Zencoder, a cloud-based video transcoding service, on Monday published benchmarks comparing Google's new infrastructure-as-a-service (IaaS) offering, Google Compute Engine (GCE), to Amazon's Elastic Compute Cloud (EC2), the current IaaS market leader. The benchmarks compare GCE, which runs entirely on the Sandy Bridge architecture, and Amazon's Cluster Compute instances, which run on Nehalem and Sandy Bridge architectures. The results show that Google has a powerful and competitive offering in the IaaS space, especially in terms of hourly price--but Amazon's instances outperform Google's, both in terms of raw performance and price-per performance.
Zencoder's benchmarks compared the following instance types:
-- Google Compute Engine's n1-standard-8-d server, running on an 8-core, 22-GCEU Sandy Bridge processor, at $1.16 per hour.
-- Amazon's EC2 cc1.4xlarge, running on an 8-core, 33.5 ECU Nehalem processor, at $1.30 per hour.
-- Amazon's EC2 cc2.8xlarge, running on a 16-core, 88 ECU Sandy Bridge processor, at $2.20 per hour.
Zencoder tested a range of different video transcoding scenarios, from one simultaneous encoding process at 640 x 360, to six simultaneous encoding processes at 1280x720. In all scenarios, Amazon's price-per performance beat Google's. For example, the EC2 cc1.4xlarge costs $2.92 per thousand, versus GCE's n1-standard-8-d's $3.66 per thousand when running six simultaneous encoding processes at 1280x720.
[ Learn more about GCE's partner ecosystem. See Google Compute Engine Leverages Third Party Support. ]
As I discussed in my hands-on review of Google Compute Engine, computing performance is only one piece of the overall picture in comparing IaaS vendors. Zencoder found, like I have, that GCE beats Amazon in some performance metrics, like boot time, and appears to have more consistent performance than EC2 has had historically. And Zencoder has compared two more costly Amazon machines to the current most-expensive Google entrant, which means that if your performance needs are met by Google's offering here, you would save money by using it over the Amazon Cluster Compute instances.
Amazon has been in the IaaS game for years, and has a much larger number of instance/machine type offerings, so it's not surprising that they can outperform Google's initial offerings. Ultimately, Google will need to add more higher-powered machine types to its IaaS product line, but these results show that Google looks like the first real competitor to Amazon in this space.