Optimise Windows for dual-core

Jake Carroll gets under the hood of Microsoft’s finest and processes a few threads of his own.

DIFFICULTY Intermediate

You have a new Intel QX6800 in your box, right? Every man, woman and Morris has one. The truth is the hardware you have – in most situations – is akin to having a Porsche on a dirt road out the back of your uncle’s farm. All hotted-up, with no room to move. This month, we will show you a few ways to exploit the potential of your thread-eating, multi-core pride and joy through application, operating system and driver optimisation.

Processing in symmetry
SMP or symmetric multi-processing is the lifeblood of your new multi-core CPU. SMP – in this instance – means that multiple processor cores are directly connected to a single shared main memory region. This SMP architecture allows any one of your processor cores to work on any set of threads in the pipeline no matter where the data for that set of threads is located in memory. More importantly, with an appropriately written operating system and thread scheduler, your threads can be balanced across multiple cores at once.

Recent trends in benchmarking have shown us that while multi-core processors are benchmarking well in highly-threaded, floating point-intensive applications – video encoding, audio encoding, scan-line rendering et al – the processors have for the most part only shown small gains in the things that many computing enthusiasts care most about. Desktop applications, gaming and operating system performance tend to fall by the way-side. There are a few things we can do to rectify some of these performance discrepancies.

Tuning desktop applications for multi-core
Threads themselves are the separate programmatic running processes that are spawned by an instruction from your executable binaries. The more threads you can run at one time, within a single application, the better the ability of the application to take advantage of physical processor resources – within reason. If your applications are running on a single thread, the chances are they will still offload processing to more than
a single core – hence, you’ll rarely see any of your cores completely idle – but will rarely attain their maximum potential, load-balancing across several cores without intervention. It can, in some cases, be detrimental to application performance when poorly-threaded applications attempt to load balance across multiple cores.

Pump your process priority up
Like putting nitrous-oxide in your transport, changing process priorities has its fair share of risks. What it does is give your applications more of a chance to take advantage of multiple-core time, before other processes get close to fetch-execute-decode time. What happens when important system processes don’t make it on time? Do they stall? Our general advice is that increasing a running processes priority to real-time is not something you should do in mission critical situations under an NT kernel – WinXP, Windows Vista, Win2k3. 32-bit Windows operating systems schedule threads on 32 priority levels, numbered 0 to 31. The thread with the highest priority is sent to the execution stack first. As threads wait to be processed, their priority levels are increased. The base application priority determines the priority level at which a threads begin. The following table shows the priority levels for each of the definable settings, when you right click on a process in task manager.



Process priority table for the NT kernel
Try increasing the priority on your next DVD encoding session, or perhaps your next Adobe Acrobat Professional 8.0 large volume PDF writing session. You’ll find in such situations, while your processor usage will be increased across multiple cores, other aspects of system responsiveness may be degraded.