Messages in this thread | | | Date | Fri, 27 Mar 1998 22:38:01 +0100 (CET) | From | MOLNAR Ingo <> | Subject | RE: The GGI and EvStack debate -- Linus and such persons please reard. |
| |
On Fri, 27 Mar 1998, Jordan Mendelson wrote:
> Right now, graphics cards are one of the few devices which must be accessed > directly from userspace with no kernel abstraction layer. How can anyone > expect to run a real multi-user server when you have multiple programs > trying to access video cards at once? While we are at it, why don't we pull > the entire network card subsystem and put that into userspace too and see > how long that survives in a real multi-user system. [...]
the problem is: video cards are 10 times more complex than networking cards, they do 100 times more operations per second, and have 1/100th of the latency. And, their perceived speed is 10 times more important to Joe Average User.
unfortunately, kernel entry speed is constant, no matter what device you use.
these 'small' _fundamental_ differences make GGI a _very_ _very_ _very_ delicate and sensitive hack, which has to be designed _very_ _very_ carefully. It's not a simple 'ah, why not put it into the kernel, it's a hardware resource, isnt it?' decision. It might even be that it's not possible to put it into the kernel, without accepting unacceptable tradeoffs. Whats the problem with this, it's a _technical_ question. The answer is yes or no, just _please_ dont try to answer that question without thinking about the issue. [note that i'm not trying to answer the question, i'm just trying to show you that it's not trivial to answer it at all.]
and yes, _even_ networking drivers are being put into user-space to get more speed and smaller latencies: take a look at Beowulf and U-Net ...
-- mingo
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu
| |