The director of product management for Google announced the arrival of VM E2, a large family of virtual machines that was launched in a beta state for Google Compute Engine. E2 is a versatile family of virtual machines with "dynamic resource management" capabilities that deliver reliable performance with flexible configurations and the best total cost of ownership (TCO) of all other virtual machines on Google Cloud.
This new family of general-purpose virtual machines is suitable for various types of workloads. Google says that except for the most demanding workloads, it should be able to deliver the same performance as the "N1" machine type at a lower price. According to Google, this family of virtual machines is developed to efficiently manage its latency-sensitive and user-oriented services.
June Yang said on her blog:
“E2 virtual machines are ideal for a wide range of workloads, including web servers, mission-critical applications, small databases, and development and mid-size environments. If you have workloads that perform well on N1, but don't require large instances, local GPUs, or SSDs, consider moving them to E2. He added that "for all but the most demanding workloads, we expect the E2 to deliver similar performance to the N1, at a significantly lower cost."
Yang further reports that dynamic resource management functions ensure E2 virtual machines provide performance comparable to Google's N1 family, but compared to an average savings of 31 percent in the total cost of ownership.
The new virtual machines are also very flexible, since have 15 predefined settings. These settings go from just 2 vCPUs and 2GB of memory, up to 16 CPUs and 128GB of memory.
Therefore, virtual machines acquire long-lasting performance at a consistently low price. E2 virtual machines can handle high CPU load without complex pricing, unlike comparable options from other vendors on the market, according to Google.
E2 virtual machines they will only distribute the resources that the user needs or with the ability to use custom machine types. Custom machine types are ideal for scenarios where workloads require more processing power or more memory, but do not require all the updates provided by the next level of the machine.
In other words, vCPUs are implemented as threads scheduled to run on demand like any other thread on the host: when the vCPU has workloads, it is allocated a physical CPU available to run until 'to be freed after the task is completed.
Similarly, virtual RAM maps to physical host pages through page tables that populate when a physical guest page is accessed for the first time. This mapping remains fixed until the virtual machine indicates that the physical guest page is no longer needed.
Once virtual machines are placed on a host, their performance is continuously monitored, so if the demand for virtual machines increases, live migration can be used to seamlessly transfer E2 load to other hosts in the data center.
The workload often changes without waiting time, thanks to a predictive approach from Google. According to Google, live migration of virtual machines is a proven part of the computing engine and its performance has continually improved over time.
To achieve the performance goals of E2 virtual machines, Google has developed a custom CPU scheduler with better latency and better behavior joint programming than the default Linux programmer.
The new scheduler produces less than a microsecond average wake-up latency times with rapid context switching, helping to keep dynamic resource management overhead negligible for almost any workload.