WebAssembly and JavaScript Challenge: Numerical program performance using modern browser technologies and devices

• Table of Contents
  • Devices
  • Experiments
  • RQ1: Old versus New JavaScript Engines
  • RQ2: JavaScript versus WebAssembly
  • RQ3: Portable versus Vendor-specific Browsers
  • RQ4: Server-side Node.js versus Client-side Browsers
  • RQ5: Best Performers
  • Raw Timings
  • iPhone 10
  • Samsung S8
  • Pixel 2
  • Samsung Tab S3
  • iPad Pro
  • MacBook Pro 2013
  • Ubuntu Deer
  • Windows Bison

Devices

Experiments

RQ1 Old versus New JavaScript Engines

Conclusion: The overall performance of JavaScript against native C versions remained within a factor of 2. The current Firefox browser has presented an overall improvement, compared to the older Firefox version. The current Chrome browser, however, has presented a decrease in overall performance compared to the older Chrome version.

Figure 2. JavaScript Performance of the MacBook Pro 2013 laptop against native C, using the old and current versions for Chrome and Firefox

Figure 3. JavaScript Performance of Ubuntu Workstation against native C, using the old and current versions for Chrome and Firefox

RQ2 JavaScript versus WebAssembly

Conclusion: All browsers demonstrate significant performance improvements for WebAssembly, in the range of 2x speedups over the same browser’s JavaScript engine. Furthermore, WebAssembly achieves an overall performance close to 1 against native C, achieving an even greater performance for Firefox57 in the ubuntu-deer platform.

Figure 4. WebAssembly performance relative to C on the MacBookPro 2013.

Figure 5. WebAssembly performance relative to C on the different platforms.

Figure 6. WebAssembly performance relative to JavaScript on the different platforms.

RQ3 Portable versus Vendor-specific Browsers

Conclusion: In general the performance of the proprietary browsers is close for all the mobile and tablet devices, while presenting some differences in the windows-bison workstation and the mbp2013 laptop, where the proprietary browsers are performing worse overall. Therefore no clear trend of vendors taking advantage of their hardware architecture knowledge has been found.

Figure 7. Performance of browsers relative to proprietary respective browsers.

RQ4 Server-side Node.js versus Client-side Browsers

Conclusion: Node.js was overall slower than native C code for both JavaScript and WebAssembly, although the best WebAssembly gave reasonable performance of 0.8 the speed of native C. Server-side Node.js matched the best browser performance for both JavaScript and WebAssembly for all the devices, except for the ubuntu-deer device.

Figure 9: Performance of Node.js in different workstations relative to C. Table 6: Browser speedup performance relative to their respective WebAssembly and JavaScript Node.js versions for each device.

RQ5 Best Performers

Conclusion: The best overall browser performance was by the Firefox 57 browser, the order of device performance has the two workstations first, surprisingly followed by the iPhone 10, the rest of the order is: MacBook Pro 2013 laptop, iPad Pro, Samsung S8, Pixel 2, Samsung Tab S3 and lastly, Raspberry Pi Model B.

Table 7: Device performance across environments using the native C raspberry pi implementation as baseline for geometric means.

Raw Timings

iPhone 10

Figure 1: Timing in seconds for the Apple iPhone 10

Samsung S8

Figure 2: Timing in seconds for the Samsung S8

Pixel 2

Figure 3: Timing in seconds for the Google Pixel 2

Samsung Tab S3

Figure 4: Timing in seconds for the Samsung Tab S3

iPad Pro

Figure 5: Timing in seconds for the iPad Pro

MacBook Pro 2013

Figure 6: Timing in seconds for the MacBook Pro 2013

Ubuntu Deer

Figure 7: Timing in seconds for the ubuntu-deer

Windows Bison

Figure 8: Timing in seconds for the windows-bison