stornewer

The term ‘global hot spare’ indicates that a hot spare may replace any failed disk, no matter its location in the array. To be able to replace any size of disk, it should have the same capacity as the largest disk in the array. Hot spare sizes may be mixed in an array, but the system will choose the first available hot spare (which may be larger than the failed disk), even if one of the exact size is present, but has a higher disk address.
Vault drives are not usable as hot spares.
Should a disk fail, and a hot spare be available, data will be rebuilt onto the hot spare. Even if the rebuilding is still in progress when the failed disk is replaced, it will run to completion. Once the rebuild onto the hot spare completes, data will be copied onto the replacement disk in a process known as equalization.


在做rebuilding的时候,还有一个hot spare磁盘,再坏一个磁盘,继续顶上,然后rebuilding也可以完成,我这样理解对不?

klin121628

字面意思：坏盘被替换掉后，没有完成的rebuild工作仍然会继续............

halex1227

第二段的意思是，当rebuild的时候，数据将会重建到热备盘。甚至在rebuild同时换掉损坏的磁盘，数据重建到热备盘的过程会继续一直到全部完成。当rebuild至hotspare结束后，数据将会复制到已被替换的磁盘。
这个过程的完成是由EMC CX3的UltraPoint技术来保证，即在一个扩展柜能够对单个硬盘探测和隔离，当一个磁盘损坏替换时，不会影响到同扩展柜中的其他磁盘。CX3可以在单个磁盘驱动器级别执行故障检测和隔离操作，因此可以采取纠错措施。
其实还有主动热备盘，和磁盘I/O记录等提高存储可用性的技术，也比较有特色。

stornewer

明白意思了,hotspare磁盘顶上之后,正在rebuilding,这时换掉坏盘换新盘,此时不会停止rebuilding来做新盘的rebuilding,而是继续做完rebuilding,然后再拷贝到新磁盘上

[ 本帖最后由 stornewer 于 2008-3-26 13:27 编辑 ]

halex1227

不向hotspare上进行rebuild，那么要hotspare干什么？

stornewer

halex1227谢谢你的热心帮助;!

stornewer

引用:

原帖由 stornewer 于 2008-3-26 13:21 发表
明白意思了,hotspare磁盘顶上之后,正在rebuilding,这时换掉坏盘换新盘,此时不会停止rebuilding来做新盘的rebuilding,而是继续做完rebuilding,然后再拷贝到新磁盘上

The page size selected for the array is global, applying to both SPs for reads as well as writes. The sizes listed are rounded – 2 KB is strictly 2048 bytes, but because of the extra 8 bytes per CLARiiON disk sector, a 2 KB page in cache is really 2080 bytes. A page will hold the data (or part of the data) from one I/O operation only – pages cannot be shared by different I/Os, except where sequential reads are occurring. The cache page(s) will be filled in that case.

Cache page size is one of the parameters with a large influence on cache performance.  Setting the page size too large results in unused memory; as an example, if cache page size is set to 16 KB, a host write of 49 KB requires 4 pages, or 64 KB, of cache.  If the page size was 4 KB, 13 pages would be used – a total of 52 KB. Setting the page size too large wastes memory, while setting it too small uses memory efficiently, but forces the SP to work too hard.  In applications that access user data through a file system, cache page size should typically be set to match the standard I/O size of the file system. For Windows NT and Windows 2000, this is typically 4 KB. Many Unix systems use 8 KB block sizes.
In the event that I/O sizes vary, it is recommended that the page size be left at the default, 8 KB.
上面两种情况分别page size 2KB和16KB,假设写入49KB;
Page size 2KB的情况,49KB总共占用25个pages,即50KB,实际占用的字节数为52000(2080*25)B
Page size 16KB的情况,49KB总共占用4个page,即64KB,实际占用字节数为66560(2080*64/2)B;
实际情况Page size 为2KB占用的字节数少,为什么说除了顺序读,cache page很容易被占满?除了cpu负担重一点,没看出什么问题

saintdragon

3楼正解

saintdragon

顺序读时，数据读的快，因此cache page容易写满。不知道是否这样？

stornewer

1)page size小,顺序读的时候,pages can be shared ?不是顺序读pages就不能shared?这个如何推断的?
2)随机读的时候,page size小,为什么容易满?