forward. But it seems to not make much of a performance difference from what I've checked so far. I'll do some more research into it before I jump to any conclusions though.
BTW Does someone remember why the check for wmem_alloc*2 < sndbuf was removed from write wakeup routine? It also would (hypothetically) improve scheduling, though could have bad some interaction with tcp logic. Maybe, tcp should have its own set of socket callbacks, which ignore sndbuf, but depend more intelligently on tcp state.
I think it disappeared when Alan moved all generic SOCK initialization things into core/sock.c
But just like the sock_wfree_donotwakeup() thing it may prove to be not much of a performance gain at all.
Later, David S. Miller davem@dm.cobaltmicro.com
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |