Use three configured for E5-2630 v4 DDR3 128 g server, set up three shard cluster each shard (3 copies), using ycsb insert 1 billion data (around 1 k per article data, using 40 threads, bulk insert), the first 100 million throughput reach more than 3 w, 300 million will see mongostat araw in is very big, dirty often reached 20%, what reason is this excuse me?
The following is my launch configuration parameters:
Pidfilepath=/data/mongo/data/shard1/log/shard1 pid
Dbpath=/data/mongo/data/shard1/data
The logpath=/data/mongo/data/shard1/log/shard1 log
Logappend=true
Bind_ip=0.0.0.0
Port=27001
The fork=true
WiredTigerCacheSizeGB=20
ReplSet=shard1
Shardsvr=true
MaxConns=20000
WiredTigerDirectoryForIndexes=true

CodePudding user response:

Normal wiredtiger btree index model is definitely more insert more slow

Constant rate of insert RocksDB engine can be used

CodePudding user response:

May be memory footprint almost full, you use htop state