| |||||||||||||
Linuxant releases DriverLoaderLinuxant releases DriverLoaderPosted Oct 14, 2003 4:19 UTC (Tue) by BrucePerens (guest, #2510)In reply to: Linuxant releases DriverLoader by mmarq Parent article: Linuxant releases DriverLoader Well, think of all the work that has gone into optimizing the network in 2.4 and 2.6 . A lot of this was concerned with zero-copy I/O where DMA happens almost directly from the disk controller to the network controller, optimizing the interrupt path, detaching entire busses while keeping the kernel running, and so on. Those are all very close to the hardware. Last year, we wouldn't have known how we were going to code some of this. We could not have forecast all of the things we'd need in such a driver API, and it would not stay the same from month to month if we wanted to continue development. Why do you think Linux performs 2.5 times as well as Windows Server 2003 while serving Microsoft's own file and printer-sharing protocol? Because we have control over things that Microsoft doesn't. What is the number one reason for Windows crashes? Badly-coded third-party device drivers. Much of Linux performance and reliability comes from the fact that we do control our device drivers. If we settle for less, we will have taken a big step toward becoming "just another run-of-the-mill operating system". Bruce
(Log in to post comments)
Linuxant releases DriverLoader Posted Oct 14, 2003 23:56 UTC (Tue) by mmarq (guest, #2332) [Link] One thing that amazed me in Linux from the start was his versability!So, there isnt nothing that prevents a stable and more static API for a split driver model where is wise to do it, and the actual paradigma, at the same time in other parts, where its better. I belive one thing is certain, there isnt a clear remedy for all deseases, and even with a large adoptance of a split model, the best argument stays with the "feature maintainer" and his experience and know-how. I belive something has to change, because as i state above that 150Mb of 2.5.* is going to take near to 2.5 years(?) to get to a decent stable 2.6.0., then, if the flush of documentation happen truth and the main kernel source gets "every driver on the planet", how many years will take about 500 Mb(example) ? Mario Marques
|
|||||||||||||