Technology for Mortals

A while back (summer of 2003), I spent my days off working at a small computer shop in the middle of nowhere. We were deep in rural South Carolina. As in, “can’t get cell phone reception” and “literally a one stop light town.” The shop itself was so small that it occupied a spare bedroom and the living room of the owner’s house, and I was the only employee. Before I started working there, I had always assumed that the “big companies” like Dell, Compaq, and so on were rip-offs. I had always assumed (and I may note, it had always been true until sometime early this decade) that building your own PC was cheaper than buying a true OEM machine. Well, by the end of that summer (believe me, our customers were not “high end” or even “mid-range”), I realized that the white box market was dead for the most part on the low end. The OEMs were selling a full machine (including a copy of Windows and shipping/handling) for less money than the parts would cost us. Ever since then, when someone has asked me if they should custom build a PC or buy an OEM PC, I tell them to go with the OEM model unless they have very specific needs that can’t be met.

Well, I have always considered my own needs very specific, so I never looked at OEM stuff before. For example, when I built my current PC almost three years ago, very few people ran dual monitors and that was a “must have” for me. In addition, I knew I wanted two RAID 1′s, and most OEMs were using motherboards without built-in RAID support. So I did it myself. Now, I am getting ready to replace this machine (technically, I will be turning the current machine into a server, and building a new desktop PC). After all was said and done, my parts list came in at just under $1,500. I could trim the price a bit, by dropping the thumb drive for ReadyBoost (I personally like ReadyBoost, I think it is an excellent idea), getting rid of the extra drive for backup purposes (not a good idea to drop it), halfing the RAM to 6 GB (do I really need 12?), reusing an existing case/PSU (instead of getting the Lian-Li/Seasonic combo I wanted), and so on. But I really don’t want to do that. I spec’ed it a certain way for a reason.

Remembering my PC shop experiences, I decided to see if the “big boys” could spank white box on the high end and well as the low end. I was stunned with what I saw. I looked at the “workstation” class PCs, since that is what I am building. First of all, what the OEMs had to offer was a lot of last year’s tech. On top of that, things that I thought would be common, were either not possible or extremely expensive (like RAIDs). And upgrades were unbelievably pricey. Basically, there is no way that the OEMs can beat a self-built system on this score.

Now, I do understand part of it. IT departments value consistency in the supply chain more than the best value; being able to swap a user’s drive into a replacement box and troubleshoot the hardware later will get a high-salaried engineer back to work a lot quicker than trying to solve the problem on their live box. Ditto for being able to just blast an image onto a machine. At the same time, I am baffled as to how the OEMs can think that they can charge what they charge for, say, a RAM increase, when every customer has a CDW or Newegg account. And of course, they may be willing to concede a lot of the market with the belief that most people looking on the high end are enthusiasts who always want a cutting edge, custom system that no OEM could profitably make. All the same, I was quite shocked to see just how bad the prices were on their high end kit.

J.Ja

Over the weekend, I took the time to (finally) work on my mother’s PC. One thing she had mentioned, is that it was a bit sluggish. Given the PC’s specs (nothing fantastic, but definitely mid-range), and her usage of it (typical usage), there was no reason in the world to think that it would be slow. When I took a look at it, I saw a whopping 10% – 20% (constantly bouncing) CPU usage on both CPU cores!

My initial reaction was to blame Symantic Utilities, which were installed and running. But, deciding to be sure before I started pointing figers, I did a touch of digging around (I cheated, and used Process Explorer. Lo and behold, it was the HP crapware that was clobbering the CPU! This piece of junk application, “HP Total Advisor” was constantly beating on Windows WMI sub-system (the sub-system that does things like show CPU usage, RAM usage, etc.) so hard, that WMI was cranking at 10% – 20% usage. Insane! Closing the main window had no effect, but quitting the system tray icon did the trick. I quickly found out how to disable this garbage.

Granted, the software did do some somewhat useful stuff for a less-than-savvy user. It was a nice little “dashboard” showing things like if it had been too long since the last backup, or when the last virus scan was performed. But the overall resource consumption was off the wall. It’s as if it was continuously monitoring all of these items, instead of refreshing the display periodically.

The funny thing is, 10% – 20% usage on a modern CPU isn’t that big of a deal, in and of itself. But it presents two problems. The first is “green IT”. The power saving systems on many systems don’t kick in unless the CPU is idle. Additionally, even if the system does step itself down, why should you be running at 10% – 20% CPU 24/7 for no reason? That’s more CPU consumption than the majority of servers out there are running. Heck, that makes this dinky desktop PC a bigger power hog than probably 90% of the computers on the planet.

The other issue is simply one of performance. Yes, your computer still has plenty of horsepower left over. But it means that for every little thing, there is CPU contention. In other words, you need to timeslice for everything. One of the wonders of multi-core CPUs, is that one task can jam a CPU core, and the rest of the system runs smooth as silk, since the other processes can still do their work on the free core(s). But when each core has even a small amount of constant activity on it, all of a sudden, things get slow. In others words, on a dual-core system, 1 core at 100% CPU usage slows the overall system significantly less than both cores at 50% usage, and probably beats even 25% usage per core too.

Why? Well, the devil is in the details of timeslicing. Essentially, timeslicing allows a CPU core to interleave computations of different processes, in a multiplexing fashion (similar to, say, a T1 line). But when it switches between processes, it needs to move all of the data for the task it is no longer working on to RAM, transition the new task’s information to the CPU, and continue. It’s a pretty slow process. Depending on how small the slices are (the smaller they are, the more overhead is dedicating to the slicing process), it is possible for the CPU to be spending more of its power doing the slicing itself than actually processing things.

When you are timeslicing, the sum of the parts is much less than the whole, thanks to this. You can keep a CPU saturated at 100% usage with, say, 5 tasks trying to run at 10% each (caution: those are fake numbers to illustrate my point). So when HP’s crapware decides to use 10% – 20% CPU usage on all cores, it’s like having a CPU that is about 70% or so as powerful as what you really have.

I’m not a huge fan of crapware. Usually, I can tolerate it. But when it decides to take that much CPU, it is inexcusable.

J.Ja