Low cost servers Cloud Excess, in other words, unused capacity of Service Providers On-demand, Reserved or a low-priced server marketing method that allows you to use it until a high-priced offer arrives. Service provider that first used this marketing method It is AWS. Then other Cloud service providers as well AWS's EC2 Spot has developed similar marketing methods to this marketing method, which he called the server. The three largest cloud providers Amazon Web Services, Microsoft Azure ve Google Cloud We will compare low-cost server marketing methods for
AWS EC2 Spot Instances
AWS'It is a low-cost Virtual Server pricing method that offers discounts of up to 90% of unused capacity in a supply and demand-driven market. First of all, the method of using idle capacity with low price AWS has created. It has been used since 2009. The system, which worked with the price offer method when it first came out, has changed to use the existing idle resources on the basis of first come first serve. The bidding mechanism can still be used optionally when requested. AWS EC2 Spot servers; Elactic Compute Cloud (EC2), EC2 Auto Scaling, Big Data and Machine Learning workloads, Elastic MapReduce (EMR), Hadoop, Spark, Elastic Container Service (ECS), Elastic Kubernates Service (EKS), Self-managed Kubernetes Clusters on Spot, Data Pipeline and AWS Batch comic AWS It can be used in many services.
PROS:
AWS Spot In cases such as insufficient capacity or pricing below the bid price, the servers can be shut down by giving 2 minutes' notice. This time other Cloud longest among service providers. This period provides time for measures to be taken for disruptions in use.
Spot for servers AWS There is no time limit as well. As long as the capacity is sufficient, the servers will remain open with no time limit. Google Cloud Preemptible example, there is a maximum time limit of 24 hours for the servers. AWS Spot The fact that there is no time limit for this is a huge advantage.
EC2 Fleet To meet the target capacity you requested with (1) Spot EC2 servers and EC2 On-Demand You can create pools by using the servers at the rates you specify, and provide the opportunity to create clusters that combine system durability with price advantage. in these pools Spot There are also options such as diversifying the types of servers, choosing the appropriate one among certain strategies for the allocation of types, and determining the maximum price. EC2 Fleet is one of the strengths of the service.
Spot Instance Advisor (2) from service Spot you can get help not to be affected by the interruptions of the servers and Spot When choosing a server, you can reduce the interruption rate and at the same time provide a cost advantage. To create predictable pools Spot Instance Advisor It is a powerful service.
AWS Spot EC2 Optional in case of an outage to your servers Terminate instead of doing Stop Start (3) or Hibernate feature allows you to Pause and Resume (4).
CONS:
AWS Spot The fact that it does not commit to providing a consistent capacity for servers can be considered a weakness for this service.
EVALUATION:
AWS EC2 Spot with its servers; Stateless Web applications, Self Managed Kubernates, Elastic Container Service (ECS), Elastic Kubernates Services (EKS) Microservices and Dokerized applications developed for container platforms such as High Performance Computing (HPC) applications, CI / CD and Test development environments, Big Data ve Machine Learning (ML) It is possible to run fault-tolerant environments such as applications.
EC2 Spot Fleet mixed with (Spot, Reserved, On-Demand) with server pools Multi Availability Zone Auto Scaling By making automations resistant to the risk of failure by using the power of services such as Spot It is possible to run on presentations.
Google Cloud Preemptible VMs
preemptible Servers of Google Cloud It is a pricing method in which it markets its unused capacity as a short-lived one with the advantage of low cost. This method AWS Spot Although similar to the marketing method, there are important differences between the two services. With this service offered at a fixed price, On Demand A savings of 80% can be achieved according to the price. Google Compute Engine (GCE), Cloud GPU, Google Kubernates Engine (GKE) comic GCP in the services preemptible servers are available.
PROS:
Opportunity to benefit from discounts without uncertainty in cost with fixed price. (5)
It can be used on all server types without limitation.
Command Line (CLI) easily with the help of preemptible server activation.
Managed instance groups (MIGs) With the help of the service, you can take back your servers, which were shut down when the capacity is not enough, automatically when the capacity is sufficient.
CONS:
preemptible Servers can be rented for a maximum period of 24 hours and are permanently shut down at the end of this period. This is the service's biggest weakness when compared to other service providers. (6)
preemptible In case of insufficient capacity, the servers are shut down in a short time like 30 seconds. Although this time is sufficient to shut down the server, it may be insufficient for operations such as failover and alternative generation, thus limiting usage scenarios.
This very short uptime quota can be changed from 24 for some server types to 6 hours. preemptible It is worryingly restricting the use cases of servers.
Managed instance groups (MIGs) When you take back your servers that were shut down when the capacity is not enough, the servers remain closed. Servers also need to be turned on.
EVALUATION:
Google Preemptible servers are suitable for Fault Tolerant applications and Batches. Because they are short-lived, additional planning and monitoring is required to keep them predictable. This will create additional workloads. preemptible platforms, test environments, etc. whose servers use automation and analytics workloads. It would be appropriate to use in areas. For Production or Mission Critical applications preemptible predictable instead of servers SLAs It is more convenient to take advantage of servers that provide
Azure Low Priority VMs
low priority servers, unused Azure It is a pricing method where server resources are marketed with a low price advantage. With this method, where you can provide 80% price advantage compared to On-Demand servers, it is possible to provide price advantage in the servers you use to run your applications. This method AWS Spot Although similar to the marketing method, there are important differences between the two services. This service, which is offered with a fixed price, is only Azure Batch ve VM scale sets available with services. These servers Batch processing apps, Media Rendering jobs, Large Scale Fault Tolerant Applications, Test and Demo environments etc. can be used for
PROS:
Opportunity to benefit from discounts without uncertainty in cost with fixed price. (7)
Using with Azure Batch service On Demand ve low priority You can run servers in the same pool and provide the opportunity to create clusters that combine system durability with price advantage.
CONS:
There are problems with capacity in general. Not available at all on some server types.
The predictability of server resources is very low. It is unclear how much the source is. This makes planning difficult.
Just Azure Batch ve VM scale sets can be used with services. Single low priority server is unavailable.
Other Azure There is limited integration with services. Opportunities to integrate with other services using software tools are limited and not sufficient.
EVALUATION:
Azure Low Priority servers Bach Processing appears to be of limited use. Other Azure Since its integration with services is limited, it can be used in a limited way by making serious work and evaluation on automation and running a usable workload.
Result:
all major Cloud service providers have offered a marketing method for affordable servers to market their unused capacity with similar logic. By evaluating the service opportunities provided by the service provider, the servers with this marketing logic can also be used in live systems to reduce their bills by taking the necessary automation and operational measures. Servers Test, Demo, CI/CD, Bach Processing, Machine Learning ve Big Data operations, Media Rendering etc. Also suitable for applications.
three big Cloud The server marketing method based on the service provider's method of marketing the idle capacity at an affordable price is summarized in the table below.
Cloud Service Provider | Service name | Pricing | Time before interruption | Maximum reservation time | Capacity management | Pricing history | Services where servers can be used |
EC2 Spot Instances | Variable | 2 Minute | None (Depending on Capacity) | Spot Fleet | Spot Instance Advisor | Elactic Compute Cloud (EC2), EC2 Auto Scaling, Big Data and Machine Learning workloads, Elastic MapReduce (EMR), Hadoop, Spark, Elastic Container Service (ECS), Elastic Kubernates Service (EKS), Self-managed Kubernetes Clusters on Spot, Data Pipeline and AWS Batch, etc. | |
Low Priority VMs | Constant | 30 Seconds | None (Depending on Capacity) | No | Fixed Price | Azure Batch | |
Preemptible VMs | Constant | 30 Seconds | Maximum 24 Hours in general, Maximum 6 hours for certain types | Instance Groups | Fixed Price | Google Compute Engine VMs (GCE), Google Kubernnates Engine clusters (GKE), Managed Instance Groups, Google Cloud Dataproc etc. |
REFERENCES
TAGs: Amazon EC2 Spot Instances, Azure Low Priority VMs, Google Preemptible VMs, Low Cost Cloud Server comparison, Cloud Cost Control, Lower Your Cloud Bill.