auto tuning tcp

[Alfred Perlstein]

On 11/12/12 10:01 AM, Andre Oppermann wrote:
> On 12.11.2012 18:43, Alfred Perlstein wrote:
>>
>>
>> On Nov 12, 2012, at 1:27 AM, Andre Oppermann <oppermann at networx.ch> 
>> wrote:
>>
>>> On 12.11.2012 09:52, Alfred Perlstein wrote:
>>>> On 11/11/12 11:28 PM, Andre Oppermann wrote:
>>>>> On 12.11.2012 08:10, Alfred Perlstein wrote:
>>>>>> I noticed that TCBHASHSIZE does not autotune.
>>>>>>
>>>>>> What do you think of the following algorithm?
>>>>>>
>>>>>> Basically round down to next power of two based on nmbclusters / 64.
>>>>>
>>>>> Please wait out for a real fix of the various mbuf-whatever tuning
>>>>> issue I'll propose shortly.  This approach may become inapproriate.
>>>>> Also the mbuf limits can be changed at runtime by sysctl.
>>>>>
>>>> What is the timeline you are asking for to wait?
>>>
>>> http://svnweb.freebsd.org/changeset/base/242910
>>
>> Very cool!
>>
>> So instead of nmbclusters, will maxsockets work? Ideas/suggestions?
>
> I've already added the tunable "kern.maxmbufmem" which is in pages.
> That's probably not very convenient to work with.  I can change it
> to a percentage of phymem/kva.  Would that make you happy?
>

It really makes sense to have the hash table be some relation to sockets 
rather than buffers.

If you are hashing "foo-objects" you want the hash to be some relation 
to the max amount of "foo-objects" you'll see, not backwards derived 
from the number of "bar-objects" that "foo-objects" contain, right?

Because we are hashing the sockets, right?   not clusters.

Maybe I'm wrong?  I'm open to ideas.

-Alfred