Cloud servers with dedicated cores
A cloud server with dedicated cores is a cloud server with vCPUs that are tied to threads of specific physical CPU cores, rather than switching between threads of different cores.
Dedicated cores ensure that the cloud server utilizes CPU resources without competing for computing power with other cloud servers. Such isolation:
- Ensures that the cloud server has a stable performance;
- Protects the cloud server from hardware vulnerabilities (e.g. Spectre, Meltdown and others) from cloud servers on the same physical node.
Dedicated cores configuration is available when you create a cloud server or when you change the cloud server configuration.
For a cloud server with dedicated cores, Hyper-Threading (SMT) is enabled by default. You can also configure mandatory placement of resources on a single NUMA node.
Tasks to be solved
Suitable for highly loaded and performance-critical systems where predictable and stable vCPU performance is required without the impact of other cloud servers:
- for running heavily loaded databases;
- processing of analytical problems and problems with high computational complexity;
- CI/CD server operations;
- hosting game servers and real-time applications that require low latency;
- deploying microservices and APIs with high request frequency;
- ensuring stable monitoring and logging operation at high event rates.
Hyper-Threading
Hyper-Threading (SMT or hyperthreading) is a technology that allows a single physical processor core to process two threads simultaneously. In this case, each pair of cloud server vCPUs corresponds to two threads of a single physical processor core.
This technology improves performance on multi-threaded tasks such as web servers and enterprise applications, or video rendering and processing. Learn more about the technology and how it works in the Selectel blog article What is Hyper-Threading.
On cloud servers with dedicated cores, Hyper-Threading (SMT) is enabled by default. For cloud servers without dedicated cores, you cannot enable Hyper-Threading (SMT).
For databases, high-load computing (HPC), or software with physical core licensing, we recommend testing system performance with the Hyper-Threading (SMT) option enabled and disabled. You can disable the option if necessary:
- for stable performance in single-threaded scenarios;
- Isolation of tasks at the level of physical kernels;
- compliance with license requirements.
When you disable the Hyper-Threading (SMT) option, threads are disabled - each vCPU on the cloud server corresponds to an entire physical CPU core. The number of vCPUs on the cloud server is halved.
For example, you create a cloud server with Hyper-Threading (SMT) enabled and 8 vCPUs. The cloud server will use 4 physical cores, each running in Hyper-Threading (SMT) mode, that is, processing 2 threads each. So the cloud server has 8 vCPUs.
If you disable the Hyper-Threading (SMT) option, the number of vCPUs on the cloud server will be reduced to 4 and will correspond to the number of physical cores rather than the number of threads. The number of physical cores occupied will not change, so the cost of the cloud server will remain the same.
Before disabling Hyper-Threading (SMT), be sure to test application performance under load - some applications without Hyper-Threading (SMT) may run slower.
Mandatory placement of resources on a single NUMA-node
NUMA (Non-Uniform Memory Access) is an architecture for multiprocessor systems. NUMA nodes are architectural blocks in multiprocessor systems, where a node includes a single processor and local RAM that can be accessed quickly and with low latency by the processor inside the node.
When resources are placed on a single NUMA node, the cloud server works only within the node - it uses CPU cores and local memory that are bound to the node. This contributes to minimal memory latency.
You can enable mandatory cloud server resource allocation on a single NUMA node when you create a cloud server or reconfigure a cloud server with only dedicated cores and 4 vCPUs or more. You cannot enable mandatory provisioning on a single NUMA node for cloud servers without dedicated cores.
If it is not possible to place the cloud server resources on a single NUMA node, the server will not be created.
Limitations
Cloud servers with dedicated cores are available only in pool segments ru-3b, ru-7a and ru-7b.
Dedicated cores configuration is available only in Standard and HighFreq lineups. The list of fixed configuration flavors of the lines that support dedicated cores can be viewed in the table List of fixed configuration flavors in all pools.
Cost
The cost of a cloud server with dedicated cores, Hyper-Threading (SMT) option and the option of mandatory placement on one NUMA-node can be viewed at selectel.ru.
The cost of a cloud server with dedicated cores does not change:
-
when Hyper-Threading (SMT) is enabled or disabled;
-
when the option of mandatory placement on a single NUMA-node is enabled or disabled.