P2L5 Flashcards
Performance metrics are
measurable and quantifiable property
of the system we’re interested in
that can be used to evaluate the system behavior
Examples of performance metrics
execution time, throughput, request rate, CPU utilization, platform efficiency, wait time, performance / $, performance / W, percentage of SLA violations, avg resource usage
Event-driven model
single address space
single process
single thread of control
dispatcher == state machine
MP and MT vs Event Driven
MP and MT 1 request per exec context
Event Driven many requests interleaved n an execution context
Most efficient way to wait for input on fd
epoll more efficient than select and poll
Helpers
designated for blocking I/O operations
pipe / socket based comm w/ event dispatcher
helper blocks, but main event loop will not
AMPED
Asymmetric Multi-Process Event Driven Model
AMTED
Asymmetric Multi-Thread Event Driven Model
AMPED/AMTED
smaller footprint than regular process/thread
Flash
AMPED with asymmetric helper processes
Apache
Combo of MP / MT
Experimental Methodology
Define comparison points
Define inputs
Define metrics
When data is in cache
SPED»_space; AMPED Flash
- unnecessary test for memory presence
SPED and AMPED Flash»_space; MT/MP
- sync and context switching overhead
With disk-bound workload
AMPED Flash»_space; SPED
- SPED blocks b/c no async I/O
AMPED Flash»_space; MP/MT
- more memory efficient and less context switching
Relevant experiments
statements about a solution, that others believe in and care for