Tfs,fastdfs Contrast
The unified storage Metadata is master server, which stores the file as slave server.
1 Master Single point: Tfs,fastdfs does not exist master single point problem.
2 Slave data backup: Fastdfs uses asynchronous inter-group backups, but there is no inconsistency in reading data (solved by master scheduling, which may cause a single slave load to increase in a short time); TFS uses synchronous backups.
3 The condition of the data lookup: Fastdfs file name contains time information, can be resolved by file name.
4 Small file Management: are small files with block storage. TFS has compressed threading fragmentation and a dedicated thread-recycling block; Fastdfs all small file information in tree management (Slave server custody information), all operations must be found before the tree.
5 pic File name inconsistency problem: TFS has a custom filename interface, but must add two identity; Fastdfs is resolved by file name Mapping.
6 and PIC interface: There are related modules can refer to.
Fastdfs single-Machine performance test:
Software Environment:
CentOS 5.4 64bit
Master server:4 NET Thread
Storage server:4 net thrread,1 disk read thread,1 disk write thread
Hardware environment:
CPU: Dual Core 2.2hz
Memory: 1.5G
Write operation:
Current QPS
1 518.41
2 562.05
4 647.39
6 540.53
8 539.78
10 527.65
Storage server+ Upload Program CPU occupancy rate of about 80%, io wait 15% or so
4.1G image data, then extract URL, use siege to do stress test
Random reads:
Current QPS
10 266.67
20 425.66
50 656.93
100 807.45
200 828.73
240 833.33
280 837.34
300 801.79
Concurrent 200 or so, siege and Nginx run full Cpu,io pressure is not (Nginx should not cache