OSNews

Ok, I submitted the DF articles and received no credit for the postings. Glad to see OS news is on their toes.

There were multiple submissions I molded into one item, and as such, I would have to have listed all three. Since I can only list one, I found it unfair to pick one and discredit two others. But, if you are hell-bent on it, I will add your name to the item.

Even though I don't really see the point in adding the name you used to submit the items... "Anon_Poster". But, as you wish...

Edited 2005-11-25 21:50

Ahhh! Good. You cleared up my confusion.

Speed was one of the things that sucked and some people said it was inherent to Microkernels but regardles Mach is probably a good kernel for research and that is all this is.

It says the first experiments worked but were slow, they incorporated some changes and it then sped up 3 fold.

The speed problems in Microkernels are not inherent, they seem to be related to using a synchronous API with them (like any *nix).

Edited 2005-11-26 15:21

Umm... DesktopBSD looks nice.. that's their impression!

Yes I'm sure there are, but, the whole point of DesktopBSD is as a desktop os. This brings the user interface into one of the primary features of the OS. Granted KDE is a desktop system that can be used on any OS and of course the default theme on DesktopBSD can be changed to anything you want. But the fact that the default theme right off the bat has alot going for it is quite important for a desktop os.

yes but most reviews of an OS fall into the category of "hey it looks nice and its KDE or Gnome"

I agree with that.

Anyway, where did they get RC3 from? On DesktopBSD's web site all I see is RC2.

Go to www.distrowatch.com and look in the bottom left hand corner. The link to rc3 is there. It could be elsewhere but that is how I downloaded it the other day.

It wouldn't make sense for Apple to re-port the kernel and mold it into something they can use again, as opposed to just improving what they already have.

Well, I'm not saying to dump what they have, I should have re-phrased it better; are they going to adopt some introduced ideas from 4.0 and port them back into 3.0 or have they moved MacOS X so far away from the original 3.0 idea that it barely resembles a Mack kernel?

The main reason is because if there have been definate improvements of 4.0 over 3.0 - would it be in Apples best interests to bring alot of the 4.0 features accross.

With that being said, however, it would be nice if Apple were a little more open in their vision of the next version of MacOS X - Leopard, and the areas they are focusing on; I can assume, with pretty much certaintity tha they'll be *really* focusing hard on scalability as to ensure that the OS is as threaded and fine grained as possible as to take advantage of dual core (in the future quad core) processors as they become a more regular site on desktops - but then again, its only speculation.

Mach 4.0 is a pretty significant change from Mach 3.0, and Apple's version of 3.0 has been quite heavily customized. Moving it to Mach 4.0 would be possible, but would probably be so much work that it'd make more sense at that point to just ditch XNU and use something like FreeBSD instead.

However, I don't think there is a lot for Apple to gain from focusing "hard" on fine-grained multithreading for Leopard. You don't neeed sophisticated fine-grained locking to take advantage of dual-core (or even quad-core) CPUs. You need them to take advantage of 16+ way server machines. Once you've gotten the low-hanging fruit, fine-grained locking in an OS is complicated, and there is little use for it on a desktop OS like OS X.

...Altho' they ARE trying to carve a niche in the server market as well. And after that Ars Technica (I think?) article slamming the OS X kernel as the problem with it's horrible lack of scalability, you would think they would want to do something about their threading/locking mechanisms.

"well. And after that Ars Technica (I think?) article slamming the OS X kernel as the problem with it's horrible lack of scalability"

No, It's Anandtech. Ars Technica doesn't make crap like this article.

True, the benchmarks used were also brought up, for example, the MySQL issue was more a situation of HFS+ being over caucious when it came ot reading/writing data - a number of Apple and MacOS programmers pulled it to bits.

Now sure, there are some improvements that need to be made, but I think the article painted MacOS X to be worse than it really was.

Well, good, actually. I did not see those followup articles. It should have been more loudly defended. I am. btw, a Mac user and have no issues.

In some ways it did, but the lmbench numbers are particularly damning. Lmbench is a very simple benchmark, but since UNIX is a very simple system, it gets to the heart of a lot of UNIX performance issues. It'll tell you what the basic primitives of your program cost, and on OS X, these primatives are very expensive.

However, I don't think there is a lot for Apple to gain from focusing "hard" on fine-grained multithreading for Leopard. You don't neeed sophisticated fine-grained locking to take advantage of dual-core (or even quad-core) CPUs. You need them to take advantage of 16+ way server machines. Once you've gotten the low-hanging fruit, fine-grained locking in an OS is complicated, and there is little use for it on a desktop OS like OS X.

The issue isn't so much fine grained so that it can scale to the moon and back; the issue is scalability so that MacOS X adequately takes advantage of dual core AND processor Mac's. That need to get that sorted out, whether its massive kernel fine grain or replacement of parts - thats up to the engineers, but the fact remains that the performance boost should have been alot better.

To take care of dual-core to quad-core Macs, Tiger's course-grained multithreading probably should have been adequate. With only four possible concurrent processes, you don't need a lot of seperate locks in the kernel to ensure that they don't hold each other up.

Even so, one has to realise that there will also be quad cores on the books, so looking at 2 years into the future, how about a PowerMac with 8 cores all together? like I said, there has to be something in the OS that is holding it up on an SMP configuration.

There has been several benchmarks and there are some scalability issues with in MacOS X when compared to Linux - which includes issues like slow thread creation etc. Its not about bashing MacOS X, but trying to find what is causing the problem(s) and addressing them in a timely manor.

I believe it's Mach+Lites, there is no Mach-lite, even if it sounds interesting.

But this is FreeBSD-related news not the general BSD news. That's why the FreeBSD's new logo is used. Find that our by clicking on the logo!