Method executed inside a Spring Rest Controller is significantly slower than in a pure java project

user_7667292 0
Viewed 143

Background: I am working on a spring web application that provides an optimization algorithm to the users. The performance of my algorithm is acceptable when executed in a pure java project. However, calling it via a rest controller in a spring project leads to a massive performance gap which I would like to understand.

Test/Observation: Of course, all tests are executed on the same hardware. The optimization algorithm is independent of spring, i.e. it receives some basic java objects as input and returns a result. As you can see in the table below, executing my algorithm takes at least twice as long inside spring which holds for 5 iterations and for 100 iterations (i.e. no constant overhead in the beginning).

Iterations [No.] Pure Java [ms] Spring [ms]
5 ~200 ~ 400
100 ~1400 3500 - 5000

I totally understand that a web application leads to some overhead on my machine (spring, angular, etc.). However, would you really expect such a massive performance impact? Essentially, this is only a small test data set and real world data sets will be more challenging.

To explain how I measured this in more detail:

  • I have an Optimization.class which contains a public static Result optimize(Data data){}. Data.class is a POJO without any reference to spring entities.
  • optimize() is an evolutionary algorithm with 5 or 100 iterations. The time given above is only measured inside optimize() but around the loop of iterations using System.currentTimeInMillis(). Hence, my method is only triggered once but executed for different number of iterations and I am only measuring how long these iterations in my algorithm take.
  • Pure Java means that I created a public static void main(String[] args) that prepares Data.class and calls optimize().
  • Spring means that I created a rest controller that prepares Data.class and calls optimize().

Main Questions: I would highly appreciate any help to understand this issue in more detail. Currently I am thinking about the following questions:

  • Which factors influence the performance of a method executed inside a spring rest controller?
  • Can I somehow configure how the given hardware is used by spring?
  • Is there any alternative to access an algorithm from the client, i.e. some alternative to the rest controller?
  • Would it help to execute my algorithm in some asynchronous way and call the server again later to receive the result?
spring-boot spring java
1 Answers

Without knowing the exact way you measured, these numbers could mean exactly zero (as in it is irrelevant what they are).

If you simply measured 5 and 100 executions, you have only measured "cold start" of those methods, and the fact that Spring is significantly slower is very expected; if you have an idea how Spring works under the hood. Spring will create proxies for all of your rest controller methods, that will go under a tomcat under the hood too... To simplify what I mean, throw an Exception in your @RestController and look at the stacktrace - you would be surprised of it's depth, I think.

The good news is that once you have many invocations of those rest methods, the JVM will optimize a lot of the things and calls will get a lot faster than the ones you made initially. Still, you will never be able to beat a plain method invocation vs calling it via a spring controller.

user_10593720
Related