>> 	I'd like to propose to raise MSIZE (non-cluster mbuf size) to 256
>> 	for all ports.  IPv4 with IPsec should perform better (depending on
>> 	how much encapsulation you plan to add), and IPv6 should perform better.
>> 	comments?
>Do you have any statistics about the mbuf usage?

	based on statistics on transpacific backbone, on Jan 1 2000:
	http://tracer.csl.sony.co.jp/mawi/samplepoint-A/2000/2000010113

		# of packet	sum		coverage
 [   32-   63]:     457221	457221		21%
 [   64-  127]:     305510	762731		36%
 [  128-  255]:     118732	881463		42%
 [  256-  511]:      83766	965229		46%
 [  512- 1023]:     771087  	1736316		82%
 [ 1024- 2047]:     358876	2095192		100%

	we may need more detailed histogram, but anyway, this gives us some
	idea.

>Since our MINCLSIZE is now defined as (MHLEN + 1), I hope raising MSIZE
>will also help to reduce the memory usage for limited memory PC.

	when i commit some more changes to IPv4/v6 processing code, we will
	start using m_pulldown and MINCLSIZE will go badk to MHLEN + MLEN + 1.

itojun