Is that normal permormance?

[n.dobryukha]

Hi All,

I've just tried to write simplest benchmark test.

Book.java

BookMarshaller.java

book.proto

infinispan.xml

TestRealHotRodServer.java

@Test
public void testBooksCacheBenchmark() {
   final int MAX_CACHE_COUNT = 100_000;
   RemoteCache<Long, Book> cache = remoteCacheManager.getCache(BOOKS_CACHE_NAME);
  for (int i = 0; i < MAX_CACHE_COUNT; i++) {
  cache.put(UUID.randomUUID().getMostSignificantBits(), new Book("title" + i, "description" + i, 2000));
   }
  QueryFactory qf = Search.getQueryFactory(cache);
  for (int i = 0; i < 250; i++) {
  Query query = qf.from(Book.class).having("title").eq("title" + i*100).toBuilder().build();
  long startSearch = System.currentTimeMillis();
   List<Book> list = query.list();
  long finish = System.currentTimeMillis();
   LOG.info(String.format("search #%d: time = %.3fs", i, (finish - startSearch)/1000.0));
   }
  cache.clear();
}

I've got following results:
100k cached records -- ~0.087s per search

250k cached records -- ~0.210s per search

500k cached records -- ~0.420s per search

Is that normal performance?

I use Infinispan Server 8.2.1. Infinispan instance is on the same machine with test application.

Intel Core I5-4570 @ 3.20 GHz, 16gb RAM

[gustavonalle]

Could you re-run the benchmark annotating your .proto so that its fields are indexed? An example is  [1]

Also,  make sure to follow some guidelines [2] when measuring java code to avoid common pitfalls.

[1] https://github.com/infinispan/protostream/blob/master/sample-domain-definition/src/main/resources/sample_bank_account/bank.proto

[2] http://stackoverflow.com/questions/504103/how-do-i-write-a-correct-micro-benchmark-in-java/513259#513259

[n.dobryukha]

Hi Gustavo Fernandes,

I've changed book.proto with your recommendation. Is it correct now?
Also I've used a JMH for create benchmark - MyBenchmark.java

Results are in attached files.

Could you give any comments?

[n.dobryukha]

Hey guys,

I need a piece of advice about improving of querying performance.

[gustavonalle]

Due to some configuration issues (ISPN-6577), your cache is running in non-indexed mode, that explains why the query performance is poor.

Try changing from

<replicated-cache name="booksCache" configuration="indexed-cache">

to

<replicated-cache name="booksCache" mode="SYNC" start="EAGER" remote-timeout="20000">

I run the benchmark on my environment and the perf with 500_000 went from

0.346 ±(99.9%) 0.015 s/op

to

0.003 ±(99.9%) 0.001 s/op

so about 2 orders faster.

[gustavonalle]

Btw, when the cache is indexed you will notice it takes a while to put half a million entries. To speed-up the process you have 2 choices:

• Populate the cache with multiple threads instead of 1 as the benchmark is doing;  or
• Enable async indexing, performance should be roughly the same as non-indexed caches at the cost of small delay to have the indexing done after the puts are finished (1 second or so).  So you need to check when index is done, like:      

int resultSize = 0;
while(resultSize < MAX_CACHE_COUNT) {
  resultSize = Search.getQueryFactory(cache).from(Book.class).build().getResultSize();
}

To enable async use the following property:

<property name="default.worker.execution">async</property>