Redis/hbase/tair comparison

KV System Comparison Table TD class= "CONFLUENCETD" style= "border:1px solid #dddddd; PADDING:7PX 10px; Vertical-align:top; " > Mainstream language, currently Java, Ruby available

contrast dimension	redis	redis Cluster	medis	hbase	tair
access mode   &NBSP,	support value size		theoretically configurable (default configuration 10M, this can be larger)	256m (larger value also needs to be tested)
	support V Alue structure	byte[]/list/map/set	supports column family storage, supports multiple quantifier (quantifier supports real-time additions and deletions, and does not need to be pre-defined in the schema) under the family of columns Family TD class= "CONFLUENCETD" style= "border:1px solid #dddddd; PADDING:7PX 10px; Vertical-align:top, "colspan=" 1 "> (1) kv/map /list (2) Support Big_ List (list no length limit) (3) support to create schema , cmd query
	supported Total data volume		1000+instance	group cluster by business, single group No more than 100GB	scale out (the industry already has a single cluster of hundreds of servers in a clustered scale application case)	scale out, theoretically total data volume unrestricted
	appropriate read/write ratio	memory type , suitable for				support multi-engine, suitable for a variety of read and write ratio. Read more write less (MDB+LEVELDB), read less write more (LEVELDB).
	data is No can overwrite	y	y	support
	does IT support Hold Scan/range Query			scan not currently supported (will be supported), Range query does not support		support scan support range query
	cap	 	cp	CA (with scheme extensible to CP)	CP
	language support	mainstream language	Java only (with other language support schemes)		php,restful,java,c/c++
accessibility Can	Dot Write Latency	virtual machine on average 1~2ms		5~8ms (Write Through
	point write swallow Spit rate	A instance 5w, single machine overall performance according to the CPU to determine the			estimated single node up to 20K/S (key + value = 1k)	limited NIC bandwidth bottleneck (100MB), single machine Pure memory 8w~10w QPS (key+value=1k)
	bulk Write Throughput rate	limited NIC bandwidth bottleneck	TD class= "CONFLUENCETD" style= "border:1px solid #dddddd; PADDING:7PX 10px; Vertical-align:top, "colspan=" 1 "> Limited NIC Bandwidth bottleneck (100MB), single-Machine pure memory 8w ~10w batch/s (batch=10keys,key+value=100,batch_size=1k)
	read LA Tency	virtual machine on average 1~2ms		same room memory 1ms, disk 5-8ms (latency does not increase with the capacity of the single-machine data)
operational dimension	Scalability (automatic expansion, online expansion)		support horizontal extension	Support Online expansion (temporarily does not support data migration when expanding), Support Online data migration	Support Online smooth expansion.	Automatic expansion under read-and-write service
	Availability (whether there is a single point, data Migration/single machine error if there is service interruption, overload protection, slow query protection)	Use keepalived or official Sentinel to maintain high availability	No single point	No single point Automatic failover when data migration/standalone error occurs No overload protection (limit the number of single-client to Redis connections only 500) No slow query protection	The current cluster Hmaster is a single point of service (provides a process-level recovery mechanism). Follow-up deployment of multi-hmaster scenarios. When a single regionserver exception occurs, when the region service is migrated to another server, the data on that server is not accessible at short time. No slow query protection	< P style= "margin:0px; padding:0px; background-color:transparent; " > (1) No single point (2) uninterrupted service (3) overload protection (4) No slow query protection, But there are slow query recursive tree monitoring
	reliability (How to prevent data loss, including machine power loss, hard disk damage, etc.)	0~n data slave Backup (actual usage is basically 0 backups)	data persistence in the underlying HDFS, the default configuration is three backup.	(2) no loss of power data, replay redo log
	business Support for hybrid deployments (performance isolation)	process-level isolation (special handling can be isolated between machines)	cluster-level isolation	hard soft isolation: (1) support for different group of business mixed steps (2) support the same group business mix (through namespace isolation) to support network and memory isolation.
	access The controllable nature of the permission	The Web operations administrator authorizes api operation supports authentication and non-authentication modes	Provide safety certification for machine access,
	implementing language Words, code amount	java,java client 14000 row, central Administration 22000 row	Java. client,30k The entire project (including client and test case) 500k	core code c++,10w line around

Note: Redis, Redis cluster, Medis if there are no special instructions, the core metrics are the same.

Redis/hbase/tair comparison