Open Replicator - Hot Push/Dedicated FAs question

Yes, dnynamox. The rest of the configuration for OR will be done as required.. should have mentioned that !

Girish,

The COFW impact will be largely dependent upon the host write workload and the FA utilization... but you can mitigate the COFW effect by running the sessions in precopy mode for at least one cycle before activating them. I'm not aware of any docs/KBs that directly confirm your original question about dedicated FA for ORS. I've done it myself though, as have several others.

Thanks,

- Sean

all FAs that devices are mapped to will need to be zoned to target array and participate in OR session.

Thanks Sean. How do we estimate whether COFW will have an impact on FA performance ? Is there a document/KB article on what you just mentioned that we can take to the customer ?

Can this be done for hot pull ?

Sure, you could do the same thing for a hot pull. The same idea would apply here -- you'd set a 0% ceiling on the control FAs that your host(s) use, and set a 100% ceiling on your pseudo-dedicated migration FAs. This would send the background copy traffic over the pseudo-dedicated FAs, but the CopyOnFirstWrite/CopyOnFirstAccess traffic would still be on the host-facing FAs.

It works this way because there's a separate instance of Open Replicator that runs on every control FA slice. A particular instance of OR on one FA cannot delegate tasks to another FA (i.e. there is no mechanism for one FA to forward an outstanding IO to another FA for servicing). So when a host IO arrives on a particular control FA that triggers a COFA or COFW, that FA has to handle the COFA/COFW work, even if there is a ceiling of 0% on that FA. The background copy process is not tied to incoming host IO though, so that traffic can be evenly distributed among those control FAs with non-zero ceilings.

Dilipchalla,

Not sure I'm entirely following your question... but any COFW/COFA activity will use the "production" FAs. Only the background copy activity will use the dedicated FAs, assuming you configure your ceilings as described previously in this thread. For the background copy activity using the dedicated FAs, there is no IO transfer from the "production" FAs to the dedicated FAs required -- the dedicated FAs just read blocks from disk as necessary, and transfer these blocks to the remote array.

All IO goes through global cache; there is no avoiding that.

If you're concerned about the utilization of the array that will be running Open Replicator, you can throttle the rate of the Open Replicator sessions (and thus the resources that they consume) by applying a lower ceiling to the dedicated FAs.

Thanks,

- Sean

Hi Sean,

This was very informative but do you know how the copy happens in backend. Because we will be using dedicated director ports which are different from director ports getting host IO's will the tracks from host director ports use global cache to transfer data from them to dedicated ports used jut for OR. If it uses global cache then there will be impact on Servers right and also what if the array is utilized 95%. Your reply will be appreciated.

Dilip,

For the background copy, there is no need for the the host ports (2f0 and 15f0) to internally transfer the data to the dedicated ports (5f0 and 12f0). The dedicated ports can perform the background copy on their own, and the host ports can perform COFA/COFW activity on their own.

All IO operations involve global cache, but whether you use dedicated or shared FAs for Open Replicator has no effect on the amount of cache Open Replicator uses.

Thanks,

- Sean

Hi Sean,

I will explain you my scenario

Let's say 2f0 and 15f0 are the ports that are connected to host. Actually we have oracle db servers which are throwing out errors during OR. Experiencing high latencies on few of the devices and db servers went down.

Then if I choose 5f0 and 12f0 as dedicated ports for OR. Then I will zone all these ports (2f0, 15f0, 5f0 and 12f0) for OR session. Then will set ceiling 0 on 2f0 and 15f0 and ceiling 70-100 (depending on copy rate and best option) on 5f0 and 12f0. Here my question comes actually.

My other team mate is saying as these dedicated ports are not zoned to hosts, for copying data from those devices the host ports 2f0 and 15f0 should transfer the data to 5f0 and 12f0 using global cache.

But as per my understanding we don't need global cache for the backend copying right? we don't need those devices to be zoned to active hosts for copying data directly to target devices.

So will there be any extra utilization of global cache due to the OR I am planning which involves dedicated ports. Will it be more worse or better or same than not using dedicated ports regarding the utilization of global cache.

As you said earlier any IO's will use global cache I agree with that but why do we need global cache to copy from dedicated ports? It doesn't make any sense to me saying that using dedicated ports will cause more utilization of global cache. Does it make sense for you? Once the IO's come to global cache from hosts then target disks should access the same cahe for OR session right? If you agree or not could you please share your thoughts..

Best Regards,

Dilip Challa.

Thank you Sean for the quick reply. That answer's my question.