Hazelcast put is running very slow

[dharamt...@gmail.com]

Hi Team,

Could some one help me with below confusion?  I really don't understand why Hazelcast vs Ignite put timings are varying so much. [8.33 Mins vs 15 Seconds]

Hazelcast:
I am trying to run simple test with Hazelcast 3.6 and observing put/s statistics from Management Center. Person class is using "StreamSeriazer" [Though I tried first using plain java serialization as well]

/*
Person class: [Backup count is set as 0]
public Person(String personId, String personName, int age, String address, String cellNumber, String companyWorking,
            long salary, boolean isMarried, boolean ownsTwoWheeler, boolean ownsFourWheeler, int numOfDependands)
*/


        HazelcastInstance hz = HazelcastClient.newHazelcastClient();
        IMap<Integer, Person> personMap = hz.getMap("person");
        System.out.println("** Entering into critical section ***");
       
        long time = System.currentTimeMillis();
        final int iterations = 1000000;
       
        for(int i=0;i<iterations;i++) {
            personMap.put(i, new Person(String.valueOf(i),"Person"+i,i,"Address"+i,"99309271"+i,"JPMC"+i,(2000+1)*(i+1),true,false,false,i+1));
        }
       
        long time1 = System.currentTimeMillis();
        System.out.println("** Put time : " + (time1-time) + " **");

Output:
** Entering into critical section ***
** Put time : 536168 **

From Management Center:
Total Heap  :  6060 MB
Used Heap : 624 MB
Heap Usage Percentage : 11%
Put/s statistics : Average 1700 puts per second , latency : 0.02 ms


Apache Ignite:

Ignite ignite = Ignition.start("appContext.xml");
        IgniteCache<Integer, Person> cache = ignite.getOrCreateCache("SampleCache");
       
        Thread.sleep(15000);
        System.out.println("** Entering into critical section ***");
       
        long time = System.currentTimeMillis();
        final int iterations = 1000000;
       
        for(int i=0;i<iterations;i++) {
            cache.put(i, new Person(String.valueOf(i),"Person"+i,i,"Address"+i,"99309271"+i,"JPMC"+i,(2000+1)*(i+1),true,false,false,i+1));
        }
       
        long time1 = System.currentTimeMillis();
        System.out.println("** Put time : " + (time1-time) + " **");
       
        System.out.println("Size is:" + cache.size(CachePeekMode.ALL));

Output:
** Put time : 15795 **
Size is:1000000

May 20, 2016 5:43:51 PM org.apache.ignite.logger.java.JavaLogger info
INFO:
Metrics for local node (to disable set 'metricsLogFrequency' to 0)
    ^-- Node [id=8d25690b, name=null]
    ^-- H/N/C [hosts=1, nodes=1, CPUs=12]
    ^-- CPU [cur=0.17%, avg=0.71%, GC=0%]
    ^-- Heap [used=880MB, free=85.47%, comm=6060MB]
    ^-- Public thread pool [active=0, idle=24, qSize=0]
    ^-- System thread pool [active=0, idle=24, qSize=0]
    ^-- Outbound messages queue [size=0]

Thanks,
Dharam

[Jaromir Hamala]

Hello Dharam,

If I'm reading your code correctly then I believe you are using client-server topology for Hazelcast, but just a local member for the other test. This makes it a bit unfair. Can you replace `HazelcastClient.newHazelcastClient();` with `Hazecast.newHazelcastInstance()` in your code? Also make sure you have the same number of members running. 

Please bear in mind benchmarks where just a single member is used typically don't have a massive value. A cluster of 10 members running on 10 different boxes would be a better test. 

Also both products use different defaults: I believe they use async backups and they also allow reading from backups by default. This is basically a trade-off between consistency and performance and we tend to be more conservative here. You can enable reading from backups and set async backups. 

Cheers,

Jaromir

[Dharam Thacker]

Thank you Jaromir ! Got it.