HPE EVA Storage

Theoretical question regarding BC & CA

 
SOLVED
Go to solution
eric lipede_1
Regular Advisor

Theoretical question regarding BC & CA

Hi

Scenario
=-=-=-=-

Site A ====== 1 Host
1 SAN
BC running locally
CA running replicating
data to Site B

Site B
======
1 Host
1 SAN
BC running locally
CA receiving


My questions are:

1. when a write is performed on a disc in site A. Is a SIMULTANEOUS write also made to the BC
copy and the CA Journal (async/ Journal used)?

2. On site A: We have PVOL and SVOL. Depending
on the HORCC_MRCF variable, the SVOL refers
to either the local BC or remote SVOL.
So if the remote SVOL is also performing a remote BC (Where it is the remote PVOL), how does the remote Site B, manage the fact that the its PVOL is also an SVOL ? - or is this irrelevant as the 2 states are mutually exclusive when HORCC_MRCF is set or unset ?.

3. In practical terms, does BC really impact on performance ? We can set thresholds etc but at its worst, how bad can it get ?. Does anyone have experience of performance issues and what they did to resolve ?

4. if CA is stopped, is there a recommendation as to where pairsync should be run from. ie does it matter if the sync command is run from site A or site B ?. What would happen if CA is started (unknowingly) from site A. & B at the same time.

5. In the scenario above (ie CA is stopped) BUT from site B. Is the site B BC affected in any way?
Does site A's BC become faster ?

All answers will have points assigned...at least those that dont respond with "see xxx manual". Many thks in advance.
10 REPLIES 10
Steven Clementi
Honored Contributor

Re: Theoretical question regarding BC & CA

1. The only way you can have simultaneous writes would be if your host is mirroring between the 2 arrays. CA operates...
Sync: Incoming Write --> Outgoing Replicated Write Request --> Remote Write --> Remote Acknowledgement --> Local Acknowledgment --> i/o complete
Async: Write Request --> Local Acknowledgment --> Outgoing Replicated Write Request --> Remote Write --> Remote Acknowledgement --> i/o complete (approximate, might be a little off)

BC on the other hand.. is a point in time copy of your local data.

That being said... if you BC(snapshot/clone) a vdisk... new writes to that vdisk are not set to the snapshot/clone.


2. Not sure...

3. For thatpoint in time that you take a snapshot or snapclone of a vdisk... you might experience some performance impact.. but it is not an ongoing thing.

4. CA can not replicate 2-way. There will always be a "source" and a "target". The only time it replicates in reverse is when you failover to the target and start the replication

5. CA and BC have nothing to do with each other. CA is remote replication. BC is local replication. If you have a schedule to snapshot/clone, it should continue to do so from the source vdisk(s).

Since BC is a local operation, there is no effect on a remote EVA if BC is not doing anything.


Steven
Steven Clementi
HP Master ASE, Storage, Servers, and Clustering
MCSE (NT 4.0, W2K, W2K3)
VCP (ESX2, Vi3, vSphere4, vSphere5, vSphere 6.x)
RHCE
NPP3 (Nutanix Platform Professional)
eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

Hi Steven
thks. A couple of points of correction.....

To say CA & BC have nothing to do with each other is not entirely correct. Yes the actual operations are different but both replicate data. BC provides the facility to ...if you like, mirror your data locally. Ie when your primary is written to, it will replicate to your BC. In fact, depending on how you set your instances up, one of the few differences would be HORCC_MRCF and of course HORC_INST.

These concepts become more relevant with XP's etc

BC can replace LVM mirror discs etc.

You can even regulate how quick BC copies sync up and
what to do in the event of heavy system i/o.

The concept of snap cloning et al though is correct but dont exclude the wonders of BC local disc sync'ing from your repertoire !

Thks for your input - Ill assign points to all at the end of the thread.
eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

.......also steve

You do realize that whilst BC allows for the provision of snap-cloning etc, it does provide a "continuous" copy right ?. Bearing in mind a true - safe copy would be achieved with a pairsplit (to allow all pending writes to finish etc)....blah blah blah
eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

A small addition to my original list
====================================

In the same host a and host b scenario....if a pairsplit is issued on the remote BC. All the pending writes to the remote P-VOL are flushed to the remote S-VOL to ensure the remote S-VOL is a "secure" snap shot copy of its P-VOL.

.......but if the remote P-VOL which is the locals S-VOL in CA, isnt pairsplit 1st, would the data on the remote BC split S-VOL be "safe".

....I hope I made sense.....
PP BIJU KRISHNAN
Trusted Contributor
Solution

Re: Theoretical question regarding BC & CA

Hi Eric,

I try my best to answer based on my understanding.

1. Yes to the BC and with a slight lag to the SVOL on RCU. If there is a problem with the link, then the updates are written to the primary journal volume and transmitted only when the link is available.

2. The CA PVOL can also be a BC PVOL in the PAIR or PSUS state. and the CA PVOL can be the BC SVOL in SSUS state.
You HORCM file must be creating one local copy and the remote copy from the same PVOL.

3. I dont see this happening unless you have BC worsening performance issues caused by other factors. So an already saturated DKA with you running split will aim disaster.

5. Wherever u run pairresync the series of operations would be the same. I dont think this matters.

6. BC performance is no way affected since this is a local operation. CA performance can vary depending on a variety of factors.

eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

thks - great answers btw !!
eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

.
PP BIJU KRISHNAN
Trusted Contributor

Re: Theoretical question regarding BC & CA

>All answers will have points assigned...at least those that dont respond with "see xxx manual". Many thks in advance.

both steve and me deserve to get some points I think, although this is not compulsary.
eric lipede_1
Regular Advisor

Re: Theoretical question regarding BC & CA

? - Both of you deserve point ?...what part of my actions indicated that you didnt ? I actuallu assigned point before you wrote that.
Also, points assignation is NOT mandatory the moment a response is written.

....or am I missing something ?