Flavors¶

Binero cloud provides different flavors of compute instances. A flavor defines the compute, memory, storage capacity for some flavors (see NVMe flavors below) and other hardware configuration that allows you to customize for performance or features needed by your application.

We provide different groups of flavors and each group has different performance characteristics and optimizations for different use-cases.

Each group of flavors has a version number meaning the higher version number the newer the hardware giving a performance advantage on previous versions, for flavors without an explicit version number in the name it implies version 1.

A flavor can have specific features such as NVMe storage, GPU, Pinned CPU. We add a CPU vendor identifier as a suffix to the flavor name, see the CPU vendor identifiers list below.

• a for AMD (amd64)

• i for Intel (x86_64) – Version 1 implies Intel

• r for ARM (ARM64/AARCH64)

An overview of the flavor groups and their types looks like below.

• General purpose

  • Version 1 (flavors without a version number) implies Intel, example gp.2x4

    • Pinned CPU for CPU intensive workload (dedicated CPU), example gp.4x8-pinned

  • Version 2 with AMD (CPU vendor identifier is a), example gp-v2a.2x4

• High memory

  • Version 1 (flavors without a version number) implies Intel, example hm.2x8

    • Pinned CPU for CPU intensive workload (dedicated CPU), example hm.8x32-pinned

  • Version 2 with AMD (CPU vendor identifier is a), example hm-v2a.2x8

• High performance

  • Version 1 (flavors without a version number) implies Intel, example hp.8x32

    • GPU, example hp.8x24-gpu4 where the number after GPU means 4 GB vRAM

    • NVMe storage, example hp.4x8-nvme50 where the number after NVMe means 50 GB disk

    • GPU and NVMe, example hp.12x64-gpu8-nvme250

Te flavor hm.8x48 is a High memory (hm) version 1 using Intel CPUs with 8 vCPUs and 48 GB memory.

The hp.12x64-gpu8-nvme50 flavor is High performance (hp) version 1 with 12 vCPUs, 64 GB memory, 8 GB GPU vRAM and 50 GB of local NVMe storage.

The gp-v2a.8x16 flavor is Generala purpose (gp) version 2 with 8 vCPUs and 16 GB memory.

See the list of flavors more in detail below, you can also see them in the portals or by listing them using the OpenStack terminal client with openstack flavor list --sort-column Name

General purpose¶

This general purpose flavor group has a well rounded combination of CPU and memory and is suitable for most general application use-cases that does not have heavy or intensive CPU requirements.

This flavor group provides a good cost for performance and we recommend it for use-cases with low to medium usage requirements.

Version 1 with Intel CPU¶

The version number 1 (implied in the flavor names) is using Intel as CPU vendor with Intel Xeon Gold 6138 2.0 GHz processors from the Intel Xeon Scalable Processors product family.

Pinned CPU¶

These flavors gives you gives you a pinned CPU, this means that we pin your vCPU to a dedicated CPU core (pCPU) giving you exclusive access to the computation time on that core.

In certain use-cases guaranteeing low-latency and fast access to CPU computation time is critical for time sensitive applications that suffers when scheduling of CPU time to shared CPU cores takes more time causing the application to experience lag or higher tail latencies due to for example noisy neighbours.

We also guarantee that your allocation also get the CPU core thread siblings so that you’re never exposed to any transient execution CPU vulnerabilities.

When allocating pinned CPUs we reach across all NUMA nodes to optimize CPU usage but due to memory being local to a NUMA node we recommend ultra sensitive workloads to not run its workload on different cores if you need low-latency memory access as that will traverse NUMA nodes to read memory.

Version 2 with AMD CPU¶

The version number 2 is using AMD as CPU vendor with AMD EPYC 7742 2.25 GHz processors from the EPYC 7002 series.

Performance testing has indicated that General purpose version 2 has 1-4x better performance than version 1 depending on your use-case.

High memory¶

The high memory flavor group has optimizations to provide the best value for applications that needs a larger amount of memory compared to more CPU cores that as provided by general purpose.

The flavor group targets more memory heavy use-cases such as a database application with heavy caching of the data set in memory that doesn’t have a high or intensive computation need.

Version 1 with Intel¶

The version number 1 (implied in the flavor names) is using Intel as CPU vendor with Intel Xeon Gold 6138 2.0 GHz processors from the Intel Xeon Scalable Processors product family.

Pinned CPU¶

See the pinned CPU section in General purpose version 1 for a detailed explanation of these flavors. These are the same but with a higher memory to CPU ratio.

Version 2 with AMD CPU¶

The version number 2 is using AMD as CPU vendor with AMD EPYC 7742 2.25 GHz processors from the EPYC 7002 series.

Performance testing has indicated that High memory version 2 has 1-4x better performance than version 1 depending on your use-case.

High performance¶

The high performance flavor group has optimizations for applications with either or both heavy and intensive CPU requirements. This results in faster task completion, a faster and more predictable access to both memory and storage giving you a lower latency.

The flavor group targets more CPU intensive applications causing your workload to run and complete faster improving serial throughput, use-cases such as caching servers or heavily utilized database applications or other application that needs more performance and lower latency.

The flavor sizes is more rounded to being equal in CPU and memory.

Version 1 with Intel¶

The version number 1 (implied in the flavor names) is using Intel as CPU vendor with Intel Xeon Gold 6154 3.0 GHz processors from the Intel Xeon Scalable Processors product family.

NVMe storage¶

These flavors provides local ephemeral NVMe based storage with high performance and low-latency access times, with the limitation of being local there is no data redundancy.

This flavor provides the best possible storage performance for IO intensive workloads that needs to write to disk and is great for ephemeral storage or as disk if you have an application with data replication or redundancy already built-in.

GPU¶

These flavors provides GPU based compute and allows gives you a GPU allocated to your instance with the specified vRAM / vMEM memory.

This gives you access to a NVIDIA A10 GPU that you can use for rendering, image processing, AI, ML and inference workloads that can run 50-200x faster on a GPU than a CPU.

GPU and NVMe storage¶

These flavors provides both GPU and NVMe storage and is great when you need to read or write fast to local NVMe storage for your GPU workload.