Tested: Google's C3D as well as C2 and N2 virtual machines to customers (r)

Sep 24, 2024
An illustration representing a bar chart and the words, 'C3D speed test.'

-sidebar-toc>        -language-notice>

Below, we will look at figures from tests conducted on real WordPress sites.

What's new with Google's C3D machines?

  1. The X4 is a fourth-generation AMD EPYC CPU (formerly codenamed "Genoa") that can be operating at 3.7 7 GHz. It supports as many as 360 virtual CPUs and more than 2,880GB in DDR5 memory.
  2. A infrastructure Processing Unit (IPU) which improves network performance as well as data I/O, while freeing the CPU from those tasks.

WordPress sites that are hosted by them can take advantage of this power since each one runs in a container isolated from other sites and includes all supporting software, like Linux, NGINX, PHP and MariaDB.

Performance comparisons of VMs

To test the performance of C3D's VM in comparison to other types that use machines, we made 3 identical WordPress (v6.5) sites that were created using a C3D machine, one on one of the C2 one, and the third on an N2 computers which are available at a couple different data center. (Although the usage of N2 computers isn't heavy at , many other hosts use them and we've added this analysis to show you how much impact it would have in the speed of your website switching to .)

It was the WordPress Hosting environment shared by the three websites:

  • WordPress version 6.5
  • Version 8.2 of PHP. 8.2
  • Ubuntu version 20.04.6
  • NGINX web server version 1.25.2
  • MariaDB Version 15.1

In order to simulate the load of our test sites, we used an Apache ab HTTP server benchmarking program, which is capable of emulating multiple concurrent users, making a variety of pages.

Demanding content that is not cached

In the absence of cached content WordPress must query the database and build the target page for each request. That's not efficient, but it's often required when the content is unique to an individual site visitor such as what's in a shopping cart.

For all three of our tests our test sites, the default shopping cart view produced a 235KB HTML web page.

So, our test protocol of the Apache ab test protocol looked like this:

  • Size of the page: 235 KB
  • Simulated concurrent users: 50
  • Run time: 60 seconds

The results (successful requests in a second):

  • C3D: 207.72
  • C2: 141.47
  • N2: 88.93

Summary Takeaway: The C3D server provided an average of 46.8 percent more cached pages than the twin site on the C2.

Chart showing results of uncached page-request testing for N2, C2, and C3D virtual machines.
Test results for speed of performance on cache-free (cache-bypass) web pages.

Requesting cached content

With caching enabled, our WordPress sites can deliver pages, without the need to run PHP workers or search the database. Indeed, thanks to the in-memory cache of NGINX, certain created content does not even need to be read off storage.

Our test content cachable was an identical blog post on each site which weighed in at 114KB. So our Apache ab testing protocol was like this:

  • Size of the page of 114 KB
  • Simulated concurrent users: 50
  • Run time: 60 seconds

The results (successful requests in a second):

  • C3D: 19,722.58
  • C2: 13,043.27
  • N2: 7,861.23

Summary Takeaway: The C3D VM thanks to its upgraded I/O stood out here when moving much more content faster than all other machines. The C3D site produced 51.2 percentage more page cached than C2's machine.

Chart showing results of cached page-request testing for N2, C2, and C3D virtual machines.
The results of the speed tests of requests for cached pages.

A prime example of raw processing power

We used image manipulation to evaluate processing on our VMs which wasn't directly connected to delivering web pages -however, resizing uploads and creating copies that have different sizes is a standard practice on a lot of WordPress websites.

We applied the ImageMagick extension in PHP to reduce the size of an 35 MB JPEG image to approximately 29KB (from 7,362 4.702 x 4,702 pixels down to 640x408 pixels) using that software's resizeImage() function and the Bessel filter.

The resulting average processing times:

  • C3D 1.484 seconds
  • C2 2.090 seconds
  • N2: 2.305 seconds

Takeaway: While resizing of the image was fairly fast on all of our test platforms, the C3D machine completed the task almost 30 percent faster over the C2 machine:

A chart showing the speeds at which C3D, C2, and N2 VMs resized a large image.
Test results for speed of image processing.

There aren't any C3Ds within the data centre? No problem!

The C3Ds are a clear benefit to operators of WordPress websites. Particularly, if you're site generates content that can't be saved, it's worth consider if those advantages make it worthwhile to house your content in one of the Data centers that support these latest machines.

Edge caching could be the ideal solution to speed up performance for those who need to locate their sites in a country that has speedier VMs are unavailable.

With edge-cachable content distributed throughout Cloudflare's worldwide data centers, response times for websites on the C3D, C2, and N2 test machines were practically the same when compared using times to the first bytes (TTFB) at various outside places.

A chart showing time-to-first-byte on C3D, C2, and N2 VMs when all are supported by edge caching.
An example of the time-to-first-byte result using Edge Caching activated.

However, even with Edge Caching, average response speed over time can be a bit faster on C3D machines since sites hosted on them can be populated and refresh caches faster.

Where do you look for C3D machines?

At present, C3D machines are available in the following 8

Google Cloud data centers:

  1. Jurong West, Singapore (asia-southeast1)
  2. Sydney, Australia (australia-southeast1)
  3. St. Ghislain, Belgium (europe-west1)
  4. Frankfurt, Germany (europe-west3)
  5. Eemshaven, Netherlands (europe-west4)
  6. Council Bluffs, Iowa, USA (us-central1)
  7. Moncks Corner, South Carolina, USA (us-east1)
  8. Ashburn, Virginia, USA (us-east4)

Within My dashboard My Dashboard, regions with the C3D machines enabled are labeled as "Boosted in the data center drop-down list when creating a new WordPress website:

A screenshot of the My dashboard showing the dialog used to select a data center.
Data centers that have C3D machines are flagged as boosted within My.

For now, Google's compute-optimized C2 VMs have been the best performers in these data centers:

  1. Changhua County, Taiwan (asia-east1)
  2. Hong Kong (asia-east2)
  3. Tokyo, Japan (asia-northeast1)
  4. Osaka, Japan (asia-northeast2)
  5. Seoul, South Korea (asia-northeast3)
  6. Mumbai, India (asia-south1)
  7. Delhi, India (asia-south2)
  8. Jakarta, Indonesia (asia-southeast2)
  9. Sydney, Australia (australia-southeast1)
  10. Melbourne, Australia (australia-southeast2)
  11. London, United Kingdom (europe-west2)
  12. Frankfurt, Germany (europe-west3)
  13. Zurich, Switzerland (europe-west6)
  14. Montreal, Canada (northamerica-northeast1)
  15. Toronto, Canada (northamerica-northeast2)
  16. Sao Paulo, Brazil (southamerica-east1)
  17. Columbus, Ohio, USA (us-east5)
  18. The Dalles, Oregon, USA (us-west1)
  19. Los Angeles, California, USA (us-west2)
  20. Salt Lake City, Utah, USA (us-west3)
  21. Las Vegas, Nevada, USA (us-west4)

Summary

Our testing suggested that moving to Google's new, faster C3D VMs could benefit the owners of many websites, especially those who have sites that provide information that isn't cacheable.

Compared to just our C2 machines -- previously the fastest used for hosting websites in Google's data centers -- the C3Ds showed:

  • An improvement in performance of almost 47% on uncached page request
  • An improvement of approximately 52% on cached page request
  • A 30percent improvement in processing time for a task such as the resizing of images

Steve Bonisteel

Steve Bonisteel is a Technical Editor for the site. He began his writing career as a newspaper reporter, who was chasing ambulances and fire trucks. He has been covering technological developments on the Internet since the 1990s.