<html>
<head>
<meta content="text/html; charset=utf-8" http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
On 20/01/16 15:04, <a class="moz-txt-link-abbreviated" href="mailto:dpreed@reed.com">dpreed@reed.com</a> wrote:<br>
<blockquote cite="mid:1453302261.02626297@apps.rackspace.com"
type="cite">
<meta http-equiv="Context-Type" content="text/html; charset=UTF-8">
<p><a class="moz-txt-link-freetext" href="http://arstechnica.com/gadgets/2016/01/numbers-dont-lie-its-time-to-build-your-own-router/">http://arstechnica.com/gadgets/2016/01/numbers-dont-lie-its-time-to-build-your-own-router/</a></p>
</blockquote>
<br>
Definitely a missed opportunity :), many of those nice fast
connections are unfortunately over-buffered.<br>
<br>
I think it's interesting in its own right. The 10kB x 100 row is
horrible, and it's not that far-fetched for modern bloated web
pages. A new frontier for the modern speed test :).<br>
<br>
My reflex was to ask what's actually being measured. Is it faster
with IPv6 (no NAT)? Would the connection tracking overhead still be
significant? (I also hear Dave screaming about offloads, but I
don't think that's why we see (low-power) Ivy Bridge cpu brought
down to only 200mbps of packet forwarding in the last row).<br>
<br>
Stateless firewalls should work quite well for TCP, you just drop
incoming SYN. It's connection-less UDP that breaks it. (And often
you still don't need high-volume connection-less, but you want
unreliable datagrams and the network wasn't designed with security
in mind...).<br>
</body>
</html>