What do we all look like

[Pradeep]

Fortunately without the inherent doubling of failure rate

 

[e_dawg]

But if you and others are suggesting that there is no extra responsiveness during multitasking, then my whole point of wanting a dual CPU setup is gone.

I'm afraid I sort of see dual CPUs as being the IDE RAID0 of extra day to day performance, if you get my drift.

Ouch. A damning statement against dual CPU's if there ever was one. I spit on IDE RAID and the very silicon its controllers are made out of.

 

[James]

Oh, it's stable and all that - not like IDE RAID0 - but I was alluding more to all those SR newbies putting in RAID and expecting their day to day performance to go through the roof, and it doesn't. It's the same with dual CPUs IMO. If you have a specific use that is helped by 'em, by all means go for it. If you're just an average user looking for the Tim Allen "more power" experience, you're looking in the wrong place.

 

[Mercutio]

"Real men sleep on rocks."

No no no.

Real men sleep on jagged rocks.
Or broken glass.

Or beneath real women.

 

[Cliptin]

"Real men sleep on rocks."

No no no.

Real men sleep on jagged rocks.
Or broken glass.

Or beneath real women.

Or red ant hills!

 

[e_dawg]

Oh, it's stable and all that - not like IDE RAID0 - but I was alluding more to all those SR newbies putting in RAID and expecting their day to day performance to go through the roof, and it doesn't. It's the same with dual CPUs IMO. If you have a specific use that is helped by 'em, by all means go for it. If you're just an average user looking for the Tim Allen "more power" experience, you're looking in the wrong place.

No traditional dual CPU applications in mind (like rendering, for example), but I very much dislike how my computer is unresponsive when the CPU is pegged at 100% by various processes while multitasking. Something I do quite frequently is compiling articles for research in Acrobat files. I frequently encounter 100% CPU usage and a semi-frozen computer when I am (1) web or paper capturing to Acrobat, (2) opening webpages that have JavaScript pop-up windows, and (3) saving Acrobat files. It is frustrating having to wait anywhere from 5-20 seconds for my computer to "unfreeze" and respond to my inputs before I can do anything. This, to me, is multitasking by water torture. If a dual CPU setup will allow me more responsive multitasking and minimize those "frozen CPU" moments, that is all I am asking for.

What say you, James?

 

[Tea]

Your problem, o doggy one, is not the lack of a second CPU. It is the lack of a decent multi-tasking OS. :wink: All jokes aside, no version of Windows that I have ever seen can multi-task particularly well.

 

[time]

That statement will probably earn Cas' ire.

 

[e_dawg]

Your problem, o doggy one, is not the lack of a second CPU. It is the lack of a decent multi-tasking OS. All jokes aside, no version of Windows that I have ever seen can multi-task particularly well.

We shall have no comments from those running obsolete OSes like Warp!

... but yes, I agree with you to a point. I switched to W2k from 98SE for several reasons, including access to the far superior NT kernel with pre-emptive multitasking. I ran NT4 Server as a test case for a few weeks before determining that its multitasking capabilities were significantly better than 9x. I figured W2k Pro, being based on the same kernel, would be just as good. Sadly, I find it is not as good as NT4 Server. Maybe somewhere in between 9x and NT4. Strange. I have tried adjusting the priority boosting quanta in the registry without much success.

I wonder how those massive enterprise DB's running MSSQL on W2k seem to handle so much load so gracefully...

 

[The JoJo]

e_dawg, can't imagine you not benefitting from using a dual cpu system from the sound of your problem.

On the other hand, you might want to switch over to linux and not upgrade your system , that's what I did.

 

[Tea]

It may well be obsolete, but it can multi-task far better than any Windows version ever did, o Doggy One. Or, no doubt, given the massive code-bloat that MS can't seem to avoid these days, ever will. But yes: the multi-tasking advantage of the NT family Windows products over the 9X family stuff is substantial. I can't say I've noticed a great deal of difference between NT 4.0 and W2K though: perhaps because the only machine I have run both on is this home unit, and it is already massively over-powered for the work it does. I mean - what is the point of 512MB DDR, X15 and Athlon XP 1800 for a machine that spends its entire life running an email client and a handful of web browsers? In other words, I could run Windows 3.1 on this beast and, from the point of view of multi-tasking, probably not notice too much difference.

These days I run a mixed stable of machines: W2K here at home (plus the Linux-based firewall I built a couple of hours ago), ECS or OS/2 on most of the office machines (which I do expect to be able to multi-task flawlessly on, and don't expect to ever crash), NT 4.0 on a workshop machine, 98SE on a little office box that I seldom use, even DOS 6.3 on the workshop DOS box, which plays audio CDs to amuse us and creates various floppy disc images on demand). The other machines I am responsible for (family ones) all run W98SE (because it's easy to fix) or NT 4.0 (because it doesn't often need fixing). But even now, I still snort whenever I hear a Windows user talk about multi-tasking.

PS: if you still think I'm an OS snob, well, you just ain't ever met a Linux user. And as for Mac users ... well ... We will say no more.

 

[e_dawg]

what is the point of 512MB DDR, X15 and Athlon XP 1800 for a machine that spends its entire life running an email client and a handful of web browsers?

It's one of life's luxuries to be able to do everything you want at top speed without ever having to wait for the computer. This, coincidentally, is my goal.

PS: if you still think I'm an OS snob, well, you just ain't ever met a Linux user. And as for Mac users ... well ... We will say no more.

My former boss was a Linux snob. And I have seen a bunch of people on SR that are strong advocates of Linux.

I don't consider myself an OS snob, as I have repeatedly tried to give Linux a chance (and will again in a few months), but I absolutely need to run W2k for its applications. I am literally dependent on Office, IE, and other such products and require full functionality for advanced features and formatting that are just not cross-platform compatible. And I have still found it to be the fastest and most stable of all the OS's I've tried (Linux incompetence notwithstanding).

As for Mac zealots, I find it amusing how they preach to me as I am now a "reformed" Mac user. Been there, done that. They really were superior IMO back in the days before Windows 95, but ever since then, there has been nothing but hype and empty promises until perhaps OS X 10.2.

 

[James]

No traditional dual CPU applications in mind (like rendering, for example), but I very much dislike how my computer is unresponsive when the CPU is pegged at 100% by various processes while multitasking.
[...]
This, to me, is multitasking by water torture. If a dual CPU setup will allow me more responsive multitasking and minimize those "frozen CPU" moments, that is all I am asking for.

What say you, James?

I say that Acrobat distilling is not dual processor aware (I just gave it a try) and is CPU bound. Ditto Javascript. You'd be better off buying either a faster CPU or two faster computers and networking them than a dual CPU machine.

Windows 2000/XP may be multiprocessor aware in the OS, but it seems IE isn't particularly - Java is particularly hopeless at using the second CPU I notice (Javascript is presumably similar, I can't find a page with enough Javascript to slow down a XP1700+ though).

Photoshop recognises the second CPU and I get about a 20% performance improvement when using large filters, and this is about the norm for multiprocessor-aware apps (ie. I could probably use a single XP2000+ or 2100+ instead and get the same result). The only thing that I have done that sees almost exactly a 100% performance improvement in a dual system is DiVX encoding.

 

[James]

One other thought.

Looking at your present system, if your .sig doesn't lie, you would get about a 10% preformance "feel" increase moving to a system with DDR RAM, and about 10% improvement in feel moving to an 800MHz box, then 10% more moving to 1200MHz, and then about 15-20% more going to around the XP1800+ or so mark. People will dispute my numbers I'm sure.

But my point is really that you can get a 50% or more improvement in the speed "feel" of your machine going to an Athlon XP-based system at the cheap end of things, which would be a pretty minimal expenditure (US/Canada-based people will be able to better estimate the exact amount).

 

[Cliptin]

My vote is for the javascript. My 1800+ bogs down for 5-10 seconds depending on how many windows are opening. This from a machine with 256 MB and only a mail client and one IE window open. Let me take that back. W2k is not slow: I can browse the start menu and windows explorer windows but IE is slow.

James, after you start Acrobat can you assign it to a particular CPU?

e_dawg, For minimal investment (under $150US) You can get a board that will support all your ram and an XP. Dozer just bought a board that that supports DDR and SDRAM.

 

[cas]

That statement will probably earn Cas' ire.

Tannin has been well schooled. He seems to be confused about a number of related issues, which makes it difficult for him to articulate a strong position on this matter.

 

[i]

No traditional dual CPU applications in mind (like rendering, for example), but I very much dislike how my computer is unresponsive when the CPU is pegged at 100% by various processes while multitasking.
[...]
This, to me, is multitasking by water torture. If a dual CPU setup will allow me more responsive multitasking and minimize those "frozen CPU" moments, that is all I am asking for.

What say you, James?

I say that Acrobat distilling is not dual processor aware (I just gave it a try) and is CPU bound. Ditto Javascript. You'd be better off buying either a faster CPU or two faster computers and networking them than a dual CPU machine.

Can someone explain how Acrobat not being "dual processor aware" impacts what e_dawg has stated his problem is?

James, you advocate a faster CPU as the solution. I don't understand how that's an effective solution to e_dawg's problem. For example, say he just picked up a single processor system that's 50% faster than the one he started off with. Perhaps being that much faster, the system will become more responsive to other applications he decides to launch while Acrobat is running.

Ok, sure ... assume the responsiveness does improve. But what happens on the very next day, when e_dawg installs another, even more CPU-intensive application? One that maxes out his new CPU? Or when he decides to run two instances of Acrobat at the same time? What do you suggest then, that he upgrades his CPU again?

It seems to me that, when you get to the level of application use that e_dawg is apparently at, you have to accept the fact that, sooner or later, you're going to run an application that peaks your CPU's capabilities. When that happens, you get an unresponsive system (unless you're running OS/2 or whatever).

That's where I don't understand what the problem is with a dual processor solution. I would accept the fact that the occasional application I run is going to max out one of the CPU's. Big deal. The OS will notice that one CPU is maxed out and simply run any other application I happen to run during that "high load period" on the second processor. Hence, the dual processor system will be consistently more responsive - and be capable of being that way for a longer period into the future - than the single processor system.

I'll assume I'm wrong about this ... someone tell me where.

 

[The JoJo]

I seem to think about this matter exactly as i.

So what if the program hogs CPU 1 for X seconds or minutes, as e_dawg can still happily use the computer because of CPU 2? Is there something I'm not comprehending?

Yes, you can have multithreaded programs (written some) , but you can also run 2 instances of a singlethread program when you have 2 cpus.

 

[flagreen]

Yes, you can have multithreaded programs (written some) , but you can also run 2 instances of a singlethread program when you have 2 cpus.

Or four instances if you have dual Prestonia Xeons.

 

[The JoJo]

Only in my dreams flagreen

 

[flagreen]

I'm sitting here typing while frame serving a movie I captured out of Adobe Premier into virtual dub which is encoding in Divx while at the same time ripping a movie with the smartripper. CPU usage is only 27%. There are very few multi threaded programs I've seen which can achieve 100% cpu usage on my machine (dual 2.2 prestonias) . Xmpeg and Tmpgenc being two which will when taxed enough. I can also run four instances of SETI at once which will also run at 100% cpu usage. I don't need the prestonias or even duals for the stuff I do with my PC but hey, other guys my age don't need Corvettes or Cabin Cruisers either, but they're sure nice to have.

 

[e_dawg]

Bill,

How much did that dual Prestonia setup cost you (could you break it down into mobo, RAM, CPU, case/PS)? I would like to see how far I would have to stretch (okay, explode) my budget... I haven't been able to find any dual Xeon mobo under $400, BTW. That seems to be the most expensive component.

James,

I am with i and The Jojo on this one (maybe the blind leading the blind, but the rationale seems logical). If Acrobat maxes out a single CPU, can I not use the other CPU to zip around in Windows and work on other apps? That is the whole point -- not making Acrobat run faster, but making everything else run "not as slow" when Acrobat is tying up CPU #1.

 

[flagreen]

Well I bought both the board and the cpus right after they became available and prices have droped since then. So I'll price them at todays prices. The 2.2 xeons are $311 each. The Supermicro P4DC6+ is $617 or $398 without the onboard scsi (P4DCE). 512mb RDRAM (2x256) $206 TOTAL. Lite-on FS120 case $125.00 and the PC Power and Cooling 450w PS was about $200. So the bottom end total would be $1551.00. Prices are from www.micropro.com

 

[James]

I am with i and The Jojo on this one (maybe the blind leading the blind, but the rationale seems logical). If Acrobat maxes out a single CPU, can I not use the other CPU to zip around in Windows and work on other apps? That is the whole point -- not making Acrobat run faster, but making everything else run "not as slow" when Acrobat is tying up CPU #1.

The problem is I think that you guys don't realise that XP doesn't (reliably)farm out CPUs to applications based on which is the least busy. (At least, I can't get it to do that unless I set processor affinity individually for applications beforehand based on the order I'm going to launch them.) It's not like there's a pool of twice the available CPU power which is doled out as required.

If one CPU is maxed out doing Acrobat distilling, you can certainly do other OS-related tasks using the other CPU, but that's a pretty small list. If you then start doing something complex in Word the chances are the system will chug just as much as if you had one CPU, 'cos the other one is essentially unused.

This is why having dual CPUs is particularly useful when you have one application that knows how to use both of them at the same time(database, Photoshop, DiVX...) and get tasks done quicker that way. It's not much use when you have multiple applications which aren't multithreaded or dual CPU aware. You may find that tasks get split across both CPUs but no more than 50% of either CPU gets used - in other words, it's of limited usefulness unless you're going to go back to your desktop and play around with it (since the OS is multithreaded) in which case the extra available CPU overhead indeed makes the desktop more responsive. But that's not really what you want, is it?

All this IMO, of course. There may well be a way to things the way you suggest, and indeed I think that's sort of how I imagined the world of dual CPU goodness would be before I got into it. It's just that it doesn't work that way for me.

 

[time]

I'd be interested to hear what Cas has to say about this. He has previously disparaged BEOS as offering no advantage over NT, yet I am under the impression that it was rather good at load balancing, and supposedly the API encouraged all apps to be multithreaded.

Cas, is the problem solely down to the way NT apps are written?

 

[flagreen]

My system never bogs down. Of course I probably don't run the intense apps that James does. But even so I have been running four instances of SETI in the past which results in 100% cpu usage and still no bogging unless I set the priotrity for SETI at the maximum level. From I've seen XP Pro does a pretty decent job of load sharing. It could also have to do with the differences in the chipsets and how they handle smp.

 

[James]

I'd be interested to hear what Cas has to say about this. He has previously disparaged BEOS as offering no advantage over NT, yet I am under the impression that it was rather good at load balancing, and supposedly the API encouraged all apps to be multithreaded.

Cas, is the problem solely down to the way NT apps are written?

I'm not Cas, but that's my reading of it. BeOS was designed from the ground up as a multi-CPU OS, NT was not - or, at least, it has to be backwards compatible with a lot of applications that are not.

Bill, SETI won't bog down your computer because the task is set as a low priority. Almost anything else takes precedence.

 

[Mercutio]

Without exception I can't think of an operating system that handles multiple CPUs really well and really gracefully. They're all basically retarded when it comes to that.

OK, since fushigi will probably come by and mention it, mainframes OSes do it pretty well (mainframes are lots of little CPUs).

 

[Tea]

Ahh, I was about to ask about mainframes. And in-between things of the AS-400 flavour, come to that.

 

[James]

If you know your target platform has multiple CPUs, you'll write your app to take advantage of them (or at least be multiple-CPU friendly). If you're unsure, you'll tend to write for the lowest common denominator. And with mainframes, if you're a contractor, you'll write whatever is the most complex and takes the longest time - that's something that works really well in a multiple CPU environment if at all possible.

 

[time]

:mrgrn:

 

[cas]

link

 

[Fushigi]

Without exception I can't think of an operating system that handles multiple CPUs really well and really gracefully. They're all basically retarded when it comes to that.

OK, since fushigi will probably come by and mention it, mainframes OSes do it pretty well (mainframes are lots of little CPUs).

I was keeping quiet, but since you mention it... The general rule is that on an AS/400 a.k.a. iSeries, SMP overhead is as little as 4% of each additional CPU. For instance, a dual-processor model 730-2066 has a performance rating of 1050. The 4-way version, 730-2067, has a rating of 2000, netting a loss of 4.76% due to SMP overhead. There are a handful of hardware white papers at http://www-1.ibm.com/servers/eserver/iseries/whpapr/#hardware although most are a couple of years old and could use some updating.

Of course, as has been said the gains from SMP also depend on application design. A single-threaded app will not generally benefit from SMP with the exception that the app isn't slowed down by whatever else might be running. We have a number of apps that, due to their design, can utilize only one of the four CPUs in our main system. However, if the app makes a system call, that system call can be SMP-capable. I've seen plenty of times where a dynamic index over a multi-GB file is built and all four CPUs crank to build it in just seconds.

In terms of multithreading, http://www-1.ibm.com/servers/eserver/iseries/whpapr/threads.html describes the threading architecture. IBM has benchmarked the prior generation of hardware at over 400,000 concurrent threads; the new hardware should up that by 80% or so. The only real limit is memory & storage.

- Fushigi

 

[cas]

The general rule is that on an AS/400 a.k.a. iSeries, SMP overhead is as little as 4% of each additional CPU. For instance, a dual-processor model 730-2066 has a performance rating of 1050.

Even in the case of applications with careful SMP support, this metric is very application specific. Certainly, very low common resource contention would be required to maintain an SMP penalty below 5% for an arbitrary number of CPUs. Any lower, and it would seem that multiple individual machines would be a better solution.

 

[cas]

IBM has benchmarked the prior generation of hardware at over 400,000 concurrent threads;

Certainly, you wouldn't want a system that limited threads in some arbitrary way, but I hope most readers realize what an absurd number even 40,000 threads is.

The ideal number of threads for a process in a 4-way machine is exactly 4.

 

[The JoJo]

I might be a bit out of the loop, but what do you do for a living cas? Or, how are you "into" threads/SMP/Clusters/mainframe etc program design?

After a few years with beowulf clusters and MPI programming, yup, 4 threads/4 cpus sound about right...

 

[cas]

...what do you do for a living cas?

I used to design real time operating systems for a living. More recently, I have worked on general purpose operating systems as well.

These days I am working on a low cost storage networking technology.

 

[Fushigi]

Certainly, you wouldn't want a system that limited threads in some arbitrary way, but I hope most readers realize what an absurd number even 40,000 threads is.

The ideal number of threads for a process in a 4-way machine is exactly 4.

Of course it's absurd. As far as I can tell, they did it to prove what the system can handle without barfing all over itself. The AS/400 has a history of being a closed, proprietary system. While that was true enough back in the early 90s, the iSeries today is the most open platform on the market bar none. Within a single chassis it can run multiple OS/400 invocations (including multiple OS versions), most AIX apps (true AIX coming by 2004), 64-bit PPC Linux, and multiple Windows NT/2000 servers all at the same time. The zSeries can't even do this.

But the image problem persists. A good deal of that is because IBM doesn't know how to properly market the machine. Benchmarking it helps (best JVM around, very high database TPM, award-winning Linux support, over 100,000 concurrent Notes email users on a single system while maintaining subsecond response time, support for all those threads, etc.) but they haven't managed to take the awards & benchmarks and present them in a way that will drive more business.

It's far and away the best selling midrange platform ever, but that's due to the applications, system reliability, security, and OS features. It's most certainly not IBM's marketing.

Security .. Klez just swept through our environment the other day. Dozens of NT/2000 boxes hit. System managers stayed up until 2AM fixing everything. Meanwhile, I was in Atlanta playing Whirlyball and having steak & drinks on the company. iSeries systems don't get infected with virii.

Oh well, enough ranting.

- Fushigi

 

[Buck]

A good deal of that is because IBM doesn't know how to properly market the machine.

If my memory serves me well, isn't this the same reason (as was explained to me by Tony and others) why OS/2 flopped? When will IBM learn their lesson?

 

[Tea]

Never. IBM are just like that. Always have been, always will be.

PS: Buck, just how do you define "flop"? By Microsoft standards, sure, but by any other standards it has been a resounding success. I don't think anyone could regard an OS that, at its market peak, was selling ten million copies a year, and which retains significant market share and is still selling at a slow and steady pace seven years after IBM were illegally forced to drop active development and marketing of it as a "flop".