Updating our TCP and socket sysctl values...

[b. f.]

On 8/25/11, b. f. <bf1783 at googlemail.com> wrote:
>> >> I believe it's time to up these values to something that's in line with
>> >> higher speed
>> >> local networks, such as 10G.  Perhaps it's time to move these to 2MB
>> >> instead of 256K.
>> >>
>> >> Thoughts?
>> >
>> >
>> > This never happened, did it?  Was there a reason?
>> >
>>
>> I went back and looked at the mail thread.  I didn't see any strong
>> objections
>> so I think you should commit this for 9.x.
>>
>> np@ did point out that nmbclusters also lags on modern hardware so
>> consider upping
>> that at the same time.
>
> I thought Bruce's observation, in:
>
> http://lists.freebsd.org/pipermail/freebsd-arch/2011-March/011193.html
>
> that:
>
> "...there is an mostly-unrelated bufferbloat problem that is
> purely local.  If you have a buffer that is larger than an Ln cache (or
> about half than), then actually using just a single buffer of that size
> guarantees thrashing of the Ln cache, so that almost every memory access
> is an Ln cache miss.  Even with current hardware, a buffer of size 256K
> will thrash most L1 caches and a buffer of size a few MB will thrash most
> L2 caches."
>
> , and his suggestion for some sort of auto-tuning, deserve
> consideration.  Are you going to address this problem (at least the L2
> and higher cache thrashing), or give some suggestions for tuning in
> UPDATING and the relevant manpages?

Sorry, I should have sent this to -arch instead of -current.

b.