引出这个话题是来自一个需求:
目前只有两台PCServer,没有共享存储的规划, 在这种情况下,要求搭建RAC的架构,也就是存储使用两台机器上本地的盘。
在这种需求下,关联到RAC的本质,就是使用共享存储,不管是集中式的,还是分布式的。 对于共享存储可选的方案是NFS和ISCSI来实现,
从可用性排序,
ISCSI的可靠性要比NFS强一些,所以设想通过ISCSI实现(相比大家在再早之前都做过openfile系统的iscsi来搭建RAC环境),两节点互为服务端和客户端,
假设节点1有a1,a2,a3,a4,a5,a6 ,节点2 有b1,b2,b3,b4,b5,b6 一共12块盘,
通过iscsi实现两节点的操作系统上都能看到这12块盘,且映射关系是完全相同,这种方式只是50%的可用性,怎么理解呢,
主要是voting disk盘决定。
这里插播个小插曲,
介绍一下RAC的存储方案:
共享存储请求的网络存储的方式:
–iSCSI
–FibreChannel/SCSI
–NFS
一般的网络技术:
–Ethernet
–FibreChannel
网络式存储:
–Centralized pool
–Storage adminsallocate it out
–Designed for scale
效率性:
–Block-or file-based
基于块的存储
–Traditional disk devices
–Operates at a low-level
–Below the filesystem
–Basic IO primitives –read, write,“how big is the
device”
–Provides a generic way to access block storage, abstracts
underlying
–FibreChannel, SCSI, iSCSI
基于文件的存储
–Higher up the stack at an OS level
–More intelligence resides in the OS
–NFS and CIFS (CIFS not Oracle-supported!)
–Metadata lives within the protocol
Creation, access time
File sizes, owners, permissions
–Much richer set of semantics:
opendir, read, write, stat
File locking
四种一般的存储方式storage on RAC:
–Raw devices
–Clustered filesystems
–ASM
–NFS
voting
disk的磁盘组在normal冗余模式下,要求有3块盘,high冗余模式下,要求有5块盘。那我们的所搭建的架构大体如下面:
voting
disk使用的dg,由节点1的两块盘,和节点2的一块盘组成。
data使用的dg,节点1和节点2的盘组成normal
下failgroup的形式。(也就是HOS MIRROR)
为什么说是50%的高可用呢,
如果单从DATA的DG来看,两节点完全是对等的, 节点1宕机,不会影响节点2的正常使用,
节点2宕机也不会影响节点1的使用,因为是互为FG,但是但是,从voting disk来看, 节点2宕机,不会影响节点1的正常服务,
节点1宕机的话,节点2也就宕机了。因为normal 冗余下,允许损坏的盘的数量是有要求的。
External redundancy
An external redundancy disk group requires
a minimum of one disk device. The effective disk space in an
external redundancy disk group is the sum of the disk space in all
of its devices.Because Oracle ASM does not mirror data in an
external redundancy disk group,Oracle recommends that you use
external redundancy with storage devices such as RAID, or other
similar devices that provide their own data protection
mechanisms.
Normal redundancy
A normal redundancy disk group requires a
minimum of two disk devices (or two failure groups). The effective
disk space in a normal redundancy disk group is half the sum of the
disk space in all of its devices. For Oracle Clusterware files, a
normal redundancy disk group requires a minimum of three disk
devices (two of the three disks are used by failure groups and all
three disks are used by the quorum failure group) and provides
three voting files and one OCR and mirror of the OCR. When using
a normal redundancy disk group, the cluster can survive the loss of
one failure group.
High redundancy
In a high redundancy disk group, Oracle ASM
uses three-way mirroring to increase performance and provide the
highest level of reliability. A high redundancy disk group requires
a minimum of three disk devices (or three failure groups). The
effective disk space in a high redundancy disk group is one- third
the sum of the disk space in all of its devices. For Oracle
Clusterware files, a high redundancy disk group requires a minimum
of five disk devices (three of the five disks are used by failure
groups and all five disks are used by the quorum failure group) and
provides five voting files and one OCR and two mirrors of the OCR.
With high redundancy, the cluster can survive the loss of two
failure groups.
在现有资源下,设计的架构就如上所说, 但是存在性能和可用上的隐患,复杂度也提升,
所以对于生产环境,繁忙度和重要性上要求的话(iscsi通过网络传输数据,且是开源软件),
不建议这个方案,当然搭建成功可能性上没有问题的,建议使用传统共享阵列的方式。
我们能不能实现100%的可用性, 当然, 只要我们把voting disk
dg的盘使用第三台主机做共享存放就可以, 可以使用nfs,iscsci,array都可以。
那么大家看,是不是最后就变为extended distance cluster rac的架构。
extended distance cluster
rac的架构
extended distance cluster
rac的架构 通过ASM实现的mirror的DG配置。