Efficiency Guide

User's Guide

Version 9.3

• User's Guide
• PDF
• Top

• Expand All
• Contract All

Chapters

• Introduction
  • Top of chapter
  • Purpose
  • Prerequisites
• The Seven Myths of Erlang Performance
  • Top of chapter
  • Myth: Tail-Recursive Functions are Much Faster Than Recursive Functions
  • Myth: Operator "++" is Always Bad
  • Myth: Strings are Slow
  • Myth: Repairing a Dets File is Very Slow
  • Myth: BEAM is a Stack-Based Byte-Code Virtual Machine (and Therefore Slow)
  • Myth: Use "_" to Speed Up Your Program When a Variable is Not Used
  • Myth: A NIF Always Speeds Up Your Program
• Common Caveats
  • Top of chapter
  • Timer Module
  • list_to_atom/1
  • length/1
  • setelement/3
  • size/1
  • split_binary/2
  • Operator "--"
• Constructing and Matching Binaries
  • Top of chapter
  • How Binaries are Implemented
  • Constructing Binaries
  • Matching Binaries
  • Historical Note
• List Handling
  • Top of chapter
  • Creating a List
  • List Comprehensions
  • Deep and Flat Lists
  • Recursive List Functions
• Functions
  • Top of chapter
  • Pattern Matching
  • Function Calls
  • Memory Usage in Recursion
• Tables and Databases
  • Top of chapter
  • Ets, Dets, and Mnesia
  • Ets-Specific
  • Mnesia-Specific
• Processes
  • Top of chapter
  • Creating an Erlang Process
  • Process Messages
  • SMP Emulator
• Drivers
  • Top of chapter
  • Drivers and Concurrency
  • Avoiding Copying Binaries When Calling a Driver
  • Returning Small Binaries from a Driver
  • Returning Large Binaries without Copying from a Driver
• Advanced
  • Top of chapter
  • Memory
  • System Limits
• Profiling
  • Top of chapter
  • Do Not Guess About Performance - Profile
  • Memory profiling
  • Large Systems
  • What to Look For
  • Tools
  • Benchmarking
• Retired Myths
  • Top of chapter
  • Myth: Funs are Slow
  • Myth: List Comprehensions are Slow

11 Profiling

11.1  Do Not Guess About Performance - Profile

Even experienced software developers often guess wrong about where the performance bottlenecks are in their programs. Therefore, profile your program to see where the performance bottlenecks are and concentrate on optimizing them.

Erlang/OTP contains several tools to help finding bottlenecks:

• fprof provides the most detailed information about where the program time is spent, but it significantly slows down the program it profiles.

• eprof provides time information of each function used in the program. No call graph is produced, but eprof has considerable less impact on the program it profiles.

  If the program is too large to be profiled by fprof or eprof, cprof can be used to locate code parts that are to be more thoroughly profiled using fprof or eprof.

• cprof is the most lightweight tool, but it only provides execution counts on a function basis (for all processes, not per process).

• dbg is the generic erlang tracing frontend. By using the timestamp or cpu_timestamp options it can be used to time how long function calls in a live system take.

• lcnt is used to find contention points in the Erlang Run-Time System's internal locking mechanisms. It is useful when looking for bottlenecks in interaction between process, port, ets tables and other entities that can be run in parallel.

The tools are further described in Tools.

There are also several open source tools outside of Erlang/OTP that can be used to help profiling. Some of them are:

• erlgrind can be used to visualize fprof data in kcachegrind.
• eflame is an alternative to fprof that displays the profiling output as a flamegraph.
• recon is a collection of Erlang profiling and debugging tools. This tool comes with an accompanying E-book called Erlang in Anger.

11.2  Memory profiling

eheap_alloc: Cannot allocate 1234567890 bytes of memory (of type "heap").

The above slogan is one of the more common reasons for Erlang to terminate. For unknown reasons the Erlang Run-Time System failed to allocate memory to use. When this happens a crash dump is generated that contains information about the state of the system as it ran out of mmeory. Use the crashdump_viewer to get a view of the memory is being used. Look for processes with large heaps or many messages, large ets tables, etc.

When looking at memory usage in a running system the most basic function to get information from is erlang:memory(). It returns the current memory usage of the system. instrument(3) can be used to get a more detailed breakdown of where memory is used.

Processes, ports and ets tables can then be inspecting using their respective info functions, i.e. erlang:process_info/2, erlang:port_info/2 and ets:info/1.

Sometimes the system can enter a state where the reported memory from erlang:memory(total) is very different from the memory reported by the OS. This can be because of internal fragmentation within the Erlang Run-Time System. Data about how memory is allocated can be retrieved using erlang:system_info(allocator). The data you get from that function is very raw and not very plesant to read. recon_alloc can be used to extract useful information from system_info statistics counters.

11.3  Large Systems

For a large system, it can be interesting to run profiling on a simulated and limited scenario to start with. But bottlenecks have a tendency to appear or cause problems only when many things are going on at the same time, and when many nodes are involved. Therefore, it is also desirable to run profiling in a system test plant on a real target system.

For a large system, you do not want to run the profiling tools on the whole system. Instead you want to concentrate on central processes and modules, which contribute for a big part of the execution.

There are also some tools that can be used to get a view of the whole system with more or less overhead.

• observer is a GUI tool that can connect to remote nodes and display a variety of information about the running system.
• etop is a command line tool that can connect to remote nodes and display information similar to what the UNIX tool top shows.
• msacc allows the user to get a view of what the Erlang Run-Time system is spending its time doing. Has a very low overhead, which makes it useful to run in heavily loaded systems to get some idea of where to start doing more granular profiling.

11.4  What to Look For

When analyzing the result file from the profiling activity, look for functions that are called many times and have a long "own" execution time (time excluding calls to other functions). Functions that are called a lot of times can also be interesting, as even small things can add up to quite a bit if repeated often. Also ask yourself what you can do to reduce this time. The following are appropriate types of questions to ask yourself:

• Is it possible to reduce the number of times the function is called?
• Can any test be run less often if the order of tests is changed?
• Can any redundant tests be removed?
• Does any calculated expression give the same result each time?
• Are there other ways to do this that are equivalent and more efficient?
• Can another internal data representation be used to make things more efficient?

These questions are not always trivial to answer. Some benchmarks might be needed to back up your theory and to avoid making things slower if your theory is wrong. For details, see Benchmarking.

11.5  Tools

fprof

fprof measures the execution time for each function, both own time, that is, how much time a function has used for its own execution, and accumulated time, that is, including called functions. The values are displayed per process. You also get to know how many times each function has been called.

fprof is based on trace to file to minimize runtime performance impact. Using fprof is just a matter of calling a few library functions, see the fprof manual page in Tools.

eprof

eprof is based on the Erlang trace_info BIFs. eprof shows how much time has been used by each process, and in which function calls this time has been spent. Time is shown as percentage of total time and absolute time. For more information, see the eprof manual page in Tools.

cprof

cprof is something in between fprof and cover regarding features. It counts how many times each function is called when the program is run, on a per module basis. cprof has a low performance degradation effect (compared with fprof) and does not need to recompile any modules to profile (compared with cover). For more information, see the cprof manual page in Tools.

Tool Summary

dbg

dbg is a generic Erlang trace tool. By using the timestamp or cpu_timestamp options it can be used as a precision instrument to profile how long time a function call takes for a specific process. This can be very useful when trying to understand where time is spent in a heavily loaded system as it is possible to limit the scope of what is profiled to be very small. For more information, see the dbg manual page in Runtime Tools.

lcnt

lcnt is used to profile interactions inbetween entities that run in parallel. For example if you have a process that all other processes in the system needs to interact with (maybe it has some global configuration), then lcnt can be used to figure out if the interaction with that process is a problem.

In the Erlang Run-time System entities are only run in parallel when there are multiple schedulers. Therefore lcnt will show more contention points (and thus be more useful) on systems using many schedulers on many cores.

For more information, see the lcnt manual page in Tools.

11.6  Benchmarking

The main purpose of benchmarking is to find out which implementation of a given algorithm or function is the fastest. Benchmarking is far from an exact science. Today's operating systems generally run background tasks that are difficult to turn off. Caches and multiple CPU cores does not facilitate benchmarking. It would be best to run UNIX computers in single-user mode when benchmarking, but that is inconvenient to say the least for casual testing.

Benchmarks can measure wall-clock time or CPU time.

• timer:tc/3 measures wall-clock time. The advantage with wall-clock time is that I/O, swapping, and other activities in the operating system kernel are included in the measurements. The disadvantage is that the measurements vary a lot. Usually it is best to run the benchmark several times and note the shortest time, which is to be the minimum time that is possible to achieve under the best of circumstances.
• statistics/1 with argument runtime measures CPU time spent in the Erlang virtual machine. The advantage with CPU time is that the results are more consistent from run to run. The disadvantage is that the time spent in the operating system kernel (such as swapping and I/O) is not included. Therefore, measuring CPU time is misleading if any I/O (file or socket) is involved.

It is probably a good idea to do both wall-clock measurements and CPU time measurements.

Some final advice:

• The granularity of both measurement types can be high. Therefore, ensure that each individual measurement lasts for at least several seconds.
• To make the test fair, each new test run is to run in its own, newly created Erlang process. Otherwise, if all tests run in the same process, the later tests start out with larger heap sizes and therefore probably do fewer garbage collections. Also consider restarting the Erlang emulator between each test.
• Do not assume that the fastest implementation of a given algorithm on computer architecture X is also the fastest on computer architecture Y.