Lecture 4 Flashcards
What are the Cloud Resource Management Policies?
- Admission Control – prevent the system from accepting workload in violation of the high-level policies.
- Capacity allocation – allocate resources for individual activities of a service.
- Load balancing – distribute the workload evenly among the servers.
- Energy optimisation – minimise energy consumption
- Quality of service (QoS) guarantees – ability to satisfy timing or other conditions specified by Service Level Agreement
What is scheduling?
- Responsible for resource sharing + time to allocate: CPU Cycles, memory, secondary storage space, I/O + network bandwidth between users and tasks
- Scheduler: A program that implements a particular scheduling algorithm
Affects 3 criterias:
- Functionality
- Performance
- Cost
What are the policies and mechanism for resource allocation
- *Policies** – principles for guiding decisions
- *Mechanisms** – Implement policies
What are the common Scheduling algorithms?
- Round-robin
- First-Come-First-Serve
- Shortest-Job-First
What are the Scheduling Algorithm Policies?
- Best-effort policies – Impose requirements regarding enter the amount of resource allocation to an application when scheduled.
- Soft-requirements policies – require statically guaranteed amounts + timing constraints
- Hard-requirements policies – demand strict timing and precise amounts of resources
What are the Cloud Scheduling Subject to Deadlines?
- Hard deadlines – Task not completed by the deadlines, other tasks depend on it will be affected with penalties and expressed as milliseconds
- Soft deadlines – More of a guidelines with no penalties, can be missed by fractions of the units
What are the considerations required in Resource Management in Virtualised Servers?
Considerations
- Many different workloads
- Finite numbers of workloads can be hosted on each server
- Timing varying workload requirements
- Performance and resource isolation among the workloads
What is a Proportional Share (PS) and it’s evaluation?
- allocates CPU in proportion to the number of shares_,_ VM’s are assigned to
Evaluation –
- Fairness – time interval over the scheduler provides fair CPU allocation
- Allocation error – difference between the allocation and real demanded allocation
Fair-share Vs Proportional Scheduling
Fair-share schedulers attempt to provide time-averaged form of proportional shared based on the usage over time periods
Work-conserving (WC-mode)
- CPU shares are merely guarantees: there is work to be done; all clients have used shares the CPU will be used
Non work-conserving (NWC mode)
- CPU shares are caps. Clients will get their share of CPU
What is the difference between Non-Preemptive and Preemptive schedulers?
Non-Preemptive schedulers
- Schedulers allow running clients to finish their CPU slice. Schedulers make decisions once the running client gives up the CPU.
Preemptive schedulers
- Running clients can be preemptive for other clients to run. Schedulers rerun their scheduling decisions when a new client arrives
What is Workload Management?
- Workload assigns (CPU, memory, I/O) resources to applications.
- Applications provide service levels on their desired performance + workload managers assign resources to comply to these levels.
What are the Workload management approaches?
Static: Resources are estimated once + assigned statistically applications
Dynamic: Dynamically allocate resources to match applications workload resource demands
What are the Xen CPU schedulers?
- Borrowed Virtual Time (BVT)
- Simple Earliest Deadline First (SEDF)
- Credit Scheduler
What is Borrowed Virtual Time (BVT)?
- a fair-share scheduler based on the concept of virtual time, dispatching thread with runnable VM with the lowest virtual time first
- Allows latency-sensitive applications borrows virtual time to gain scheduling priority + applications borrow virtual time from its future allocation
Features:
- Preemptive, WC- code only
- Optimally-fair;
- Low overhead implementation on multiprocessors
What is Simplest Earliest Deadline First (SEDF)?
Runnable domain with the earliest deadline is picked + scheduled
Features:
- Preemptive, WC, NWC modes
- Fairness depends on the value of a period
- Implements per CPU queue, lacks global load balancing on multiprocessors