How Workload Prioritization Reduces Your Datacenter Footprint
Download as PPTX, PDF1 like1,344 views
The document discusses workload prioritization in data centers, focusing on balancing OLTP and OLAP workloads to minimize resource conflicts. Eliran Sinvani from Scylla introduces concepts like scheduling shares and dynamic adjustment to optimize resource utilization and enhance system performance. Future enhancements include improving visibility and isolation within workload prioritization frameworks.
How Workload Prioritization Reduces Your Datacenter Footprint
- 1. How Workload Prioritization Reduces Your Datacenter Footprint Eliran Sinvani, Core team leader
- 2. Presenter Eliran Sinvani, core team leader @ScyllaDB Eliran is a core team leader in Scylla for the past year. Before that, he had 6 years of experience developing real-time and Linux- based embedded systems. He started at Marvell as an L1 comm stack engineer. Most recently was a low-level infrastructure team leader at Airspan where he was involved in the hardware and software planning and execution of the second generation of Sprint MagicBox. Eliran has a BSc in electronics and computer engineering. In his spare time, he creates simulations and games for VR systems and tinkers with open-source embedded projects.
- 3. Agenda ■ Theory: ● some basic concepts ● Define the problem ■ Getting Technical: ● Look at the most common 2 existing solutions ● Overview of workload prioritization implementation ● Configuring Workload prioritization overview ■ Getting Practical: ● Commands example ● Some examples ● Real World example ■ Conclusion ■ Questions
- 5. Different types of loads ■ OLTP ● Small work items ● Latency sensitive ● involves narrow portion of the data ■ OLAP ● Large work items ● Throughput oriented ● Performed on large amounts of data
- 6. OK, so why can’t I simply do both? ■ We will try to explain: ● What does it mean that OLTP and OLAP conflicts. ● How can some workloads conflict and what is the impact on the Datacenter.
- 7. What happens if we just try it?
- 10. What happened?
- 11. Traditional Solutions For Conflicting Workloads
- 12. Existing Solutions ■ Divide and conquer! ● Division is space (Multi DC) ■ Wastes resources and money ● Division in time (off peak OLAP) ■ Impacts the QOS for OLTP during The OLAP periods.
- 13. Putting it in Numbers - Multi DC solution (based on AWS i3.metal) ■ Example: ● Capacity per instance: 15TB ● Minimum amount of instances: 10 ■ Assumptions: ● Real time workload is latency sensitive. Only uses 60% of resources (of 10 instances). ● Analytics don’t run constantly, therefore only runs 60% on 100% of resources. ■ HW: HW Costs Estimated Waste % Estimated Waste $ OLTP DC (10 instances) USD 278,560.00 40% (Resources) USD 167,136.00 OLAP DC (10 instances) USD 278,560.00 40% (Time) USD 167,136.00 ● Increased maintenance costs and additional complexity
- 16. How does it work?
- 17. Schedulers Basics ■ Schedulers work with Shares
- 18. Schedulers Basics 100 shares 100 shares
- 19. Schedulers Basics 100 shares 50 shares
- 20. Schedulers Basics 200 shares 100 shares
- 22. Schedulers Basics - operation highlight ■ Shares are really all there is to it :) ■ Schedulers only kicks in when there is a conflict on the resource. ■ Schedulers maintain fairness by trying to optimize ratios ● aggregate throughput is not the goal. ■ Schedulers can be dynamic ● meaning you can change the amount of shares in real time. ■ Limits the impact of one Shareholder on another.
- 24. Scylla controllers workload changes: ● automatic adjustment ● new equilibrium
- 25. Advantages ■ Better system utilization ■ Easier setup ■ Dynamic adjustment
- 26. How Does it work? ■ Schedulers ● Easy to configure ● Dynamically adjusted ● Doesn’t harm system utilization ● Limits the impact between different loads.
- 27. How Does it work? ■ Schedulers ■ Converting data processing paths from serial to parallel
- 28. How Does it work? ■ Schedulers ■ Converting data processing paths from serial to parallel ■ Operation priority classification
- 30. So…. Does it work?
- 31. So…. Does it work? (behind the scenes)
- 32. Configuring Workload prioritization 1. Make users that generates the same workload be part of the same group. ● Priorities are attached to groups or individual users. 2. Create a service level for the workload and set its shares: ● Share determine the amount of importance of the service level. ● It is always relative to other service levels. 3. Attach the service level to the group of users. ● This will grant the shares to the group of users. ● At that point the workload prioritization mechanizm will start to ● Treat their requests according to priorities.
- 33. Configuring Workload prioritization 1. Make users that generates the same workload be part of the same group. ● CREATE ROLE super_high_priority; ● GRANT super_high_priority TO special_user; 2. Create a service level for the workload and set its shares: ● CREATE SERVICE_LEVEL 'important_load' WITH SHARES=1000; 3. Attach the service level to the group of users. ● ATTACH SERVICE_LEVEL 'important_load' TO ‘super_high_priority;
- 34. Making OLTP and OLAP coexist ■ To create the effect of - OLTP always get its way and OLAP utilizes all free resources: ● OLTP gets 1000 shares and OLAP gets 10 shares.
- 35. Prioritizing between some workloads ■ Workload prioritization in general facilitates resource division between several loads. ■ There are a lot of effects that can be achieved. ■ One constraint: The number of different workloads: ● Due to latency requirements the system can only use 16 scheduling groups. ● Some scheduling groups are used by background processes ● Workload prioritization can take advantage of the remaining scheduling groups. ● Currently we have 8 unassigned scheduling groups.
- 36. Prioritizing between some workloads ■ Load1: 200 shares, Load2: 400 shares, Load3: 800 shares
- 37. Prioritizing between some workloads ■ Load1: 200 shares, Load2: 400 shares, Load3: 800 shares
- 38. Conclusion
- 39. Summary ■ The goal is to minimize inter-workload impact without breaking the bank. ■ Schedulers are in the heart of the solution: ● Shares are all there is to it.
- 40. Future work ■ Increase visibility with per scheduling group metrics. ■ Achieve even better isolation by canceling serialization points. ■ Increase the number of available workload prioritization scheduling groups.
- 41. Thank you Stay in touch Any questions? Eliran Sinvani eliransin@scylladb.com
Editor's Notes
- #2: When transferring this to power point can you make sure that the embedded videos autoplay when presenting? 10x :)