> > By the way, I think TCP_SACK_MAX (declared in tcp_var.h) should be 4.
>
> Yes, but I was working under the assumption that if a flow is SACK capable,
> it will more than likely also be using the timestamp option.  It's trivial
> enough to checdk for it, I suppose.

In my experience, Windows XP does not use the TCP Timestamps option by
default, while it uses the SACK Permitted option.  That says, in a case
where IE6 on a Windows XP machine accesses to Apache on a NetBSD
machine, the TCP connection uses SACK while it does not use Timestamps.

Although there would be only limited situations where IE6 would send
large data to a web server (e.g., uploading large pictures), I think
it is better to send SACK blocks as much as possible for such case.

Regards,
Noritoshi Demizu