June 02, 2008 | Comments: (0)
The myth of "MinWin" and a thinner Windows 7
Urban legends are strange creatures. Even when they're exposed for what they are - tall tales seemingly "legitimized" through frequent retelling -- people continue to believe the lie.
Case in point: "MinWin." For months, so-called industry "experts" were speculating that Microsoft would make a clean break with Windows 7 -- that core elements of the OS would be rewritten from the ground up and that backwards compatibility would be relegated to the domain of virtual machines and emulation.
[ Add your Windows systems to the exo.performance community, plus monitor how they specifically perform, with InfoWorld's Windows Sentinel tool. ]
Central to this theory was "MinWin." Citing the now infamous "Eric Traut demo," they claimed as fact that Microsoft was retooling the Windows kernel to make it lighter and less monolithic. Never mind that doing so would likely break the entire Windows hardware/software ecosystem. "MinWin" was the future. It was new. It was "cool." And as any industry media professional will tell you, it's the "cool" new technologies that drive page views.
Of course, now we know better. The whole "MinWin" bubble burst last week when, through various Microsoft web postings and interview comments, it was revealed that Windows 7 would in fact be more akin to "Windows Vista Second Edition": An evolutionary update that builds upon the existing NT 6.x kernel architecture as manifested in Windows Vista.
Undaunted, the "MinWin" true believers continue to cling to the legend. "If not Windows 7, then some future version," they say. "MinWin is coming." In fact, it could be here "today" if Microsoft would just "strip away all the user-mode bloat they've tacked onto Vista and its derivatives."
That last point seems to be a common theme among the "MinWin" faithful: That if you could somehow "pare down" Vista, removing unnecessary background services and dumping all that flashy Aero baggage, you'd end-up with a much leaner OS, something comparable to Windows XP or even 2000. It's an intriguing idea, one that adds to the whole "MinWin" mystique. It's also patently false.
The truth is that, when you strip away all of the new services and UI flashiness, the core Windows Vista/7 architecture is still quite bulky -- much more so than Windows XP and/or 2000. People who claim otherwise simply haven't done their homework. Fortunately, we make it easy to do just that by providing tools and resources of the Windows Sentinel project
For example, if you start with a basic Windows Vista Business (SP1) installation, you are looking at a workload consisting of nearly 600 threads spread across some 60+ processes. Whittle this down a bit -- by disabling Superfetch, Indexing, ReadyBoost and a few other non-critical services -- and you can get the thread count down into the 450-500 range, 98 of which are, incidentally, owned by a single process: System (i.e. the Windows Vista "kernel").
By contrast, a default Windows XP Professional installation spawns just over 300 threads across roughly 40 processes. Whittle this down a bit and you can get the thread count well under 300. Cut the OS to the bone and you can get it into the low 200 range spread across 20 or fewer processes, with 57 of those threads belonging to the Windows XP kernel process (i.e. System).
Compare this to Vista which, when cut to the bone (every non-critical service disabled, all UI goodies turned-off), still spawns 41% more (340 vs. 241) threads spread out across 50% more (30 vs. 20) processes, and you see how much harder it is to put Vista on a "diet." The Vista kernel alone spawns 72% more (98 vs. 57) threads than Windows XP.
Even Windows 2008 running in its "Server Core" configuration -- a scenario often cited as a precursor to "MinWin" -- still spawns roughly 300 threads, and it doesn't even have a proper shell running (just a command line). And, of course, the "Server Core" kernel accounts for 98 of those threads -- just like Vista SP1.
Given the above, it really comes as no surprise that, even stripped bare and with all the extraneous UI fluff disabled, Vista still takes 40% longer to execute the OfficeBench test script when running against an identically configured Windows XP system (Office 2007 used in both test scenarios). You can't have your cake and eat it too. More concurrent threads (especially in the kernel) equal more potential CPU overhead, impacting linear performance.
Bottom Line: The idea that Vista's problems are entirely isolated to User Mode is pure rubbish. Vista is fatter all the way around, and this includes the kernel and its various outlying critical subsystems. To achieve anything close to "MinWin" you'll need to break all sorts of stuff that would render Windows essentially unusable. Which is why, despite all the fanfare and breathless pontificating, "MinWin" remains an academic exercise -- and why the bloated Windows Vista/7 kernel architecture, with its DRM hooks and bulky legacy constructs, is here to stay. Get used to it.
Posted by Randall Kennedy on June 2, 2008 01:23 AM
May 06, 2008 | Comments: (0)
Vista PCs: the hardware resources they are so hungry for
Cross-tabular analysis. It's one of those nerdy-sounding terms that statisticians like to use when expounding upon their latest data-mining gems. It's also the lifeblood of the Windows Sentinel project. We use cross-tabular analysis to extract all sorts of interesting statistics from the exo.repository, including how a platform shift can directly affect workload composition.
[ Add your Windows systems to the exo.performance community, plus monitor how they specifically perform, with InfoWorld's Windows Sentinel tool. ]
For example, did you know that Vista-based PCs are working harder than ever? Aside from the obvious fact that they run slower than they did under Windows XP (in many cases, by 40 percent or more), there are some fascinating changes going on under the hood. If you compare the typical workload of a Windows-based system, breaking it down by per-process thread count and CPU utilization, you see that a Windows Vista PC is using roughly 21 percent more CPU cycles per thread than a comparably configured Windows XP PC. This is in addition to the fact that Vista is spinning almost twice as many execution threads (something we discovered in our recent cross-generational analysis of Windows/Office).
Translation: Not only is Vista "fatter" than XP, it's also more demanding at a very fundamental level. Hence the need for beefier client hardware with lots of cores.
Figure 1 - Thread Utilization Index
The above tidbit is something we pulled from the exo.repository using that nerdy, cross-tabular analysis I mentioned earlier. Basically, we extracted the process list for each system in the repository (about 1800 at last count), then compared the total CPU time for all running processes with the thread count for same, factoring in the number of physical/logical CPUs. The resulting compound index - the Thread Utilization Index (TUI) - gave us a handy, single-number ratio that describes how a given workload is behaving at a system-wide level. And when we compared the average TUI for XP and Vista-based systems, we saw a 21 percent spike in the latter.
The great thing about compound indices like TUI is that they're dynamic: We've adapted the algorithm that generates the index to compensate for factors like multiple cores and hyper-threading -- things that can directly affect workload behavior/scalability. By abstracting the data from the platform particulars, we make it possible to compare seemingly disparate systems -- for example, a quad-core uber-workstation and an aging Pentium M notebook -- because the indices acknowledge the context in which they are generated.
Another good example is the Peak Memory Pressure Index. As we analyze the index's four contributing factors -- the Committed Bytes, Pages Input/Sec and Percent Page File Utilization performance counters, as well as the duration of the peak event -- we weight each and then evaluate them against the PC's physical memory capacity. This, in turn, allows us to make those difficult heterogeneous comparisons -- for example, the aforementioned Pentium M notebook with 512MB of RAM and the uber-workstation with 8GB. The raw data says that the latter is running a much larger workload, in terms of RAM consumption, than the former. However, in practical terms, both may be comparably burdened in terms of how their respective workloads affect the available physical hardware: Available RAM; paging activity; and so on.
Figure 2 - Peak Memory Pressure Index
Again, it's our ability to abstract the data from the hardware that allows us to glean all of these interesting tidbits. It's also how we can say with confidence that Vista-based PCs are experiencing roughly 30 percent higher memory pressure, across the board, than their XP brethren.
No question: Vista is considerably "fatter" and more demanding than XP. But you don't need lots of fancy abstract statistics to tell you that. Just boot the thing ... and wait ... and wait ... and wait ...
Posted by Randall Kennedy on May 6, 2008 02:45 AM
April 18, 2008 | Comments: (0)
Application usage summary: 4/18/08
In the first of our ongoing series on "who's using what" under Windows, we focus on the state of the Task List on our cross-section of instrumented systems:
[ Monitor your own Windows performance with the new Windows Sentinel tools from InfoWorld. ]
Figure 1 - Application Usage
Based on our sampling of ~1500 Windows systems, here's what we found:
- Over 80% of you fire-up Internet Explorer from time to time. Whether it's to check an IE-specific web site or because you're using an application that incorporates some IE functionality, Microsoft's web browser shows up as one of the most common processes in the real world.
- Despite IE's near omnipresence we still find Firefox running on nearly half (47%) of the systems sampled. This would seem to bolster our above theory about ancillary IE use: Chances are, if you're running Firefox, it's because you chose to do so and simply prefer it to IE. Within our small community, at least, Firefox is sporting "market" share that's better than half of IE's on the same sample group.
- Nearly all of you run Microsoft Office. We found no sign of OpenOffice or other alternatives. By far the major player was Microsoft Word (60%), with Excel (44%) and PowerPoint (21%) making a respectable showing. And, of course, Outlook was the dominant mail client at 50% of instrumented systems.
- Another frequent visitor: Adobe Acrobat Reader (36%). This comes as no surprise as Adobe long ago locked-up the portable document format marketplace. So much for XPS and the so-called "threat" from Microsoft.
- Also not surprising: A big swath (35%) of you own iPods (or iPhones). It's like a sea of white headphones! That's just scary!
That's all for this week. Next up: Operating System usage statistics. Stay tuned!
Posted by Randall Kennedy on April 18, 2008 09:32 AM
