Re: [Chromium-dev] BufferPool problems, cont

[Mike H. <mho...@gr...>]

I'm fine with the packer making things fit in the correct MTU, I just=20
have a problem with the MTU being reset to larger (or smaller) than the=20
user specified.

I'm not convinced the UDP code actually even needs this.  Say I want a=20
1MB MTU, I can do the calculation so to find an MTU to specify to the=20
tilesort/pack SPU that will end up at the 1MB MTU I really want.  If I=20
do this, doesn't the code still work?  If so, then it begs the question=20
why we are changing the MTU size in the first place, especially before=20
the UDP code can ever realize that it is going to have to adjust it's=20
MTU to the network.

-Mike

Samuel Thibault wrote:

>Hi,
>
>Le ven 12 sep 2003 08:24:17 GMT, Mike Houston a tapot=E9 sur son clavier=
 :
> =20
>
>>Now if only the packer=20
>>could initiate a send in the correct MTU/TCP window chunks while fillin=
g=20
>>the buffer.  That one gets REALLY tricky.
>>   =20
>>
>
>Just wondering... Isn't this precisely what I tried to do for UDP ?
>Fitting buffers to the window may not be wise, since it may vary
>a lot, but at least the buffer's mtu may be set to it. In a way,
>we would really get to a reliable packet transmission protocol :)
>
>Regards,
>Samuel Thibault
> =20
>