有没有生产环境只用ONLINE BACKUP的？

wangzhonnew

强，学习了

冬瓜头

快照和snapshot都是一种东西，flashcopy就是基于snapshot的，有两种方式，copy on write，或者redirect write。

这个就像oralce里的undo segment一个道理。

askgyliu

最初由 wangzhonnew 发布
[B]强，学习了 [/B]

有兴趣的话可以看看Using SANs and NAS。现在大一点的数据库都开始用非传统的方法来备份了。

http://www.itpub.net/445019,6.html

askgyliu

flashCopy is kind of storage synchronization between two sets of disks, while snapshot is just a snapshot, no more, no less.

想做FLASHCOPY的话，得要有两份IDENTICAL STORAGE。若是DB有500GB with 10 LUNs, BACKUP SERVER也要有同样的500GB with 10 LUNs，并且FS CONFIGURATION也是一样的。在做BACKUP时，会有如下PROCEDURE：
0) Attach (?) the backup storage to the db storage(I cant remember the exact word used to describe this action), i.e. establish the link between backup storage and the db storage
1) Put DB into backup mode (Oracle) or suspend write to the DB (DB2)
2) synchronize the db storage and the backup storage (flashcopy)
3) Detach the backup storage(break the link between db storage and the backup storage)
4) Pull db out of backup mode (Oracle) or resume write to DB (DB2)
5) Import the backup storage into the backup server
6) Send the backup image to tape library

而SNAPSHOT就简单一些：
1) Put the db into backup mode (Oracle. No experience on db2 on NAS snapshot)
2) Take a snapshot of the file system(s) for data files (nothing to synchronize)
3) Pull the db out of backup mode
4) Archive redo logs
5) Take a snapshot of the file system(s) for the archived log.
6) Backup server then sends the backup image to the tape library

askgyliu

FLASHCOPY和SNAPSHOT两者有个根本的区别，就是FLASHCOPY是用TWO SETS OF HARDDISKS(PHYSICALLY)，而SNAPSHOT就象是照相一样，看着有了，实际是SAME SET OF HARDDISK。

FLASHCOPY出来的BACKUP IMAGE可以直接拿到另外一个SERVER，直接就可以RESTORE AND ROLLFORWARD。

而SNAPSHOT本身并没有什么另外的STORAGE，NOT PHYSICALLY SEPARATED FROM ORIGINAL DB IMAGE。

用个例子来说明一下NAS SNAPSHOT：

Assume a file(file0) has 4 blocks, blocka0, blockb0, blockc0 and blockd0 when created, all are filled with data, and the following in sequence:

At time1: a change to blocka0, NAS will allocate a new block blocka1 to have the updated image of blocka0, while keep blocka0 unchanged.

At time2: a snapshot is taken, so there will be a file file1 consisting of blocka1/blockb0/blockc0/blockd0.

At time3: a change to blockb0, NAS will allocate a new block blockb1 to have the updated image of blockb0, while keep blockb0 unchanged.

At time4: a snapshot is taken, so there will be a file file2 consisting of blocka1/blockb1/blockc0/blockd0.

At time5: a change to blockc0, NAS will allocate a new block blockc1 to have the updated image of blockc0, while keep blockc0 unchanged.

DB BACKUP的话已经算是有了三次BACKUP了。(EDITED)

在这个时间点上，看起来FS上有三个独立的文件，file0, file1 跟 file2。而实际上这三个文件在STORAGE上却是share a common block, blockd0。如果这时候blockd0 corrupted,这三个文件都会同时坏掉。

FLASHCOPY是不会有这种问题的。

askgyliu

理论上，FLASHCOPY可以做REVERSE SYNCHRONIZATION to restore a database in seconds，而SNAPSHOT只能是用OS COPY来RESTORE A DATABASE。两者不可同日而言。

以上是我关于两者的理解。就正于方家。

askgyliu

再有，ORACLE UNDO SEGMENT是为了能有个version consistent data by SCN，所以会不停地往回滚，until the latest possible SCN。

而NAS SNAPSHOT是as it is，没有往回滚的可能。

冬瓜头

实际上，flashcopy，就是copy的snapshot，没有snapshot，就没有flashcopy。不copy，不能保证数据安全，就这样。

askgyliu

最初由 冬瓜头 发布
[B]实际上，flashcopy，就是copy的snapshot，没有snapshot，就没有flashcopy。不copy，不能保证数据安全，就这样。 [/B]

可以把你对COPY，SNAPSHOT的定义分享一下吗？我恐怕我们在谈的是不同的东西。

冬瓜头

flash copy并不是flash速度完成的，snapshot才是。
定义和你的一样，只不过表达不一样，我只是想说，flashcopy是copy的snapshot，所以没有snapshot技术，就没有flashcopy技术。flashcopy是snapshot技术的具体实现方式。其他品牌的盘阵比如hds的，用truecopy，还有比如用其他某某名词的，其实他们都是急于snapshot的技术思想来实现的。