>> It is hard for the DIPC development team to prepare the patches for the
>> new kernels.
>
>Exactly why I mentioned it.. along with shared pid space.. Both would be
>relitavly low cost to maintain within the kernel, but I'm sure the
>developers have a helluva time keeping up.
>
>Once these basics are in, its alot less work for people trying to do more
>advanced cluster stuff..
Quite right. Adding DIPC to the standard kernel will enable others to
port/write a whole new group of applications and tools to/for Linux.
This will be a _really_new_ feature, even relative to other UNIX variants.
It allows Linux to enter new territories, and will make it interesting to
some new categories of users.
>I've played with it a bit (earley releases) and I've found it to work
>rather well (well, a bit slow when I used it.. But hey it was alpha).
Going over the network is a guaranty that things will run slower. Not much
can be done about it. Using TCP/IP as the main networking protocol (for
reliability) and doing most of the work in the user space (to reduce the
needed kernel changes) also play some role. A bit of careful programming,
like transfering more data at longer intervals, overcomes many of the
problems.
>This really should be included..
-Kamran