Practical distributed kV database ttserver

Relational databases often face challenges in Internet applications that store massive amounts of data:

• No quick hash search capability for key-value;
• The access interface does not support HTTP;
• The TPS for reading and writing is not high enough;

Since Google's bigtable, key-value databases have attracted the attention of more and more developers. There are more and more KV-structured databases in Internet projects. With the increasing demand for distributed technology, we hope that the database itself can support better distributed architecture when selecting a database. You can also increase or decrease machines quickly. Tokyo TC is a good dBm. With the support of todyo TT, TC can achieve good distributed storage and data search. In addition to TD, it can also implement distributed search. TT has good performance. I tested it on a common 1u server.

[Root @ WWW] #/usr/local/bin/tcrmttest write-port 11211 127.0.0.1 1000000 <writing test> host = 127.0.0.1 Port = 11211 tnum = 1 rnum = 1000000 Nr = 0 EXT = RND = 0 ......................... (00100000 )......................... (00200000 )......................... (00300000 )......................... (00400000 )......................... (00500000 )......................... (00600000 )......................... (00700000 )......................... (00800000 )......................... (00900000 )......................... (01000000) Record Number: 1000001 size: 32430400 time: 46.946 OK host load up to 1.6. The testing machine is 1 GB memory 2 gcpu.

TT is ideal for memcache. Tt also has some shortcomings. For example, the performance of one TT server cannot be supported when the data volume increases. Now I want to add a machine. The data on the TT server needs to be manually exported to another machine. If data separation is implemented in mod mode, the application layer needs to be modified. In this case, an intermediate layer is required. If the TT server has a donw server, the cache of all TT servers will be re-allocated, which will cause performance degradation. In this case, consistent hash will be better.