Performance Benchmark: memcached vs Kyoto tycoon

KT:

Http://fallabs.com/kyototycoon/

 

Note:

1) the client and server are both 8-core and gigabit NICs.

2) The table is horizontally the data size of value.

3) The table value is the number of operations per second.

4) KC creates a hash database.

	100b	1kb	10kb	100kb	1 MB
KT write	35599	35075	34518	33189	30562
KT read	37939	40209	38095	38197	40518
KT Deletion	39968	39541	39200	37091	37664
Memcache write	28735	29394	28977	27382	27824
Memcache read	30515	30931	30057	28968	30721
Delete memcache	32362	32278	31715	30609	32175

 

 

Note:

1) KT enables GC and memcache protocol extension. A client is used.

2) use parallel libraries and multi-thread operations during client testing.

3) values are all strings and do not involve serialized deserialization operations.

Conclusion:

Why is KT better than memcached? In addition, the disk will be released immediately after the data is removed. Unlike MongoDB and MSSQL, the disk will occupy files, unless the repair will contract.

 

However, will the data volume change when the data volume is too large? (The following test uses three machines, that is, one machine to store 1/3 of the data)

Let's take a look at the Read and Write efficiency of KT every 1 million more data records. The table value is also the number of operations per second:

	0	1	2	3	4	5	6	7	8	9
KT write	21010	11660	6888	4409	3311	2637	2387	1855	1803	1590
KT read	26063	25241	26139	25931	25810	26062	26106	26055	25465	3208

Conclusion:

There is basically no drop in reading and writing, and the total data volume is tens of millions (because it is three machines, the maximum data volume of one machine is only 5 million) when compared with empty data, the write speed is only 10%, which is inferior to that of MongoDB, And the read speed is superior to that of MongoDB.

The last read efficiency is greatly reduced. When the previously stored data is out of date, consider whether GC Recycle of expired data will affect the Read efficiency?

 

Finally, let's add one point: by default, memcache plugin does not support gets. If you want to obtain it in batches, you can directly use the GET command.