Linked by Thom Holwerda on Thu 26th Jul 2007 16:01 UTC, submitted by SEJeff
Linux After years of being relegated to server racks and the desktops of ultrageeks, Linux is finally making some headway as a viable alternative to Windows on the consumer desktop. That's the optimistic message delivered by a newly energized contingent of Linux proponents. By employing the same consumer-friendly marketing techniques practiced by Microsoft, and by taking advantage of the rising popularity of web-based applications, Linux vendors are getting ready for what they say will be a wave of consumer interest in the free operating system.
Thread beginning with comment 258555
To view parent comment, click here.
To read all comments associated with this story, please click here.
RE: yay
by kaiwai on Fri 27th Jul 2007 01:00 UTC in reply to "yay"
kaiwai
Member since:
2005-07-06

if there's one thing Linux needs, it's *more* forks. The desktop and the server aren't mutually exclusive and using one mans experience, as prominent as it may be, to justify a split seems absurd. It's also not like his project was nicked without a decent replacement.


Take the scheduler for example, there is a weigh up between throughput and 'teh snappy' - if you tweak it for optimum throughput, its almost a guarantee that you'll end up with an operating system that isn't snappy enough for the end user.

It isn't about a fork/independence for each project, but for a realisation that there is no 'one size fits all' approach. If you're a vendor for the desktop and server, you'll need to have a kernel with different optimisations for different purposes.

I don't see it as a bad thing, it means that requirements for both end users can be focused on rather than it being a situation of ideas being at logger heads - one wants a feature for the server but could effect desktop performance and vice versa.

Edited 2007-07-27 01:06

Reply Parent Score: 3

RE[2]: yay
by butters on Fri 27th Jul 2007 02:33 in reply to "RE: yay"
butters Member since:
2005-07-08

Well, you can have pluggable schedulers just as Linux has pluggable elevators. But more importantly, ongoing work suggests that you don't need to hard-code for throughput or latency. You can make it a tunable.

The primary tunable in the new "fair" schedulers is granularity, which controls the amount of time that must pass before the scheduler can switch out the running task in order to maximize fairness. Raising this value leads to more throughput, while lowering it results in lower latency. If you keep the scheduling fair, a single tunable is all you need to dial-in your particular place on the throughput/latency trade-off.

The primary difference between a desktop OS and a server OS is what kinds of applications they run. A kernel shouldn't care what applications it runs. It's job is to make sure that resources are distributed equitably and efficiently among the applications. There is indeed a "one size fits all" solution, with strategically-placed knobs to control the few fundamental trade-offs.

Specialization at the kernel level is bad. Vista gives Windows Media Player 80% of each timeslice if it wants. Is that the way an operating system should work?

Reply Parent Score: 4