难道RAID 1可以增加性能吗？

cuwiner

A RAID-1 (mirror) volume maintains identical copies of the data in
RAID-0 volumes. Mirroring requires more disks. You need at least twice
as much disk space as the amount of data to be mirrored.

After configuring a mirror, you can use it as if it were a physical slice.
With multiple copies of data available, data access time is reduced if the
mirror read and write policies are properly configured.

  看书上这句话，觉得很疑惑

   在我的理解了里面，只有RAID1只是做镜像，但也有性能调优能力？

briangao

Assuming it is a simple RAID1 ( without big cache or other fancy hardware)
Read speed will be improved a bit since you can read data set from multiple disks at the same time
Write speed will be the same or slightly worse depending on implementation. You are writing the same data to multiple disks. However they are written at the same time.

The performance benefit that you will get from mirroring is minimum. So you should not use it as a performance improvement measure.

[ 本帖最后由 briangao 于 2007-3-14 09:57 编辑 ]

cuwiner

刚才看了一下资料，里面有这个一句

Mirroring improves read performance only in threaded or asynchronous I/O situations

  那什么是 in threaded ？

briangao

Threaded means one read request can be divided into multiple smaller indenpendent read operations

cuwiner

原帖由 briangao 于 2007-3-15 11:34 发表
Threaded means one read request can be divided into multiple smaller indenpendent read operations

   打破沙锅问到底，呵呵，谢谢

   那什么样的命令才可以divided into multiple smaller indenpendent read

  还有我看书上说，做RAID1能否取得读性能的提升，还要看
  Depends on the underlying device

  这个 underlying device是什么？

bsdwelt

原帖由 cuwiner 于 2007-3-15 13:40 发表



   打破沙锅问到底，呵呵，谢谢

   那什么样的命令才可以divided into multiple smaller indenpendent read

  还有我看书上说，做RAID1能否取得读性能的提升，还要看
  Depends on the underlying d ...

1。取决于产生读操作的应用程序
2。底层设备

race

RADI1 只能使性能变差，不可能增加机器性能。

nus

原帖由 race 于 2007-3-15 15:41 发表
RADI1 只能使性能变差，不可能增加机器性能。

就硬盘IO来说，相对单个硬盘，raid1可以提高一点读的性能的， 特别在同时有多个读的进程的情况下。

FreeWare

raid 1的读写性能看软件是如何操作读写的，比如软件的sun volume manager(solstice disksuite)支持三种读方式：round-robin和geometric，fist read中文大概是轮询，几何，读第一个镜像。
轮询方式从第一个子镜像开始读，然后是第二个子镜像，以此类推，如果有多个应用同时读取数据可以自然可以提高读写速度；
几何方式将所需要读取的sector范围分解到各个子镜像上，比如要读4个sectors，可能并行地从第一个和第二个子镜像同时读取各2个sectors，理论上可以获得最高2倍读取速度，实际是读取速率稍微增加；
读第一个镜像方式只读取第一个子镜像。
至于写镜像的速度，理论上略微低于单个磁盘（或raid0），因为要同步写两个盘。

至于其它软件或者硬件存储也支持各种自有方式的raid1读取模式，原则上默认方案可以提高读速率。

不知道可解释了lz的疑惑。