Messages in this thread | | | Date | Tue, 15 Sep 1998 22:23:10 -0700 | From | "David S. Miller" <> | Subject | Re: RTO [was Re: my broken TCP is faster on broken networks [Re: Very poor TCP/SACK performance]] |
| |
Date: Sat, 12 Sep 1998 18:01:59 +0200 (CEST) From: Andrea Arcangeli <andrea@e-mind.com>
And I just seen that 2.0 increase srtt of 2*MDEV and not of 4*MDEV:
--- 2.0.35 --- } else { /* no previous measure. */ sk->rtt = m<<3; /* take the measured time to be rtt */ sk->mdev = m<<1; /* make sure rto = 3*rtt */ } --------------
2.1 instead calc mdev as m<<2.
And this probably is the cause of the stalls reported by many people on the list.
No, the real fix for the stalls will be in 2.1.122, the linear increase case of congestion window calculations was completely wrong due to a change I made circa 2.1.92 which is reverted now in 2.1.122
Setting the initial mdev to m<<1 is a bug in the BSD code which we just copied blindly in 2.0.x, the calculation in 2.1.x (m<<2) is the correct one, check the RFCs :-). I believe this is mentioned in Steven's books and the tcp-impl "known TCP implementation problems" draft, amongst other places.
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 Please read the FAQ at http://www.tux.org/lkml/
| |