Category: Benchmarking

Petaflops?

on June 22, 2011

I saw an article earlier this week about Japan’s K Computer, the latest computer to be designated the “fastest supercomputer” in the world. Twice a year (June and November), the Top500 list comes out. The list’s publishers consider the highest scoring computer on the list as the fastest computer in the world. The first article I read about the recent rankings did not cite the results, just the rankings. So, I went to another article which referred to the K computer as capable of 8.2 quadrillion calculations per second, but did not give the results of the other leading supercomputers. On to the next article which said the K Computer was capable of 1.2 petaflops per second. (The phrase petaflops per second is in the same category as ATM machine or PIN number…) The same article said that the third fastest was able to get 1.75 petaflops per second. OK, now I was definitely confused. (I really miss the old days of good copy editing and fact checking, but that is a blog for another day.)

So, I went to the source, the Top500 Web site (www.top500.org). It confirmed that the K Computer obtained 8.16 petaflops (or quadrillion calculations per second) on the LINPACK test. The Chinese Tianhe-1A got 2.56 petaflops and the American Jaguar, 1.76 petaflops.

Once I got over the sloppy reporting and stopped playing with the graphs of the trends and scores over time, I started thinking about the problem of metrics and the importance of making them easy to understand. Some metrics are very easy to report and understand. For example, a battery life benchmark reports its results in hours and minutes. We all know what this means and we know that more hours and minutes is a good thing. Understanding what petaflops are is decidedly harder.

Another issue is the desire for bigger numbers to mean better results. The time to finish a task is fairly easy to understand, but in that case, less time is better. One technique for dealing with this issue is to normalize the numbers. Basically, that means to divide the result (such as a time) by the result of a baseline system’s result. The baseline system’s result is typically considered to be 1.0 (or some other number like 10 or 100) and other results are meaningful only in relation to the baseline system or each other. A system scoring 2.0 runs twice as fast as the baseline system’s 1.0. While that is clear, it does take more explanation than just seconds.

Finding the right metrics was a challenge we faced with HDXPRT 2011. Do you think we got it right? Please let us know what you think.

Bill

Comment on this post in the forums

Knowing when to wait

By Bill Catchings

on June 15, 2011

Mark mentioned in his blog entry a few weeks ago that waiting sucks. I think we can all agree with that sentiment. However, an experience I had while in Taipei for Computex made me reevaluate that thinking a bit.

I went jogging one morning in a park near my hotel. It was a relatively small park, just a quarter mile around the pond that took up most of the park. I was one of only a couple people jogging, but the park was full of people. Some were walking around the pond. There also were groups of people doing some form of Tai Chi in various clearings around the pond. The path I was on was narrow. At times, there was no way of getting around the people walking without running into the ones doing Tai Chi. That in turn meant running in place at times. Or, put another way, waiting.

Everyone was polite at the encounters, but the contrast between me jogging and the folks doing Tai Chi was stark. I wanted to run my miles as quickly as possible. Those doing Tai Chi were decidedly not in a rush. They were doing their exercises together with others. The goal was to do them at the proper pace in the proper way.

That got me to thinking about waiting on my computer. (Hey, time to think is one of the main reasons I exercise!) There are times when waiting for a computer infuriates me. Other times, however, the computer is fast enough. Or even too fast, like when I’m trying to scroll down to the right cell in Excel and it jumps down to a whole screen full of empty cells. This phenomenon, of course, relates to benchmarks. Benchmarks should measure those operations that are slow enough to hurt productivity or are downright annoying. There is less value in measuring operations that users don’t have to wait on.

Have you had any thoughts about what makes a good benchmark? Even if you weren’t exercising when you had the thought, please share it with the community.

Bill

Comment on this post in the forums

Posted in Benchmarks in general, HDXPRT metrics, What makes a good benchmark? |

Home sweet home

By Bill Catchings

on June 9, 2011

After a long set of flights back from Computex in Taipei, I’m finally home in North Carolina. Unfortunately, I’m still not quite sure what time zone I’m in!

While awake in the middle of the night, I’ve been thinking about some of the things I saw at Computex. While I was there, it seemed like a jumble of notebooks, power supplies, gaming rigs, motherboards, cases, Hello Kitty accessories, and some things that I still don’t quite know what they were. Many of the things I saw were not brand new, but it was my first chance to see them up close. Some of them were of technologies still on the horizon like Intel’s Ultrabook concept and Microsoft’s Windows 8. I also saw all sorts of combinations of phones, 4G, and other devices.

One thing that stood out to me were the number and variety of tablets. They were in a variety of sizes (and screen resolutions). There were quite a few vendors and some were ones I would not have suspected but was pleasantly surprised to encounter, like Viewsonic and Shuttle. The OS choices included Android, WebOS, and MeeGo. ASUS had a couple of interesting hybrid approaches such as the Eee Pad Transformer and the Padfone. The former is a 10.1-inch tablet that plugs into a keyboard. The Padfone is a smartphone that can plug into the back of a larger (10.1-inch) touch screen to act as a tablet.

All of these tablet choices, as well as the iPad that they all must be compete against, left me wondering how to choose between them. Some part of the choice comes down to the size and features. As always, however, performance plays a key role. My tolerance for waiting on a tablet device is even lower than it is for waiting on my PC. The problem is how to make valid comparisons across such a wide range of platforms. I’d love to hear from you what you think about performance testing on tablets. Is it useful? What are the best ways to accomplish it?

Finally, thanks to all the folks who came by and visited our suite at Computex. I enjoyed getting the chance to meet some of the members of the HDXPRT Development Community. And, hopefully, I convinced more folks to join.

Bill

Comment on this post in the forums

Posted in Future of performance evaluation, Performance testing on tablets |

Computex – Taipei

By Bill Catchings

on May 31, 2011

It’s hot and muggy here in Taipei. Just like home in North Carolina!

Weather aside, Taipei is definitely not Raleigh. Taipei is a big city with tall buildings. Right next to the hotel is the Taipei 101 which was the world’s tallest building for a few years. The streets are full of cars and motor scooters. People here walk quickly and purposefully. All of Computex seems to be filled with similar purpose and drive. It reminds me a quite bit of COMDEX in Vegas in its prime. Technology has taken over a city only too glad to embrace that technology. In next week’s blog, I’ll let you know about some of the cool things showing here.

I’ve had some interesting HDXPRT meetings so far. One of them helped me to remember some of the non-technical challenges of a successful benchmark. We’ve mentioned benchmark challenges like reliability (it needs to run when you need it to run) and repeatability (it needs to give similar results—within a few percent—each time you run it). I discussed with folks from one PC performance Web site the importance of a benchmark having some permanence. If the benchmark changes too frequently, you can’t compare the current product with the one you reviewed a couple months ago. With HDXPRT, our goal is an annual cycle. That should allow for comparing to older results while still keeping the benchmark current.

Any folks who may be here in Taipei for Computex, please come on by the Hyatt. We can talk about HDXPRT, benchmarks in general, or what you would most like to see in the future of performance evaluation. If nothing else, come by and escape the humidity! Drop us an email at hdxprt_computex@principledtechnologies.com and set up a time to come on over.

Bill

Comment on this post in the forums

Posted in Benchmarks in general, HDXPRT release cycle |

What to do, what to do

By Mark Van Name

on May 11, 2011

When you set out to build an application-based benchmark like HDXPRT, you face many choices, but two are particularly important: what applications do you run, and what functions do you perform in each application?

With HDXPRT the answers were straightforward, as they should be.

The applications we chose reflected a blend of market leaders, those providing emerging but important features, and the input from our community members.

The functions we perform in each application are ones that are representative of common uses of those programs—and that reflect the input of the community.

What’s so important here is the last clause of each of those paragraphs: your input defines this benchmark.

As we finish off HDXPRT 2011 and then move to the 2012 version, we’ll begin the development cycle anew. When we do, if you want to make sure we choose the applications and functions that matter most to you, then participate, tell us what you want, let us hear your voice. We will respond to all input, so though we can’t guarantee to accept all direction—after all, goals and desires sometimes conflict—we can guarantee that you will hear back from us and that we will explain the rationale for our decisions.

Mark Van Name

Comment on this post in the forums

Posted in Collaborative benchmark development, HDXPRT 2012, HDXPRT development process |

Putting HDXPRT in some benchmark context

By Mark Van Name

on May 6, 2011

Benchmarks come in many shapes and sizes. Some are extremely small, simple, and focused, while others are large, complex, and cover many aspects of a system. To help position HDXPRT in the world of benchmarks, let me share with you a little taxonomy that Bill and I have long used. No taxonomy is perfect, of course, but we’ve found this one to be very helpful as a general categorization tool.

From the perspective of how benchmarks measure performance, you can divide most of them into three groups.

Inspection tools use highly specialized tests to target very particular parts of a system. Back in the day, lo these many decades ago—okay, it was only two decades, but in dog years two tech decades is like five generations—some groups used a simple no-op loop to measure processor performance. I know, it sounds dumb today, but for a short time many felt it was a legitimate measure of processor clock speed, which is one aspect of performance. Similarly, if you want to know how fast a graphics subsystem could draw a particular kind of line, you could write code to draw lines of that type over and over.

These tools have very limited utility, because they don’t do what real users do, but for people working close to hardware, they can be useful.

Moving closer to the real world, synthetic benchmarks are specially written programs that simulate the kinds of work their developers believe real users are doing. So, if you think your target users are spending all day in email, you could write your own mini email client and time functions in it. These tools definitely move closer to real user work than inspection tools, but they still have the drawback of not actually running the programs real people are using.

Application-based benchmarks take that last step by using real applications, the same programs that users employ in the real world. These benchmarks cause those applications to perform the kinds of actions that real users take, and they time those actions. You can always argue about how representative they are—more on that in a future blog entry, assuming I don’t forget to write it—but they are definitely closer to the real world because they’re using real applications.

With all of that background, HDXPRT becomes easy to classify: it’s an application-based benchmark.

Mark Van Name

Comment on this post in the forums

Posted in Application-based benchmarks, Benchmarks in general, History of benchmarking, Performance benchmarking |

Category: Benchmarking

Petaflops?

Knowing when to wait

Home sweet home

Computex – Taipei

What to do, what to do

Putting HDXPRT in some benchmark context

Check out the other XPRTs: