Cockroach Labs publishes the 2021 Cloud Report

GCP attained the highest amount of raw throughput, GCP’s machines outperformed all of AWS and Azure for network throughput. Interestingly, the three machines with AMD processors (n2d-standard-16,n2d-highcpu-16, and n2d-highmem-16) all had the lowest throughput among GCP machines.

Additionally, GCP led the way in single-core CPU performance achieving the highest coremark score by 10% over the best performing compute-optimized machines in AWS and Azure. It also maintained the most storage I/O throughput and proved to be more cost-efficient in this experiment. GCP’s top-performer (n2-standard-16) achieved 45% and 125% more read throughput than Azure and AWS, respectively. Although, it should be worth mentioning that both AWS and Azure restricted the amount of bandwidth to disks based on the machine size (16 cores) and disk type, while GCP restricted only based on the disk size. As a result, lower throughput numbers for Azure and AWS were expected. The only notable drawback the GCP had was in the area of network latency.

For three years in a row now, AWS has offered the best network latency, their top-performing machine’s 99th percentile network latency was 28% and 37% lower than Azure and GCP, respectively.

AWS machines with their custom-built Graviton2 processor performed best in the multi-core category. The processor which uses 64-bit ARM architecture enabled the top-performing m6g.4xlarge machine to achieve better multi-CPU scaling than the other machines. As regards cost, AWS provided the most cost-efficient machine (c5a.4xlarge / ebs-gp2) with a general-purpose disk, and GCP’s machine (c2-standard-16 / pd-ssd) narrowly achieved the highest maximum TPM amount with a general-purpose disk.

However, AWS demonstrated the lowest performance on seven of the 12 benchmarks, Amazon is definitely making big waves in terms of chips which was a major highlight in their most recent re:invent.

Azure’s real edge was in areas of storage surpassing AWS in storage I/O performance. Provisioned with ultra disks, their machines led all cloud providers in storage I/O read IOPS, write IOPS, and write latency. Azure machines’ IOPS performance correlated with their pricing. The machines with the ultra disks achieved 3.4 times more IOPS than the same machines with premium disks.

Its storage I/O latency performance aligned with how its machines are priced. The machines with the more expensive ultra disks had an average read latency 72% lower than that of machines with premium disks. Also, the average monthly cost of machines with ultra disks was 40% more expensive than the average monthly cost of machines with premium disks.

How each cloud provider ranked according to each benchmark is shown below

• 1 core CPU: #1 GCP #2 Azure #3 AWS
• 16 core CPU: #1 AWS #2 GCP #3 Azure
• Network Throughput : #1 GCP #2 AWS #3 Azure
• Network Latency : #1 AWS #2 Azure #3 GCP
• Storage I/O Read IOPS : #1 Azure #2 GCP #3 AWS
• Storage I/O Write IOPS : #1 Azure #2 GCP #3 AWS
• Storage I/O Read Throughput : #1 GCP #2 Azure #3 AWS
• Storage I/O Write Throughput : #1 GCP #2 Azure #3 AWS
• Storage I/O Read Latency : #1 AWS #2 GCP #3 Azure
• Storage I/O Write Latency : #1 Azure #2 GCP #3 AWS
• Max. tpm : #1 GCP #2 Azure #3 AWS
• $/tpm : #1 AWS #2 GCP #3 Azure

The researchers also added that “For the first time in our three years of putting together the Cloud Report, GCP came out on top. But declaring a winner was much harder to declare than in years past” pointing out that each of the cloud providers has shown overall growth in machine performance since their last report. In other news, Cockroach Labs, the company behind this cloud report recently secured a series E funding round of $160 million setting their valuation at $2 billion as reported by TechCrunch.