Redis stand-alone transaction test under WINDOWS

WIN 10:i5 4440, 8GB

Test the transaction speed of Redis

The Redis watch mechanism implementation transaction is an optimistic locking form (which should be a spin lock). For scenarios where concurrency is not high, you can conserve CPU resources (lightweight locks). However, when the competition is intense, the probability of failure is greatly increased, and the retry after a large number of failures consumes CPU resources.

The logic of the test

1) Read counter

2) Java-side self-increment

3) set back to Redis

The Redis incr function is not available here, the Redis server side is single-threaded execution, so the server side only executes a incr instruction, in fact, can be regarded as atomic operation, the server side directly executes 1-2 seconds, of course, depending on the computer.

The counter is calculated until 100,000. (The number of threads increases, so the load calculation per thread is reduced)

Package one;import java.text.numberformat;import java.util.concurrent.countdownlatch;import  redis.clients.jedis.Jedis;import redis.clients.jedis.Transaction;public class  transactioncontentiontest extends connectionbuilder {    private  Static final int contention_level = 5;    private static  final int TOTAL = 20000;    private static final  Numberformat nf = numberformat.getpercentinstance ();     public static  void main (String[] args)  throws Exception {         nf.setminimumfractiondigits (2);         buildpool ();         countdownlatch latch = new countdownlatch ( Contention_level);        Thread[] threads = new Thread[CONTENTION_LEVEL];         contentionclient[] clients = new contentionclient[contention _level];        jedis jedis = pool.getresource ();         jedis.set ("Testcounter",  "0");         jedis.close ();        for  (int i =  0; i < contention_level; i++)  {             contentionclient client = new contentionclient ();             client.settotal (total);             client.setcountername ("TestCounter");        &nbsP;    client.setjedis (Pool.getresource ());             client.setlatch (latch);             clients[i] = client;             threads[i] = new thread (client);        }         long start = system.currenttimemillis ();         for  (Int i = 0; i < contention_level ;  i++)  {            threads[i].start ();         }        latch.await () ;         long end = system.currenttimemillis ();     &nbSp;   system.out.println ("Elapse:"  +  (End - start)  +  " ms");         for  (int i = 0; i <  contention_level; i++)  {             double failrate =  (Double)  clients[i].getfailcount ()  / TOTAL;             system.out.println (i +  " Fail Rate:"  + nf.format (failrate));             Clients[i].getjedis (). Close ();        }         close ();    }    static class  Contentionclient implements runnable {        private  jedis jedis;        private string countername;         private int total;        private long  failCount = 0;        public CountDownLatch  Getlatch ()  {            return latch;         }        public  Void setlatch (Countdownlatch latch)  {             this.latch = latch;        }         private CountDownLatch latch;         public jedis getjedis ()  {             return jeDis;        }        public  void setjedis (Jedis jedis)  {             this.jedis = jedis;        }         public string getcountername ()  {             return counterName;         }        public void setcountername (String  CounterName)  {            this.countername  = counterName;        }         public int gettotal ()  {             return total;        }        public  Void settotal (int total)  {             this.total = total;        }         public long getfailcount ()  {             return failCount;        }         public void setfailcount (Long failCount)  {             this.failCount = failCount;         }         @Override          public void run ()  {             while  (total > 0)  {                 jedis.watch (CounterName);                 integer counter = integer.parseint ( Jedis.get (CounterName));                 transaction tx = jedis.multi ();                 counter++;                 tx.set (Countername, counter.tostring ());                 if  (Tx.exec ()  ==  null)  {                     jedis. Unwatch ();                     failCount++;                 } else {                     total--;                 }             }            latch.countdown ();         }    }}

Test multiple sets of data separately

Number of threads	1	2	4	8	16
Retry rate	0	67%	185%	435%	900%
Take	21055ms	19312	21136	34231	65964

With optimistic locking mechanism concurrency is slightly higher, the resource consumption is very serious.

After all, the use of Windows and home machines, testing data, no reference value.

This article from "Tech Doc" blog, declined reprint!

Redis stand-alone transaction test under WINDOWS