| If we set s to first packet size then s is way too low as this is not a
| data packet. On my system this is 24 bytes (skb->len) and if sending
| 1400 byte packets you are sending at a rate over 50 x too much.
|
| More conservative default is mss and follows general principle of being
| conservative rather than aggressive.
I don't agree with this patch.
1) all packets going through ccid3 are data packets
(restriction len > 0 in ccid3_hc_tx_send_packet).
2) your example is specific to iperf, but with other sources the initial
packet may look different: we should not modify generic kernel code just
because with a specific application it looks strange.
3) we already have the larger initial windows code tackling this issue:
as soon as the first feedback packet arrives, the
min(4*MSS, max(2*MSS, 4380))
kicks in - which has an even stronger effect than just using 1 * MSS.
OK?
|
| Signed-off-by: Ian McDonald <ian.mcdonald@xxxxxxxxxxx>
| ---
| diff --git a/net/dccp/ccids/ccid3.c b/net/dccp/ccids/ccid3.c
| index 1ae4f52..b40cad6 100644
| --- a/net/dccp/ccids/ccid3.c
| +++ b/net/dccp/ccids/ccid3.c
| @@ -342,7 +342,7 @@ static int ccid3_hc_tx_send_packet(struct sock *sk, struct sk_buff *skb)
| ccid3_hc_tx_set_state(sk, TFRC_SSTATE_NO_FBACK);
|
| /* Set initial sending rate X/s to 1pps (X is scaled by 2^6) */
| - ccid3_hc_tx_update_s(hctx, skb->len);
| + ccid3_hc_tx_update_s(hctx, dp->dccps_mss_cache);
| hctx->ccid3hctx_x = hctx->ccid3hctx_s;
| hctx->ccid3hctx_x <<= 6;
|
| -
| To unsubscribe from this list: send the line "unsubscribe dccp" in
| the body of a message to majordomo@xxxxxxxxxxxxxxx
| More majordomo info at http://vger.kernel.org/majordomo-info.html
|
|
-
To unsubscribe from this list: send the line "unsubscribe dccp" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
