This post is more than 5 years old
2 Intern
•
522 Posts
0
1335
February 25th, 2016 13:00
Synchronous RP Replication and Distribution
Hello,
Just a quick query on synchronous RP replication and jvol distribution. Assuming the snapshot granularity is set to dynamic, what can the user expect to have for a speed of distribution to the replica given the ACK is given back to the primary host once the write reaches the target RPA or target jvol?
Along those same lines, when looking at the max journal lag, if a user was to enable a test image and leave it enabled for a long period of time to the point where all previously distributed snapshots rolled off and only new, non-distributed snapshots were in the jvol, does the system pause replication at that point or do those new non-distributed snapshots continue to roll off as well (assuming the TSP portion of the jvol hasn't filled up from being enabled)?
I haven't had a chance to test the stuff above and was trying to remember how some of these mechanics occurred after a discussion today.
Thanks!
-Keith
forshr
2 Intern
•
1.1K Posts
0
February 26th, 2016 05:00
Re the second part question, yes, if the 'Do' queue fills because of prolonged image access full distribution cannot occur from the point post of the snapshot selection, i.e. any newer snapshots.
echolaughmk
2 Intern
•
522 Posts
0
February 25th, 2016 13:00
Thanks Rich,
So just for clarity (been a while for sync
) when the I/O is ACK'd, it is done once the I/O hits the RP or JVOL on the target site and I think the distribution to the replica is done immediately following the ACK being sent back - is that correct? I was trying to validate with some old docs, but haven't found what I thought I had seen before so was trying to confirm if the distribution was done immediately following that ACK. Basically wanted to avoid a scenario where the I/O is ACK'd, sits in the jvol, the jvol blows up, and that one write is not written to the replica in a worse case scenario.
OK, so it will pause as it seems in this setup. I know it is not recommended to have the image enabled that long to make this happen, but the question came up and I couldn't remember if it paused the CG or if it started to drop off the non-distributed snapshots that had been replicated during image access, but not distributed to the replica because of image access. I thought it might just roll them off, but it sounds like it doesn't do that if I am reading you correctly.
Thanks!
-Keith
forshr
2 Intern
•
1.1K Posts
0
February 25th, 2016 13:00
Hi Keith,
When you set the replication mode to sync the snapshot granularity is set to per I/O. The distribution overhead does no change as the replication mode does not impact this operation since the I/O is ACKed prior to distribution (unless the default is changed).
When image access is initiated the journal lag will increase due to only one of the distribution phases takes places, i.e. the first where I/O are written to the head of the 'Do' queue. Once this fills and the RPA memory fills, replication will pause. This is referred to as a temporary highload as the temporary nature of this action can be controlled by disabling image access.
Regards,
Rich
forshr
2 Intern
•
1.1K Posts
1
February 25th, 2016 14:00
Hi Keith,
When replicating remotely in sync mode the ACK occurs from the remote RPA. Distribution is primarily dependent on the journal/replica performance to ensure a minimum level of journal lag and therefore immediate distribution.
When replication is paused (on a per CG basis), RP initiates marking mode to the Prod copy journal.
Regards,
Rich
echolaughmk
2 Intern
•
522 Posts
0
February 25th, 2016 16:00
Great...thanks for confirming that first part Rich.
For the second part, what I am curious on is when it goes paused and it sounds like if I have 10 snapshots in the jvol and I enable the latest, if I get 10 more incoming, the original 10 roll off (these would have been distributed to the replica already). For the new 10 coming in, they wouldn't be distributed to the replica cause it is in image access mode. When the 11th comes in, the portion of the jvol for incoming snapshots is full - it sounds like rather than rolling off one of the 10 that aren't distributed, at this point the CG will pause - is that correct?
I'm good on the marking mode process, thanks. I was trying to find out when the CG would pause in this scenario if the TSP portion never fills up but the image is enabled for a long time such that the journal portion effectively has no snapshots left that were previously distributed to the replica.
Thanks!
-Keith
echolaughmk
2 Intern
•
522 Posts
0
February 26th, 2016 05:00
Great...thanks Rich!