PV 168
Seminar 13
Prerequisites for the seminar
● NetBeans IDE 12.0
○ Download from https://netbeans.apache.org/download/index.html
Address Database
● Loads list of addresses in Czech Republic (DataLoader)
● Find all addresses matching given (possibly incomplete) specification
(AddressFinder)
● Executes performance test to help evaluate CPU and Memory consumption
(PerformanceTest)
● There are multiple implementations of AddressFinder using various data
structures and search algorithms
● Concrete implementation is selected with dialog box when the application is
started
SimpleAddressFinder
● Stores data as simple List, no optimized structure
● Search is done sequentially, all addresses must be traversed
● Multiple search strategies:
○ ForEachSearchStrategy – based on for-each loop (#1)
○ StreamSearchStrategy – based on Streams (#2)
○ ParallelStreamSearchStrategy – based on parallel Streams. Parallel streams utilize
multiple threads with Fork/Join framework. Threads count corresponds to CPU count. (#3)
IndexedAddressFinder
● Stores data in map-based structure.
● The search process consist of two steps
○ Finding the collection of addresses with appropriate AddressBase (municipality, municipality
district, street, and district).
○ Then finding addresses within this collection with appropriate orientation number and/or house
number.
● The first step is implemented as a map lookup to avoid sequential search.
● Multiple implementations exist for the second step:
○ IndexedAddressGroup – addresses with the same AddressBase are stored in Maps, find
by number(s) is done as map lookup (#4)
○ SimpleAddressGroup – addresses with the same AddressBase are stored in simple List,
find by number(s) is done sequentially (#5)
Profiler
● Tools for evaluation of application performance
○ CPU time
○ Memory usage
● Profiler integrated in IntelliJ IDEA does not provide good support for memory
usage profiling 😕
Instructions (screenshots on next slides)
● Clone the project https://gitlab.fi.muni.cz/pv168/address-database
● Open the project in NetBeans
● Set Java Platform to JDK 11 or newer
○ Right click on project AddressDatabse in left panel and choose Properties
○ Select category Build > Compile and choose Java Platform
○ If you don’t see suitable JDK there, click on Manage Java Platforms… to add JDK
● Open profiler (menu Profile > Profile Project or Ctrl+F2)
● Configure Sesion by selecting profile Telemetry
● Run the application in profiler
● Choose AddressFinder implementation #1
● Check the results (notice the detail numbers when hovering over the graphs)
How to set JDK in NetBeans
Profiler window
Tips & Tricks: Snapshots
● Methods and Objects profiles allow to take snapshots of the results
○ Useful for later comparison results for different AddressFinder implementations
○ Unfortunately there is no such option for Telemetry
● Snapshot is taken if you confirm it in dialog when the application finishes
● Snapshots can be saved to disk
○ For long term comparison
● Saved snapshots can be renamed
○ In window Snapshots (menu Window > Profiling > Snapshots)
○ Better identification, e.g. AddressFinder #1
Tips & Tricks: Taking Snapshot
Tips & Tricks: Saving Snapshot
Tips & Tricks: Renaming Saved Snapshot
● Run the application with Telemetry profile
○ Check Memory graph to see used heap size after loading the data.
○ Check the output tab to see average time per single search
○ Write down both numbers
● Run the application with Methods profile
○ Check where the application spent most of the time in main thread
● Run the application with Objects profile
○ Check which object types occupied most of the heap
Seminar Task 1 (AddressFinder #1)
● Run the application with Telemetry profile
○ Check Memory graph to see used heap size after loading the data.
○ Check the output tab to see average time per single search
○ Write down both numbers
● Run the application with Methods profile
○ Check where the application spent most of the time in main thread
● Run the application with Objects profile
○ Check which object types occupied most of the heap
● Discuss and write down answers to these questions:
○ Is there any significant difference in CPU time or memory consumption compared to #1?
Seminar Task 2 (AddressFinder #2)
● Run the application with Telemetry profile
○ Check Memory graph to see used heap size after loading the data.
○ Check the output tab to see average time per single search and CPU count
○ Write down all three numbers
● Run the application with Methods profile
○ Check where the application spent most of the time in main thread
○ Check where the application spent most of the time in ForkJoinPool.* threads
● Run the application with Objects profile
○ Check which object types occupied most of the heap
● Discuss and write down answers to these questions:
○ Is there any significant difference in CPU time or memory consumption compared to #1 or #2?
○ How many ForkJoinPool.* threads were running?
Seminar Task 3 (AddressFinder #3)
● Run the application with Telemetry profile
○ Check Memory graph to see used heap size after loading the data.
○ Check the output tab to see average time per single search
○ Write down both numbers
● Run the application with Methods profile
○ Check where the application spent most of the time in main thread
● Run the application with Objects profile
○ Check which object types occupied most of the heap
● Discuss and write down answers to these questions:
○ Is there any significant difference in CPU time or memory consumption compared to #1 – #3?
Seminar Task 4 (AddressFinder #4)
● Run the application with Telemetry profile
○ Check Memory graph to see used heap size after loading the data.
○ Check the output tab to see average time per single search
○ Write down both numbers
● Run the application with Methods profile
○ Check where the application spent most of the time in main thread
● Run the application with Objects profile
○ Check which object types occupied most of the heap
● Discuss and write down answers to these questions:
○ Is there any significant difference in CPU time or memory consumption compared to #1 – #4?
Seminar Task 5 (AddressFinder #5)
● Which implementation was the least CPU efficient (slowest) one?
● Which implementation was the most CPU efficient (fastest) one?
● How much did parallel processing in #3 help?
● What is the cost of optimization in #4? Is it worth it?
● Which implementation would you recommend to use?
Seminar Task 6 (Evaluation)
Link to slides
https://is.muni.cz/auth/el/fi/podzim2020/PV168/um/seminare/PV168-seminar-13.pdf
Conclusion
Any questions?