每天一点点!水平有限,不正确的地方,请指出!谢谢!
1.Chapter 1.GPFS architecture(GPFS体系)
1.1 General concepts
GPFS is the successor to the PIOFS file system. It is a high performance file system designed for the needs of sophisticated parallel programs.
GPFS是一种非常成功的PIOFS文件系统。它是一种为满足需求复杂的并行程序的高性能文件系统。
Perhaps the most prominent feature of GPFS is the degree of its scalability.By scalable, we mean that incremental improvements, such as increased writing and reading performance, may be made to the file system by adding additional hardware of the same, or even lesser, capability. A truly scalable file system should provide high capacity and high throughput. GPFS file systems may be multiple Terabytes in capacity and provide throughput of multiple GB/sec to and from one file. As an AIX file system, GPFS inherently supports large files that exceed two gigabytes in size.
GPFS最出色的特征莫过于其高可拓展性。通过对文件系统增加额外的硬件或者相同的硬件来进行拓展,我们可以提高磁盘的读写性能和其灵活度。一个拥有真正高可拓展性的文件系统可以使文件系统大小更大以及提高磁盘的吞吐量。作为AIX文件系统的一种,GPFS支持超过2GB的大文件。
2013年10月22更新
1.1.1 Architecture overview
GPFS is implemented as a number of separate software subsystems, each of which may be distributed across multiple nodes within an SP system. Figure 1 on page 3 shows two nodes from a typical GPFS configuration as dark-filled boxes. The right node is an application node.That is, a node which has mounted a GPFS file system and running a user application that accesses that file system. The left node is a VSD server node.A VSD server node is one that physically has a number of disk drives attached that may be shared with other nodes in the same partition.
GPFS的实现实际是许多个独立的软件子系统功能协调发挥出来的,并且每一个单独的子系统可能是分布在多个节点上。下图从一个典型的GPFS配置中的两个节点。左节点是应用节点。也就是说,这个节点挂载GPFS文件系统同时用户运行应用程序访问这个文件系统。右节点是VSD服务节点。VSD服务节点:物理上有多个磁盘驱动连接在一起,它可能会被提供给同一分区的不同节点使用。
In contrast to the dark-filled boxes in Figure 1, which represent different nodes, the light-filled boxes represent the different software subsystems and services utilized in GPFS. The figure shows that several software subsystems utilized by GPFS must be present on the application node(s); other software subsystems may be on a node other than the application
蓝色区域(dark-filled)代表的是不同的节点,白色区域(light-filled)代表的是GPFS软件子系统和服务。从上图可以看到,GPFS子系统必须在应用节点上运行,而其他GPFS软件子系统必须在除应用节点以外的节点上运行。
这句话表明了:GPFS文件系统必须有他的GPFS服务器——VSD服务器,另外必须还有其他的服务器——跑应用的,并且接下来的table1图,会明确的告诉我们,GPFS服务器上有哪些子服务,GPFS应用服务器上又有哪些子服务。
这个图 很重要,可以和之前我们说过的GPFS简介中的哪个QUORUM 中的 N+1 冗余有很大的关系,后续,我们继续揭晓!
1.2 Data flow and potential bottlenecks(数据流和潜在的瓶颈)
One informative way to study file systems is to do an analysis of data flow for reads and writes. This is particularly true of file systems with distributed components. What follows is such an analysis plus some information on potential bottlenecks
最行之有效的研究一种文件系统的方式莫过于通过分析它在读与写过程中的数据流。特别是对于这种拥有并发式组件的文件系统。下面是这样的分析加上一些信息对潜在的瓶颈的研究。
2013年10月23
1.2.1 write data flow
Figure 2 on page 5 shows how GPFS interacts with other system components during write operations. We will consider a write of 256 KB that we assume is the size of one full GPFS block. If the write is smaller than a GPFS block, GPFS may utilize a write-behind strategy for better performance.
下图表示GPFS在写操作时如何与其他系统组件进行交互式切换,我们考虑一个写256KB的操作,假定这个恰好是一个完整的GPFS块,如果写的块大小比这个块小,那么GPFS利用延迟写策略获得更好地性能。
这个操作过程事实上就是介绍:写操作进行时,GPFS 守护进程mmfsd如何判断这个写动作可以执行,根据什么来作出判断?如果能,该怎么进行。反之,又怎样操作。
a.The application makes a write call with a pointer to a buffer in its space
应用程序调用一个写操作:一个指向缓冲区空间的
b.The mmfsd on the application node checks to see if it holds an Exclusive lock for the file. That is, it checks to see if it has the right to modify the file. If this is the first write for this node and for this file, a write token must be acquired; otherwise, (if an Exclusive token is already held) skip to step 5.
应用程序节点利用mmfsd命令去检查这个是否有排他锁,也就是说,mmfsd检查这个写操作是否拥有修改这个文件的权限,如果这是第一次写这个节点上的这个文件,那么”write token“(写令牌)必须被收回!否则,如果这个排他令牌已经被他拥有了,那么转到第 e .
2013年10月27
1.3 GPFS software structure and required services
1.4 GPFS operation
1.5 Positioning GPFS and other filesystems