Register / Log-In

Exploring 7-Zip Test Result Data

So it has been just over a week since we've launched OpenBenchmarking.org, and now we are beginning to get real data from the thousands of people who have just started using OpenBenchmarking and contributing their results. As part of one of the initial blog postings, Michael posted a HOWTO that guides the user on using the compress-7zip result set. The data-set was about 170 uploads.

So, here is the data presented with score against the submission date.

Nice random data, which is what we'd expect. Note that there is an outlier with a score of about 54,000 that system seems to be a Dell PowerEdge R910 server. A quick configuration check on Dell's website indicates that the system would cost about $40,000. Unfortunately, that system is an outlier, so we'll ignore it for the rest of this article.

Looking at the histogram, we have a not-so-interesting distribution.

So now let's look to see if there is any correlation to any system component. Firstly, I imported the data into OpenOffice.org Calc and sorted the data by some of the fields that we track for the system.

First up, let's look at simply ordered by performance - to see if there is any interesting trends there.

Yes, the data is not continuous. Investigating where the major discontinuities of note occur when we go from 4 cores to 6 and 8 cores around the 8500 mark, and when the Core i7 2600 CPUs appear at the jump to 22,000. As we get more data, I am expecting to see the data smooth out considerably, with a few disruptive products thrown in the mix.

If I were to make some predictions for the shape of this graph as we get more data, I would expect that we'd see an almost logarithmic growth at the start, a gentle ramp through the middle, to a near exponential growth towards the high end.

So now I sort the data against CPU type. Bear in mind that the sorting is being done against the CPU strings that Phodevi identify.

Given that this is expected to be a CPU benchmark, the graph appears as expected. There are clearly clusters of different CPU with different strings. None of the other system metrics showed such a strong correlation with the volume of data that we have.

Is the compiler relevant to this benchmark? Let's sort the data by compiler name, and see if there is any emergent pattern?

No, the compiler doesn't seem to have an absolute bearing on the results of this test. This is further confirmation of the compress-7zip results measured in the Phoronix Sandybridge Compiler Comparison.

So compress-7zip is clearly a CPU oriented benchmark and if you compress-7zip is a representative test for your environment, focus primarily on your CPU.

---

This is a post from the OpenBenchmarking.org Blog by Matthew Tippett. Exploring 7-Zip Test Result Data was posted in Explorations on Monday, 07 March 2011.

---

User Comments

Post A Comment