Advanced System Building Guide

Alan writes "FiringSquad has up an Advanced System Building Guide, detailing how to construct your own rig. The first half deals with hardware selection and even esoteric concepts such as PCI slot placement. The second half is focused on Windows XP, and makes recommendations such as moving the swap file and scratch disk to a separate partition." From the article: "You laugh at the so-called expertise of Best Buy's GeekSquad, and are the one doing the teaching when calling technical support. If this sounds like you, you've come to the right place if you're looking to take your system building skills to the next level."

Comments

Speedy Delivery.

[Anonymous Coward]

Move your swap file to it's own SCSI disk (small one)

Yes, reducing

[2names]

wear and tear. Each time the machine has to swap or write temp files, the physical moving parts of the hard drive experience wear.

If you can reduce the amount of this wear on your OS and data drives by placing swap and temp on a physically seperate drive, you may prevent major data loss.

I would think this would be obvious, but I guess not.

Re:Yes, reducing

[winkydink]

Insignificant in the MTBF calculation. Ask a hw engineer. The rotational assembley that spins the platters (the speed of which is constant) is by far the biggest failure mechanism.

Re:Yes, reducing

[ashmedai]

I hang out over at StorageReview, and a while ago there was a post where someone did so. The feedback amounted to that:

1. Concentrated seeks from pagefiles etc do not negatively impact the hard drive's life span because the head does not ever come in physical contact with the disk, and in fact
2. Concentrated reads/writes actually increase the hard drive's reliability because it remagnetizes that region every time it is written to.

Re:Yes, reducing

[Taladar]

Your second point is bullshit. If you constantly write a sector, let us say it is the page-file/partition you probably won't need access to this data after several months of not touching it which is about the only situation where remagnetization helps.

Re:Yes, reducing

[RapmasterT]

sure, you could reduce wear and tear on your OS drive by splitting the data out across as many drives as you can stick in there.

However, due to the realities of MTBF, every drive you add increases your chances of a catastrophic failure. If you don't have a real performance reason for adding spindles, more drives is just more points of failure.

Re:so sad

[pianoman113]

I suspect that hdd brand choice could spark a small-scale religious war.

I've had great success with almost every brand out there (those that I've tried, have worked great), and I've seen spectacular failures with most of them.

Re:Take the article with a grain of salt

[pg110404]

Reduce wear and tear?

I agree. To avoid wear and tear, it's better to have the swap file on a separate harddrive whose main task is that alone (perhaps a drive for periodic archival?). That would also give the extra performance.

If you need the extra performance by moving the swap, moving it to a separate partition will just slow everything down because the head has to move further on the platter to get there. If it's interspersed among your data, the chances it needs to hunt for the right track is that much reduced because it's already pretty close to being there already. If you're not actively using more virtual memory than physical ram, where the swap space is doesn't do a whole lot of difference because you're not doing a whole lot of swapping.

A dedicated drive gives the speed AND longevity.

Yes, mechanical parts WILL wear out

[WidescreenFreak]

You need to remember that hard drives are NOT solid state devices. They have bearings and mechanical parts. The first rule of thumb when it comes to PCs or any kind of equipment is that "The question is not if the parts will wear out but when the parts will wear out."

That being said, the hard drives will wear out. Period. End of story. Some might die in a few months, some in a few years, and some might never die before you replace them.

Even more important is the conecpt of multiple spindles to do multiple jobs. If you have one drive that suddenly hits swap because you're doing something, not only will your system grind to a halt because the drive head is loaded with contention (it can only do one job at once, obviously) but you're adding that much more wear and tear.

With the swap on a separate drive (and preferably on a separate IDE channel, assuming that that's what you're doing), the main drive can do whatever it needs to do while letting the other drive take care of the swap. So, not only are you greatly increasing potential throughput and system efficiency, you're dramatically reducing wear and tear on the drive head mechanism.

Re:Yes, mechanical parts WILL wear out

[Fizzlewhiff]

Yeah but he's talking about partitioning a drive to save wear and tear. It is still the same drive. Would you partition a drive into two equal partitions and mirror them for redundancy? Makes no sense.

Re:Yes, mechanical parts WILL wear out

[StarsAreAlsoFire]

Actually, grand parent is right. Two parts with a MTBF of, say, an hour, used in the same system at the same time, sum to a MTBF of (slightly) less than 1 hour. Not directly, which would imply an MTBF of 1/2 an hour, but they sum.

Look at the space shuttle: No single part has an failure rate worse than once in 100K launches (IIRC; it may have been one in a million. It's in the design specs)*

Now, there are some odd million parts. WHOA! No, you don't get a failure every launch, but the failure rate is WAY higher than on in a million -- think the RFP/design specs 'required' a one in 10K chance of failure.

The reason for the discussion was based on some of the 'design requirements' floated about for the next-gen 'shuttle replacement' -- one of which was a 1 in a million chance of failure -- thus necessitating a piece-wise failure rate of around 1 in a billion.

And what in the world does the Monty Hall problem have to do with this?

Try: Math World

and:
NASA

And the most directly applicable:
Hotwire article

Or just consider how they test for MTFB: Take 1000 parts. Run them until all of them die (not really for hard-drives, but this is how you do it for REALLY IMPORTANT things:~} ). Now plot the distribution and take the mean.

Cheers,

* yes, I am a rocket scientist, and this was discussed in classes I took years ago.

Re:Yes, mechanical parts WILL wear out

[Kiryat Malachi]

Given N parts, each with a probability of P of failure on a given day, with the probabilities independent the correct equation for calculating the probability of *any* system failure occurring is as follows:

(1-P) is the probability of a given widget surviving the day.

(1-P)^N is the probability of *all* widgets surviving the day.

1-((1-P)^N) is the probability of a failure within the system on any given day - this may be a single failure, or multiple failure, but is most assuredly a failure.

For a P 1/3 and N 3, the Ptotal is 0.70, which is indeed only slightly above 1/3. This is true. It is also misleading. The complexity gain is much more noticeable with more realistic failure probabilities - given a P of 1/10, and 3 parts, your failure rate suddenly goes to 0.271, near to tripling your failure rate. As you approach the limit of reliability (lower and lower P), your gain approaches the number of parts in the system; essentially, with sufficiently reliable parts, each additional part *does* more or less additively affect reliability. The bound for additive behavior is roughly (1/P) >= 9.6N, for a 5% bound (additive to within 5% accuracy, in other words).

Re:Yes, mechanical parts WILL wear out

[WidescreenFreak]

Actually, what you said is only partially true. You're discounting power saving features in most PCs and PC-based operating systems that can shut down idle drives. By pushing swap/virtual memory to another drive, that drive will probably stay up and running due to paging in/out, but the main data drive would have a much better chance of being spun down when not in use. This happened multiple times on a PC that had three drives in it. The main drive (which was also swap) never slowed down, but the two data drives frequently had the chance to shut down, sometimes for hours on end.

Re:What the heck is this supposed to mean?

[Anonymous Coward]

It means if your job depends on color accuracy, get a Mac and not a Windows or Linux based PC.

I just built my system--Lessons learned

[angle_slam]

Things that I learned from building my XP system:

• He talks about installing SP 2 after installing XP. That's fine, if you have an SP 1 CD. But if you have a pre-SP1 CD like I do, XP will not recognize any hard drive space over 127 GB. You can't partition it or anything. XP thinks the drive is 127 GB and you're stuck. The solution (and probably a better idea even if you have an SP1 disc) is to Slipstream SP2 onto your XP install disc. Here is an explanation of the process. Basically, you integrate SP2 into XP and burn a new CD. So when you install XP, it is automatically SP2 and recognizes the full size of the hard drive.
• My system would not Standby properly. The fans were still on, which defeats the purpose of putting the system into Standby. You have to go into the BIOS and enable S2 or S3 Standby mode if you want the system to appear off in Standby mode, but still have 5 second startup.
• For some odd reason, my motherboard BIOS didn't have USB 2.0 defaulted on. I have no idea why they would do that. Make sure it is changed to enable USB 2.0 support.
• Don't forget the Administrator password. I had to do a reinstall because I forgot it. Luckily, I hadn't transferred any info at the time.

Re:I just built my system--Lessons learned

[smitten0000]

You don't necessarily have to reinstall if you forget your Administrator password. Check out the following utility:

http://home.eunet.no/~pnordahl/ntpasswd/

Re:I just built my system--Lessons learned

[thecue]

Don't forget the Administrator password. I had to do a reinstall because I forgot it. Luckily, I hadn't transferred any info at the time.

There's a Linux distribution for that.

Re:I just built my system--Lessons learned

[scott_karana]

>Don't forget the Administrator password. I had to do a reinstall because I forgot it. Luckily, I hadn't transferred any info at the time.

I've almost done this myself a few times, but I googled around and discovered Peter Nordahl's 'Offline NT Password & Registry Editor', which can just reset the Admin password and avert the problem of reformatting.

Re:Take the article with a grain of salt

[poot_rootbeer]

Yes, the extra ten minutes you need to spend going through config dialogs every time you upgrade to a larger hard drive (and how often does one do that? About once a year? Once every two years?) is more than enough justification to subject your system drive to more "wear and tear" every day.

Disappointed. I assumed an article on "advanced" system building would include a lot more work with a soldering iron and tin snips.

Depends on what you're doing

[temojen]

If you regularly edit several photos at a time (or do video editing), you can have a GB or two and still hit swap. Or if you use Linux it'll automagically pre-emptively write any inactive pages to swap incase it needs to free them (this is a good thing).

Re:so sad

[badasscat]

No matter which HD brand you recommend SOMEONE is going to tell you they had bad luck with them. I've actually had fairly GOOD luck with Maxtor.

Ditto. I've got two Maxtors running right now with no problems whatsoever - neither even gets warm to the touch, and they are both inaudible to boot.

Like you, I have had one go bad on me in the past, but then I've also had two WD's and a Seagate go bad on me, and know numerous people who've had IBM's, Samsungs, and other drives go bad too. It's sort of a badge of honor to have a drive go bad - you're not a real geek if it hasn't happened to you yet. But it really doesn't matter who makes the drive; their failure rates are pretty similar (with a few notable and notorious exceptions - the IBM DeathStar drives, for example, though these were simply defective).

I don't think anybody who saw my house, with its four networked PC's, two of which are scratch-built (one of which is technically 15 years old!), one of which is controlling all my media viewing, would question my geek credentials, and I've got no problem with Maxtor at all. It's almost like a form of nerd prejudice if you really think one drive maker is significantly worse than any other - it can't be based on anything real.

Re:Here's what I think

[schapman]

I always put my system specs in my sig on hardware forums (a basic one at least) and I do it for good reason. It lets everyone know what I'm using if I'm asking questions, and that I have experience with certain things when answering theirs. I will however, agree with you on the people who list every last litte detail.

Re:Um... swap file?

[SScorpio]

I've need seen or heard of an actual one made up of placing actually SIMMS or DIMMS, but what you are wishing for are called solid state hard drives.

I did a quite Google of the term and got http://www.bitmicro.com/products_edisk_35_ide.php.

I also found a dicussion on Sharky's forums from back in 2001 about this very issue. I doubt we'll ever see one, but you never know what those crazy people in Hong Kong will hack out next.

Re:Take the article with a grain of salt

[mrjackson2000]

The Ars System Guide gives 3 levels of systems, all built with resonably good hardware

Re:Um... swap file?

[badasscat]

I don't have the details handy, but running a swap, even if RAM is bountiful and plenty is always a good idea. It's something to the effect of the system really likes seeing the swap there, even if you technically don't need it.

It's more that it's good to have it there just in case, because you never know when you will need it (even with 2GB, you can multi-task yourself straight to hell if you're doing image editing, watching videos, and running crap in the background all at the same time), and it doesn't hurt anything to have it enabled. If your system doesn't need it, it just won't use it, so no use disabling it. But that one day when you run out of RAM in a very bad way because you've disabled your swap file could kill you (or at least your data), depending on what you're doing. Windows PC's do not like it when they run out of memory without expecting to.

There's something of a myth that some people believe in that Windows is constantly accessing your swap file even with loads of RAM, and that turning the swap off will force Windows to use your RAM. Well, a) Windows XP is pretty good with memory management, and doesn't use swap when it doesn't have to, and b) even if it did use swap too much, turning it off isn't going to "teach" the OS to use memory properly. It either needs the swap file or it doesn't, and if it doesn't, what do you have to lose by leaving it on?

Re:RAM Drive

[kryptkpr]

I believe you seek one of these Speed is limited mostly by the PCI bus..

Re:so sad

[Anonymous Coward]

Take a look at the Reliability Survey on http://storagereview.com/

Windows doesn't handle swap that well.

[khasim]

Start it up and it will soon start swapping. No matter how much RAM you have.

Since it is going to do it anyway, you'll want a nice, clean, ORGANIZED place for it to do it in.

The problem is that adding a partition usually puts that partition near the spindle which is the SLOWEST portion of the disk. But it will still cut down on fragmentation and crap.

With a Linux system, I put the swap drive down first. It gets the fastest portion of the disk. It should never use it, but just in case ...

With Windows, if you do that you'll end up installing Windows to D:\, which is fine, but you'll need to make adjustments everytime something wants to install to C:\program files.

Re:The guide is useful for those who don't know...

[Sweetshark]

Things can go a few certain ways, but there's only one right way.
Oh, really?

"Rig" has a long history w/ electronic gear...

[Ellis D. Tripp]

at least among ham radio operators. One's transmitter has been referred to as a "rig" since the beginning of the hobby.