What is GFS
Chanllege
high performance -> split data to many servers -> constant faults-> fault tolerance -> replication -> potential inconsistencies -> make protocols and do more actions to solve this -> lower performance ----------------------> conflict
架构 (to meet the challenge)
有一个master机器来维护整个系统
一致性的例子
file name map array of chunk handles and version number and log and checkpoints should store in disk , the other can store in memory to gain high performance
at leaste once ---------->secondary node may fail and client retry to ensure at least once
but the design may cause the duplicate chunk in the chunkserver storage。 we should do something, but the client may do nothing.we can provide the interface and hide the details in it
master should wait the lease is due, because the lease is not due, will two primary. make a mess.
详细阅读可以看:
《The Google File System》论文翻译(GFS-SOSP2003) - 叉鸽 MrCroxx 的博客