Linked by Thom Holwerda on Thu 29th Mar 2007 09:00 UTC, submitted by zelko
Thread beginning with comment 225653
To view parent comment, click here.
To read all comments associated with this story, please click here.
To view parent comment, click here.
To read all comments associated with this story, please click here.
Holy sheep s... Batman!!! Read the paper - well written and quite an eye opener. Partially supports what I said about X, but not because of the network transparency thing. Also possibly explains why I have seen so many complaints posted about the slowness of Gnome ;-). More and more functions are being pushed out to user space, and if developers don't manage the frequency of system calls, we will all suffer.
While I'm at it, I should probably proactively dispel another popular one. When you look at how much memory X is using, keep in mind that you are looking at total mapped memory. The *vast* majority of it is *video* memory which, depending upon the driver, can be mapped 2, 3, or more times for different purposes.
Very little of it is system RAM.
Very little of it is system RAM.
There used to be a paper from John Carmack explaining that (back in the utah-glx days). Do you happen to know where I can find this? I'm tired of trying to explain this very fact to users who just won't accept what I'm saying :-)
Adam






Member since:
2005-07-24
"""
But who really uses that functionality in X today?
"""
Oh, for me about all 60 of the business desktop users that I happen to support. Probably the vast majority of Linux desktops in business.
Besides, the whole "X is slow because of network transparency" bit went out the window *years* ago when MITSHM, and later, DRI showed up.
I agree that X'less OpenGL is a good thing for some applications.
But I get so tired of hearing the same *wrong* claims made about X over and over and over and over again.
While I'm at it, I should probably proactively dispel another popular one. When you look at how much memory X is using, keep in mind that you are looking at total mapped memory. The *vast* majority of it is *video* memory which, depending upon the driver, can be mapped 2, 3, or more times for different purposes.
Very little of it is system RAM.
Xorg has its inefficiencies. But they are mostly implementation details and not fundamental design decisions.
Jeff Garzik touches on this in the context of a broader topic, in this bittersweet paper from last year's Linux Symposium.
http://tinyurl.com/357m7o
It's a good read. You'll laugh! You'll cry! You'll find out why your Core 2 Duo system takes so freaking long to perform certain more intensive operations... like shutting down.