Math intensive server stuff
I'm working on (or more accurately about to start) a very math intensive client server system, where the server has to do a metric ass-load of calculations mostly on 64-bit signed integers on behalf of client machines. The data are all going to be in ram, and multi-cpu support is a good thing.
Would you recommend a PPC machine over a x86 machine for a task like this? I guess this is mainly a chipset/etc... question, but i have been unable to find that sort of information elsewhere, and i figure who better to ask, 'cause you probably have a decent gut-feeling for how the architecture works in practice on real-world data =:-)
Wow, nothing like a nice, simple question to start your day. ;-D
If you're going to be working with 64-bit numbers, I would try to do a real-world test. That is, get Athlon, Alpha, Pentium 4, and PowerPC 7400 (G4) machines, and try some tests to see how quickly each of them handle integers that big. I know the PowerPC 7400 (G4) has an awesome vector calculation module (128-bit vector registers can be fun), but that's not the same thing as integers. The 7400 doesn't have any specialized integer units, whereas the Athlon and P4 do. (I'm pretty sure of this, but you should definitely do your homework before making any decisions based on this information.) It might be great for decompressing video, but for hard-core (sounding) tasks like handling 64-bit signed integers, I'd definitely test everything possible. Definitely try the Alpha and MIPS, too. Alpha's sort of known for handling these sorts of things.
I'd also ask on the linuxppc-devel mailing list. People that probably can help you are on that list. (http://lists.linuxppc.org/)
Re:Math intensive server stuff
One other important thing to ask is the state of the GCC complier for the PowerPC Platform. IIRC, it isn't as efficient as the ones available for the x86 and Alpha platforms. How much would LinuxPPC benefit from an optimized compiler and what sort of performance could be expected from LinuxPPC compared to Linux86/Alpha/others?
Franz Sirl, a Linux/PowerPC developer, has done a lot of work on optimizing GCC for PPC. Look at PPC vs. x86 benchmarks. Theoretically, PowerPC kicks x86's butt, but if things aren't optimized for it (as often happens in this x86 world), it may not seem like such a hot processor. With more optimization, performance should continue to improve. Again, I'd ask about this on the linuxppc-devel list.
Is LinuxPPC a viable alternative to x86 Linux? Can I run my department on a LinuxPPC-based server with the same peace-of-mind that I get on an i386-based box running Linux? Will I still enjoy the almost surrealistic uptimes I get with my current Linux server? Does the LinuxPPC code still suffer from chronic flakey-ness?
I'm currently looking into obtaining a PowerPC box to test out the current state of Linux on the PPC platform.Hopefully your answers will point me down the path of RISC utopia.
That's a good place to be. Of course it's an alternative. For some of us, it's reality. ;-) You can run on a PPC box as well as an x86 box. I know of many places that are using PPC boxes for everything from basic stuff like web servers to netatalk servers to controlling puppets. (Jim Henson's Creature Shop!) There are a lot of real world examples of Linux/PowerPC in action. We have a few in our office. Most of our servers have the same legendary uptimes that Linux is known for. Servers outside our office have them. Other people's servers have them. It's everywhere. ;-)
The PowerPC 604 is an incredibly stable processor for Linux use, and the 750 (Apple's G3) is very solid. The 7500 (G4) is getting there fast.
I'm not sure what you mean by "chronic [code] flakey-ness". Code not optimized for PPC? Yes, though not as much now. Unstable code? Yes, but that seems universal. A lot of times, it can be traced to simple things like bad RAM. Try replacing the RAM if your machine is acting quirky. It could make a big difference! (He says to an audience in which 512 MB of RAM may not be uncommon....)
ATX motherboard availability?
I think one thing that would foster Linux PPC adoption, and PPC in general would be a relatively cheap PPC motherboard. I remember IBM released their reference design some tine ago and there was some noise from 3rd parties about product anouncements, but nothing materialized. Does anyone know when we might see something?
Oh, that would rock if those came into reality. I do know about it but I can't comment about it beyond saying that I hope it actually happens.
What I can say: it needs to happen.
merge with RedHat?
by A moron
I've tried LinuxPPC several times over the years and have actually been disappointed. It just hasn't seemed polished and LinuxPPC, the company, has had some serious customer service problems.
Have you ever thought about or actually talked to RedHat as making LinuxPPC the RedHat Distro for PPC?
Yes, and I can't comment about that. In September of 2000, we committed to customer service, as things really were bad. But then, a major hunk of the company was unable to think straight, or remember what he was just about to say. (That "hunk of the company" == me!) The good news is that customer service is still a major priority, and I'm well enough that I can see to it that it stays that way, and make sure that people are helped.
Your Perception Before and After the Accident
First of all, I'd like to commend you and your wife for your courage and determination through your ordeal! I also hope they throw the book at the jerk who caused the accident!
My question is this: Do you find that your perception of the world and what your interests, passions and abilities are, different than before your accident? Has the accident changed your interests towards the computing industry?
First off, thank you. :-) I've been through hell, many times. People don't know what it's like to have a traumatic brain injury, or permanently lose vision in an eye. It's not fun. ;-)
Say, Cassie (my wife) is a major hero here. Thanks to everyone who wrote in and expressed support for her. Do it some more, she'd appreciate it. ;-D
They will be throwing the book at "Jerk Boy" (my name for the ...drunk.... who hit me). He faces three felony charges, including driving under the influence and driving with intent to cause harm. Considering that he had a 0.25 BAC (blood-alchohol content), I don't know how they could defend against it.
Yes, my injuries changed a lot of things, including how I see the computing industry. Part of me realizes how big a help it's been (and at the same time a bloody pain the ass!), and part of me wants to get out of it. To me, living well (being content with your situation, loved ones, etc.) is much more important than having the latest box, biggest monitor, or best domain name is. (It was that way before, too, but now it's even more so.)
Linux and Accessibility
During your recovery period, did you find the need to use any accessibility tools to accomplish tasks? If so, what were your impressions? Does Linux have the tools people with alternative interface needs (like text-to-speech) need to access their information?
I didn't have any experience with that stuff, unfortunately. I think the State of Georgia's health department may have been able to help me with a bit of that, but I'm sure it all would have been for Windows or the Mac OS. (And useless to me.)
A moment of victory for me was about four months after I got hit; I was back at home, and still was able to use vi. I'm not sure what that says about the effects of a brain injury. ;-D
Altivec and MP G4's?
ok, newbie questions but:
1)How much can a PPC linux distro can benefit from Altivec optimization?
2)Does LinuxPPC enjoy the same degree of improved performance from additional processors that OS-X does?
AltiVec: Apps that use features that AltiVec can help will benefit. The OS itself probably won't get much benefit. AltiVec was designed to help things like video and audio, things that people in Apple's markets are interested in. Linux doesn't have much software that can benefit from that. I don't know how much an AltiVec-friendly kernel or version of Apache would improve things, if at all. As Linux gets more "desktop" and "multimedia" software, AltiVec support will probably become a more interesting topic for Linux.
Oh, you could probably jigger Enlightenment to use AltiVec.. oh my...
SMP support is still improving. Until recently, there wasn't much in the way of SMP hardware out there. Before Apple introduced the multiprocessor G4s, there simply wan't that much at all. There were a number of MP PPC 604 machines, but they're no longer in production, unless you find an obscure Motorola box.
With Apple (the major PPC player) making MP boxes, MP support will improve. With Apple making boxes that have AltiVec, support for that'll improve too. Assuming that they keep using AltiVec.
Why should my next purchase be a PowerPC?
Intel hardware is a commodity; it's cheap, there are lots of peripherals for it, you can buy individual components and build your own box easily, and prices are very low.
AFAIK (which isn't far), PowerPC hardware is mostly proprietary, controlled by Apple, is more expensive, has less variety in peripherals, and you're more or less stuck buying a Macintosh just to get your PC. Not just that, but many components of many PowerPC-based computers have marginal to no support under Linux (USB is marginal, Firewire is nonexistent right now, etc).
Given all this, where is the major win in the PowerPC? Why ought my next purchase for a PC be a PowerPC running LinuxPPC/Yellow Dog/MkLinux?
I'm not trolling here; I'm just uneducated. :) Educate me.
Very good question! You're right: unless you get an oddity of a system, you're buying a Mac to get a PPC box. The TiVo is a Linux/PPC box, but I don't think that it can run Apache. ;-) (then again, Paul Mackerras and company have added an Ethernet card... anyway...)
For what it's worth, USB on Linux is coming a long way (printers now work, for example), and FireWire's getting there.
The advantage of PPC over x86 is/was that at a lower clock speed, you got higher performance. But 1 GHz PCs are coming out left and right, and Apple's fastest is 500 MHz. For some things, they will be the same speed. But you can get a Celery (er, Celeron) box super-cheap, with a lot more stuff that you get with an iMac. It's becoming quite difficult to have a viable alternative when you can't get a super-cheap box.
PPCs are smaller, faster, and cooler (literally and figuratively) than x86 chips. The PPC 7400 doesn't need a CPU cooling fan. Get a Mac (I'd name some other cheap PPC boxes, but there aren't any!) and you'll get a super-cool case and the cool-looking one-button optical mouse.
Where's the real advantage, though? Again, good question. As the computer world changes, and no third party desktop PPC boxes appear, it's getter harder to answer. I think the real problem here is that there are no real third-party alternatives for PPC hardware. And that needs to change. Should IBM's POP board see the light of day, ask me again, and I'll have a different answer for ya. As I said above, it needs to happen.