Clear Linux Concluding 2019 With ~7% Faster Performance For The Year, Some Open-Source Workloads Much Faster
Written by Michael Larabel in Operating Systems on 17 December 2019. Page 1 of 6. 5 Comments

The latest in our series looking at various Linux performance metrics for end of year 2019 as well as for larger comparisons in ending out the 2010s, the latest is an always fun benchmarking topic... Looking at the course of Intel's Clear Linux performance over the course of the year. Here is a look at the performance of Clear Linux over the span of 2019 for 80 different tests.

Thanks to Clear Linux's swupd packaging/bundle system, while it's a "rolling release" Linux distribution, it sees multiple "releases" per day and that allows easily rolling back the state of the operating system to an earlier state. So with that in doing a year-end comparison were tests of Clear Linux 26970 as their last release of 2018 compared to Clear Linux 31890 as we approach the end of 2019.

For Clear Linux 26970 at the end of 2018 there was the Linux 4.19 kernel, X.Org Server 1.20.3, GCC 8.2.1 + LLVM Clang 7.0.1, Python 3.7.2, and the other latest versions of software packages of its time.

Clear Linux 31890 for the current state of this Linux distribution there is the Linux 5.4 kernel, X.Org Server 1.20.5, GCC 9.2.1 + LLVM Clang 9.0, Python 3.8, and the other current versions of popular software.

Some of the defaults that remained on Clear Linux throughout the year included the use of MQ-DEADLINE for NVMe solid-state storage, the Intel P-State driver with performance governor out-of-the-box, and only minor changes to the default CFLAGS/CXXFLAGS. The mitigation details and other information can be found on the system table below.

Via the Phoronix Test Suite 80 different tests were run on both Clear Linux releases. The same system was used throughout testing, which was the Intel Core i9 7980XE with 4 x 4GB DDR4-3200 Corsair memory, ASUS PRIME X299-A, eVGA NV124 graphics, and Samsung 970 EVO 500GB NVMe SSD. The i9-7980XE was used over more recent CPUs in using an Intel platform with mature support at the end of 2018.

In jumping straight to the metrics with a measurable (+/- 2%) change in performance, there was quite a few surprises. The Intel MKL-DNN / DNNL performance is much faster now than at the end of 2018 on Clear Linux, NASA's NPB tests scored big improvements on the latest toolchain, and Clear Linux's optimizations for packages like Rscript and Golang appear to be paying off. There were also many tests with single digit percentage improvements.

In only a few cases was Clear Linux 26970 faster than Clear Linux 38190. Those regressions include PHPBench, which this year Clear Linux moved to the recently released PHP 7.4. Clear Linux has already been offering substantially faster performance than other distributions thanks to making use of profile guided optimizations (PGO) and other compiler optimizations for building its PHP package. But the pullback in performance looks to be either due to a regression in PHP itself or perhaps some of Clear's tuning needing to be updated for the latest PHP. Also seeing small hits to the performance was Facebook's RockDB, albeit in a random fill scenarios, and then Mozilla's DeepSpeech speech-to-text engine running about 3% slower.

Let's take a look at some of these interesting results in more detail.


Related Articles

Trending Linux News