Orkhan Gasimov "High Performance System Design"
•
0 likes•105 views
Orkhan Gasimov is a digital architect who works on enterprise and solution architecture projects to transform existing businesses and develop new digital products. He provides guidance to help oversee potential issues and motivate teams. He enjoys sharing his experience through talks and conferences to support the technology community on a global scale.
Report
Share
Orkhan Gasimov "High Performance System Design"
- 2. ORKHAN GASIMOV Architecture Practice Coordinator, Digital Transformation Architect, GlobalLogic I am a passionate digital architect working at enterprise and solution architecture levels. I work with projects transforming existing businesses and those aimed to create new digital products by providing guidance and help to oversee possible issues along the journey. I love to motivate people, I love technology, and I’m proud to take part in development of technologies that supercharge the world. I also love to support community by providing talks at meetups and conferences, sharing experience at global scale.
- 4. WHAT IS PERFORMANCE? Response time Throughput Resource utilization Latency Budget Operation modes
- 5. THE PROCESS Initiate Definition and validation of initial requirements Construct Design Test in small (design and code review) Deliver Test in Large Rework Assess Maintain Initiate Construct Deliver Maintain
- 6. RISK MANAGEMENT The time, scope and resources invested in performance management must be proportional to project risks.
- 7. DESIGN DOCUMENT IF YOU DON’T DOCUMENT IT, IT DOESN’T EXIST! • Architecture and detailed design documents • Verification criteria of subsystem / component • Test plan • Test cases • Requirements 1. Use cases 2. Budgets 3. Measurement and instrumentation
- 9. TRADE-OFF Conflicts are possible, compromises are required. For instance, reliability and performance. Triple-Modular Redundancy Recovery Blocks Module 1 Module 2 Module 3 Voter Logic Block 1 Block 2 Block 3
- 11. PERFORMANCE ENGINEERING TACTICS (“Software Architecture in Practice” 2nd edition, chapter 5.4 “Performance Tactics”)
- 13. INSTRUMENTATION PRINCIPLE The most basic example – Logging Instrument your code/app to see what’s happening inside Instruments should be able to operate in different modes
- 14. SYNERGETIC PRINCIPLES Resource Sharing Principle Parallel Processing Principle Load Sharing Principle
- 15. RESOURCE SHARING PRINCIPLE System resources are limited Processes compete for resources Some resource are available for shared usage (concurrent or sequential) 1. Use resources when they are available 2. Minimize resource lock/block time and free resources asap 3. Minimize execution window for planned jobs/tasks
- 16. PARALLEL PROCESSING PRINCIPLE 1. Process in parallel when communication overhead is less than performance gain 2. Measure communication and resource concurrency overhead
- 17. LOAD SHARING PRINCIPLE Balance the workload by handling conflicting tasks at different times or in different locations. 1. Plan processes so that they do not use resources at the same time 2. Divide resources in such a way that processes use different parts of the resource, and do not require the entire resource
- 18. INDEPENDENT PRINCIPLES Centering principle Fixing-point principle Locality principle Processing versus Frequency Principle
- 19. CENTERING PRINCIPLE Dominant workload functions: A subset of system functions that are frequently used (20% of functions or less) and that are performed 80% of time or more 1. Identify the dominant workload function and minimize their processing. 2. It is necessary to design / implement the dominant functions first.
- 20. FIXING-POINT PRINCIPLE The fixing point is a point in time. The latest fixing point is during execution, just before instructions are to be executed. For responsiveness, fixing should establish data connections at the earliest feasible point in time, such that retaining the connection is cost-effective
- 21. LOCALITY PRINCIPLE Locality – closeness of desired actions, functions, and results to the physical resources used to produce them. Types of locality 1. Spatial 2. Temporal (i.e., time) 3. Effectual (i.e., purpose or intent) 4. Degree (i.e., intensity or size)
- 22. PROCESSING VS FREQUENCY PRINCIPLE The trade-off between the amount of work done in processing a request and the number of requests received.
- 24. PERFORMANCE PATTERNS Fast Path First Things First Batching Alternate Routes Flex Time Slender Cyclic Functions
- 25. FAST PATH Function A Function B Function C Frequently used (dominating) Result e.g. ATM
- 26. FIRST THINGS FIRST Secondary / Unnecessary Primary & Important e.g. Radar systems
- 27. BATCHING A B C e.g. Secure messaging
- 29. FLEX TIME e.g. Report generation
- 30. SLENDER CYCLIC FUNCTIONS e.g. Cron
- 31. PERFORMANCE ANTI-PATTERNS God class Excessive Dynamic Allocation Circuitous Treasure Hunt The One-Lane Bridge Traffic Jam
- 32. GOD CLASS
- 33. EXCESSIVE DYNAMIC ALLOCATION Object Request Object Request Object Request New Object New Object New Object Object Request Object Request Object Request Pool of Reusable Objects
- 36. TRAFFIC JAM