All questions

Autonomous Performance Optimization is a new approach to IT performance optimization that leverages AI and Machine Learning techniques to automatically and continuously identify how to configure the technology stack to maximize application performance, cost savings and service resilience.

The complexity of today’s on-premise and cloud infrastructures has grown beyond the reach of even the most skilled performance experts. Software gets built faster and faster, technology platforms are complex with hundreds of possible configurations, cloud vendors offer a growing array of configuration and tuning options. Traditional manual approach to performance tuning is ineffective at this scale, which means organizations are leaving significant performance and cost savings on the table.

Akamas patented AI has been specifically designed to solve otherwise intractable real-world optimization problems that require exploring millions of system configurations. The Akamas reinforcement learning engine intelligently explores the configuration space and converges to the optimal configuration within hours. 

Akamas provides both quantifiable and unquantifiable benefits, such as improved operation knowledge and higher business agility. While results may vary for each specific application and environment, Akamas AI-powered optimization routinely generates performance improvements in the range of 20-70% – with peaks up to 2x to 4x, higher resiliency in the range of 30%-60%, infrastructure cost savings in the range of 30-80%, and increased operational efficiency (in terms of time and effort) up to 5x. With Akamas these results are achieved by setting set custom goals and constraints representing specific trade-offs among performance, resilience and costs goals and SLOs.

An Akamas optimization study consists of a number of experiments which are executed by first setting a specific configuration (multiple parameters) for the target system and measuring results when a load is applied, then scoring these results against the user-defined goal and constraints, and finally by letting Akamas AI automatically identify which most promising configuration to run next, until an optimal configuration with respect to the given goal is found.

Akamas flexible AI-driven optimizer is technology-agnostic. Akamas can optimize virtually any system, application, middleware, database, and cloud. Dozens of out-of-the-box Akamas Optimization Packs provide built-in tuning parameters, relationships, and safety guardrails for many technologies. Optimization packs can be easily created without any coding – only a simple YAML description file is required.

No, Akamas does not require any agents. Akamas can leverage any configuration interface, monitoring and load testing tool already in place. Dozens of out-the-box Akamas Telemetry Providers provide the KPIs for Akamas to score each optimization experiment against the defined optimization goal.

Akamas provides native automation operators and mechanisms, including ability to invoke API calls and SSH, to integrate with virtually any already deployed configuration management, load testing, CI/CD pipeline tool. Akamas workflows can be easily configured to smoothly integrate with your ecosystem and make the whole optimization process fully automated.

Akamas automated and continuous optimization approach is designed to fit nicely into modern CI/CD pipelines. By adding an optimization step to the pipeline before the production deployment, DevOps can ensure optimal application performance gets delivered, always.

Akamas patented AI has been designed to support real-world optimization by converging on optimal configurations faster (and more precisely) than with any alternative approach or AI technique. The actual optimization time generally correlates with the complexity of the optimization scenario, mostly represented by the number of parameters. Akamas can converge on optimal configurations in just a few hours, typically in 8 to 24 hours for medium-complexity scenarios.

Akamas optimization is guided by your custom performance and cost optimization goals. For example, you can ask Akamas to maximize application KPIs such as payments/second, minimize a database query latency, or decrease the cloud cost of your containers. You can also make sure that your SLOs or other technical and business constraints are matched. For example, you can set the goal to maximize application throughput while still matching your performance SLOs on latency and error rates. Akamas AI is able to find the optimal configuration by considering real-world constraints.

No. Akamas only focuses on application and system parameters that can deliver optimization results without requiring any code changes. While code-level optimizations may also provide significant benefits, they require (often substantial) development effort and do not represent an option that can be immediately enacted. Also, wrong configurations can be a performance drag even for the most optimized code. Akamas optimizations can be achieved in a fraction of the time and provide immediate and significant improvements. Moreover, Akamas can also optimize any application-level settings, provided that an interface (e.g. API or configuration files) is available to change them.

Akamas leverages a patented Reinforcement Learning technique specifically designed to navigate the vast configuration space of today’s tech stacks and quickly find optimal settings. Akamas does not leverage pre-defined models and best practices but autonomously experiments which configurations of your specific IT stack under your specific workloads perform well towards achieving the optimization goals you have set. This experimental approach and the ability to apply the optimal configuration, makes Akamas a true closed-loop, autonomous optimization system.

Competing optimization solutions use simple performance models, based on past resource usage, to suggest changes. Unfortunately, this approach does not work when tackling real-world optimization problems, as the effect of parameter changes on application performance is unpredictable. Akamas solves the problem by testing the configurations on real systems and measuring the resulting performance. This makes it possible to identify the best configuration for each specific application stack and workload.

Yes. Vendor default values represent recommendations based on general best practices, that of course are not specifically tuned for your specific application, workload and environment. Actually, it has been demonstrated that default values shipped by vendors (e.g. JVM garbage collection setting) may negatively impact the application end-to-end performance (e.g. transaction throughput). On the contrary, Akamas recommended configuration is based on real experiments run in your own environment by considering the full-stack, which ensures end-to-end application performance is really maximized.

AIOps tools are able to analyze incidents, identify root-causes and bottlenecks, and can recommend configuration changes for some specific technologies. Typically, these tools are reactive, as they rely on data collected after the issues have occurred and rely on best practices and customers’ troubleshooting experience. On the contrary, Akamas promotes a proactive approach that helps keeping applications and infrastructure optimized and hence prevents performance and reliability issues before applications are delivered into production.

Basics

Autonomous Performance Optimization is a new approach to IT performance optimization that leverages AI and Machine Learning techniques to automatically and continuously identify how to configure the technology stack to maximize application performance, cost savings and service resilience.

The complexity of today’s on-premise and cloud infrastructures has grown beyond the reach of even the most skilled performance experts. Software gets built faster and faster, technology platforms are complex with hundreds of possible configurations, cloud vendors offer a growing array of configuration and tuning options. Traditional manual approach to performance tuning is ineffective at this scale, which means organizations are leaving significant performance and cost savings on the table.

Akamas patented AI has been specifically designed to solve otherwise intractable real-world optimization problems that require exploring millions of system configurations. The Akamas reinforcement learning engine intelligently explores the configuration space and converges to the optimal configuration within hours. 

Akamas provides both quantifiable and unquantifiable benefits, such as improved operation knowledge and higher business agility. While results may vary for each specific application and environment, Akamas AI-powered optimization routinely generates performance improvements in the range of 20-70% – with peaks up to 2x to 4x, higher resiliency in the range of 30%-60%, infrastructure cost savings in the range of 30-80%, and increased operational efficiency (in terms of time and effort) up to 5x. With Akamas these results are achieved by setting set custom goals and constraints representing specific trade-offs among performance, resilience and costs goals and SLOs.

Core Technology

An Akamas optimization study consists of a number of experiments which are executed by first setting a specific configuration (multiple parameters) for the target system and measuring results when a load is applied, then scoring these results against the user-defined goal and constraints, and finally by letting Akamas AI automatically identify which most promising configuration to run next, until an optimal configuration with respect to the given goal is found.

Akamas flexible AI-driven optimizer is technology-agnostic. Akamas can optimize virtually any system, application, middleware, database, and cloud. Dozens of out-of-the-box Akamas Optimization Packs provide built-in tuning parameters, relationships, and safety guardrails for many technologies. Optimization packs can be easily created without any coding – only a simple YAML description file is required.

No, Akamas does not require any agents. Akamas can leverage any configuration interface, monitoring and load testing tool already in place. Dozens of out-the-box Akamas Telemetry Providers provide the KPIs for Akamas to score each optimization experiment against the defined optimization goal.

Akamas provides native automation operators and mechanisms, including ability to invoke API calls and SSH, to integrate with virtually any already deployed configuration management, load testing, CI/CD pipeline tool. Akamas workflows can be easily configured to smoothly integrate with your ecosystem and make the whole optimization process fully automated.

Akamas automated and continuous optimization approach is designed to fit nicely into modern CI/CD pipelines. By adding an optimization step to the pipeline before the production deployment, DevOps can ensure optimal application performance gets delivered, always.

Akamas patented AI has been designed to support real-world optimization by converging on optimal configurations faster (and more precisely) than with any alternative approach or AI technique. The actual optimization time generally correlates with the complexity of the optimization scenario, mostly represented by the number of parameters. Akamas can converge on optimal configurations in just a few hours, typically in 8 to 24 hours for medium-complexity scenarios.

Akamas optimization is guided by your custom performance and cost optimization goals. For example, you can ask Akamas to maximize application KPIs such as payments/second, minimize a database query latency, or decrease the cloud cost of your containers. You can also make sure that your SLOs or other technical and business constraints are matched. For example, you can set the goal to maximize application throughput while still matching your performance SLOs on latency and error rates. Akamas AI is able to find the optimal configuration by considering real-world constraints.

No. Akamas only focuses on application and system parameters that can deliver optimization results without requiring any code changes. While code-level optimizations may also provide significant benefits, they require (often substantial) development effort and do not represent an option that can be immediately enacted. Also, wrong configurations can be a performance drag even for the most optimized code. Akamas optimizations can be achieved in a fraction of the time and provide immediate and significant improvements. Moreover, Akamas can also optimize any application-level settings, provided that an interface (e.g. API or configuration files) is available to change them.

Optimization AI

Akamas leverages a patented Reinforcement Learning technique specifically designed to navigate the vast configuration space of today’s tech stacks and quickly find optimal settings. Akamas does not leverage pre-defined models and best practices but autonomously experiments which configurations of your specific IT stack under your specific workloads perform well towards achieving the optimization goals you have set. This experimental approach and the ability to apply the optimal configuration, makes Akamas a true closed-loop, autonomous optimization system.

Competing optimization solutions use simple performance models, based on past resource usage, to suggest changes. Unfortunately, this approach does not work when tackling real-world optimization problems, as the effect of parameter changes on application performance is unpredictable. Akamas solves the problem by testing the configurations on real systems and measuring the resulting performance. This makes it possible to identify the best configuration for each specific application stack and workload.

Yes. Vendor default values represent recommendations based on general best practices, that of course are not specifically tuned for your specific application, workload and environment. Actually, it has been demonstrated that default values shipped by vendors (e.g. JVM garbage collection setting) may negatively impact the application end-to-end performance (e.g. transaction throughput). On the contrary, Akamas recommended configuration is based on real experiments run in your own environment by considering the full-stack, which ensures end-to-end application performance is really maximized.

AIOps tools are able to analyze incidents, identify root-causes and bottlenecks, and can recommend configuration changes for some specific technologies. Typically, these tools are reactive, as they rely on data collected after the issues have occurred and rely on best practices and customers’ troubleshooting experience. On the contrary, Akamas promotes a proactive approach that helps keeping applications and infrastructure optimized and hence prevents performance and reliability issues before applications are delivered into production.