<html><head>
<meta content="text/html; charset=ISO-2022-JP" http-equiv="Content-Type">
</head><body bgcolor="#FFFFFF" text="#000000"><br style="font-family:
Courier New,Courier,monospace;">
<br style="font-family: Courier New,Courier,monospace;">
<span style="font-family: Courier New,Courier,monospace;">Yasuhiro
Orange Morishita / $B?92<BY9((B wrote:</span>
<blockquote style="font-family: Courier New,Courier,monospace;"
cite="mid:20130911.022021.46342557.yasuhiro@jprs.co.jp" type="cite">
<blockquote type="cite"><pre wrap="">And I know the IP specification defines the minimal MTU size to 576.
So, we may need a very short RFC for updating the definition of MTU,
</pre></blockquote>
<pre wrap=""><!----> ^
to 1280
-- Orange
</pre>
<blockquote type="cite"><pre wrap="">in RFC 791.</pre></blockquote>
</blockquote>
<br style="font-family: Courier New,Courier,monospace;">
<span style="font-family: Courier New,Courier,monospace;">i do not think
that the definition of mtu is wrong. if i were going to update 6891-bis
(which is itself 2671-bis) the logic i would draft is:</span><br
style="font-family: Courier New,Courier,monospace;">
<br>
---<br>
<br style="font-family: Courier New,Courier,monospace;">
<span style="font-family: Courier New,Courier,monospace;">"A DNS UDP
responder shall, when transmitting a message which does not include
embedded cryptographic signatures such as TSIG or DNSSEC signatures, use
an effective DNS message payload size which is calculated as
MIN(OFFERED, MIN(DISCOVERED, ESTIMATED) - OVERHEAD)) where OFFERED is
the EDNS BUFSIZE received from the initiator, and DISCOVERED is the path
MTU if known or else the outbound interface MTU, and ESTIMATED is
chosen as 576 for IPv4 or 1280 for IPv6 as the minimum guaranteed size
of an IP datagram, and OVERHEAD is chosen as 64 for IPv4 or 48 for IPv6
as the maximum likely size of the IP and UDP headers.<br>
<br>
This specification does not define a maximum for any future IP transport
protocol, and so both initiators and responders should be prepared to
receive messages as large as the 9 kilobyte ethernet jumbogram size in
preparation for future transport protocol development."<br>
<br>
---<br>
<br>
i'm trying to decide whether that "should" ought to be a "must". and, i
know the number is 9K not 64K, because of buffer memory considerations
on today's embedded servers.<br>
<br>
vixie<br>
<br>
</span>
<blockquote style="font-family: Courier New,Courier,monospace;"
cite="mid:20130911.022021.46342557.yasuhiro@jprs.co.jp" type="cite">
</blockquote>
</body></html>