Hi
@Mladen, it's not the case anymore. Most recent snapshot technologies do
the so-called "redirect on write" (or "copy on first write") mechanism that
limits the necessity of additional I/O, at the expense of higher file
fragmentation. ACFS is an example easy to verify with `acfsutil info file`.
@Jeremy: beware that if you use TTS in conjunction with snapshots, you will
end up by having inter-dependencies between 1 LUN and many Databases, so
plan carefully your strategy, in particular naming conventions and
snapshots and LUNs creation and deletion.
just my 2 cents...
--
Ludovico
2016-02-23 5:10 GMT+01:00 Mladen Gogala <gogala.mladen@xxxxxxxxx>:
Jeremy, snapshot is just a set of pointers. Using snapshot will just move
the CPU usage from your DB server to your disk array, if you are using SAN
snapshot. Snapshot is NOT a copy. It's a set pointers which is kept
accurate to point in time by combining the pointers with COW. There are two
possibilities:
1. You are doing SAN snapshots and what is being snapped is a LUN. In
this case, reading the snapshot will use CPU of the SAN and, because of
optimized algorithms, will not have a profound impact on your system.
2. You are doing file system snapshots, which are done using the COW
method (copy-on-write). That is as bad as it can get. Each snapshot will
triple your IO rate. The explanation is simple: to write the new data to
your original location you must: 1) read the old data 2) write the old data
to the new location and 3) write the new data. That means that for every
write, you need to do 3 IO operations. If this is on your DB system, I wish
you a good luck.
A word of advice: test your ideas before putting them in practice. People
often confuse snapshots for copies. Let me repeat: snapshots are NOT copies.
On 2/19/2016 11:10 AM, Sheehan, Jeremy wrote:
Hello Guru’s,
Has anyone used JFS2 Snapshots with Oracle in any capacity? We have an
old process that uses a filesystem copy in combination with transportable
tablespaces to move datafiles from one db to another. We are looking to
eliminate the filesystem copy and replace it with JFS2 snapshots. This is
being done to eliminate CPU resources and speed up the process (about an
hour per tablespace grouping x 3/day x 3 tablespaces).
If you have used something in a similar manner, any gotchas to keep an eye
out for? Any advice? Please let me know.
Thanks in advance!
Jeremy
--
--
Mladen Gogala
Oracle Consultanthttp://mgogala.freehostia.com
DISCLAIMER: I am solely responsible for any opinion expressed in this email
