[ILUG] D11, (again, i havn't given up)

[Caolan McNamara]

On 09-Aug-99 Justin Mason wrote:
>Caolan McNamara said:
>
>> Well fair nuff, some digging throws up the showstopper problem that I can't
>> actually do it the 100% natural way, but have to instead fall back to some 
>> standard (ish) ipc route to communicate between the Server and Client
>(though
>> doors, see later on), do appear to do most of what D11 would require.
>
>BTW one thing you could do that would probably be of a higher speed than
>vanilla X11 would be to send data in plain structs without marshalling.
>
>I would imagine (I've never looked though) that X11 does some kind of
>marshalling of data to make it byte-order, byte-size, etc. independent.
>
>If D11 is just running on one machine, you don't need that kind of thing.
>You may even be able to write some of the data from client to server
>without a copy (apart from what TCP/IP itself does).

Yeah i reckon that incremental improvement in this order could be achieved.

a) device independant marshalling vs device dependant marshalling (funny eh?)
b) removal of domain socket vs just about anyother version of ipc methinks
c) implementation using doors as ipc (looks good i reckon),
d) per call custom optimizations for doors, once you are guaranteed to be on
the same server you can ditch loads of block data passing and move to pointer
depends how much of this actually exists.

>Dunno, but it's an idea,

I might take a few evenings off this week and take a look at how XFree and 
libX11 are organized and play with a few toy door client/servers. Maybe get
a few benchmarks out the door for calls similiar to how to X calls look. See
if I can find out why it hasn't been done before, theres surely a few 
showstoppers lying around that I havn't noticed yet.

C.

Real Life: Caolan McNamara           *  Doing: MSc in HCI
Work: Caolan.McNamara at ul.ie          *  Phone: +353-86-8790257
URL: http://www.csn.ul.ie/~caolan    *  Sig: an oblique strategy
Do we need holes?