MONGODB3.0 Performance Test Report II

Test environment:

Server: X86 pcserver total 6 units

CPU: Single 8-core

Memory: 64G

Disk: RAID 10

Operating system: CentOS 6.5

mongodb:3.0

Java Driver: 2.13.0

jdk:1.6

Network: Gigabit Ethernet

Test scenario:

single Monodb service, one with configuration server as a pressure server, the amount of data does not exceed the size of memory.

The background in the library is 100 million data of size 10K. The query is randomly inserted at the ID, inserting 10k of data at a time.

10 thread Insert, 10 thread query.

Before testing, a large number of random queries are performed until the memory does not grow

The following are the MONGODB 3.0 test results:

Insertion speed average 9229 per second

Query speed average 2102 per second

The following are the MONGODB 2.0 test results:

Insertion speed Average 169 per second

Query speed average 254 per second

Analysis: 1. Limited by the network, corresponding to 10k data, insert or query theory maximum speed should be in 12500, 3.0 version of the insertion efficiency should be the largest embodiment.

2. In mixed mode, the 2.4.4 version behaves poorly, and it seems that the 3.0 version of the document-level lock has a significant performance improvement.

3. Insert and query for 2.4.4 Test 10 thread case, insert in 9,000 times/second, query in 800 times/second, look at the library level of the lock on the mixed mode effect is still very large.

4.3.0 in the case of a single 10 thread, the query is 1900 times/second, and the Btree index improves query performance by one time.

5. The same amount of data, 3.0 version than the 2.4.4 version, memory almost doubled, appears to be a space for performance.

For 3.0, the insertion has reached the network limit, but the query does not, increase the query to 20 threads, that is, 10 thread insert, 20 thread query.

Insertion speed average 4622 insertion speed very unstable

Query speed average 2530 per second

It appears that a large number of queries can affect the insertion speed, which can result in very unstable insertions. Creating a replica set is a good choice for distributing queries across replicas.

Finally, the Shard performance test will be performed.

MONGODB3.0 Performance Test Report II