This post is more than 5 years old
28 Posts
0
7856
How does replication work?
Just wondering if someone can give me a rundown of how 'smart' replication is between Equallogic SANs.
The reason I ask, I now have two SANs that are replicating, and I've noticed the following behaviour:
1. If a single file on a replicated volume changes day-to-day, the replication that happens is using some kind of differential technology, such that the entire volume isn't replicated, merely the deltas.
2. If a single file on a volume is deleted then replaced with an updated version, no differential is used - the entire contents are re-copied in its entirety.
3. If a single file on a volume is deleted, but then replaced with an identical copy of itself, still no differential is used - the entire file is re-copied.
Scenario 2 I can kind of understand, but scenario 3 makes no sense to me. There is no real change, the data is the same, yet the entire file is re-copied during the replication schedule.
Any insight or articles I should read up on?
Thanks!
Origin3k
4 Operator
4 Operator
•
1.7K Posts
1
January 22nd, 2013 14:00
its all about blocks and not about files. EQL use a huge block size of 15MB so it might be happened the at your data is within a block. this is one of the reasons why snaps need a lot of space. if only one bit have changed a new 15MB page is allocated.
Regards,
Joerg
Times Prince
27 Posts
1
January 22nd, 2013 21:00
You have made some good observations. Both scenario 2 and 3 contain the same basic premise - you deleted a file. Once deleted the storage system would have no way to know that you rewrote the same data, because technically the original has been deleted, so there is no valid image with which to compare a new file.
Check out the EqualLogic support web site for white papers for articles.
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
1
January 23rd, 2013 09:00
Joerg,
Re: Page size. That's not 100% correct. When you do a write, a new page might be allocated. If the write falls within an existing allocated page then a new page isn't allocated. If not, then yes a new page will be allocated.
Also with replication, each page is examined using a smaller blocksize for actually changed data. So the amount of data actually replicated maybe smaller than what the allocation shows.