Rr Suspend And Resume With Delta Log Resynchronization - Dell PowerVault MD3600f Functional Manual

Dell PowerVault MD3600f/MD3620f Remote Replication Functional Guide

RR Suspend and Resume with Delta Log Resynchronization

It is important to be able to size the communications link for throughput and for IOPs. Peak period
measurement for the application's set of virtual disks is necessary in order to determine the bandwidth
required for remote replication. Peak period IOPs measured will determine whether the intended
distance of the remote site can support the peak IOPs. Additional factors that impact the peak period
requirements include future growth needs, the need to share a single communications link, and
whether there is a need to use redundant communications links for safety.
In the event that peak period write traffic exceeds the throughput and/or the IOPs capability of the
communications links, it will be necessary to consider using RR's Suspend and Resume features.
RR provides the ability to suspend replication by explicitly using the Suspend command and under
certain circumstances, to automatically suspend replication (consistency group suspension,
asynchronous replication when link bandwidth exceeded). This allows RR to continue to capture all
write I/Os but not use the overloaded communications links until a later time when the peak period
traffic eases.
Asynchronous replication will automatically suspend when the peak period traffic exceeds the
communications link capability (after exceeding the RR asynchronous queue).
AWOC will not automatically suspend when the peak period traffic exceeds the communications link
capability. Instead host I/O will be slowed down to keep the peak period traffic below the link
capability. This preserves data integrity of the remote virtual disks without resorting to RR suspension.
Suspending replication will create a delta log for each affected virtual disk at the time of suspension.
This delta log tracks the missing write I/O to the remote virtual disk for the duration of the suspension
and marks a data range for the write and will use this delta to sequence writes to that remote virtual
disk when RR resumes.
Resume will cause RR to resume replication for the suspended virtual disk. The delta log will be read
front to back from the primary virtual disk for all the missing I/Os in the secondary virtual disk and RR
will send these missing I/Os to the secondary MD36X0f. Once the delta log is completely processed,
RR resumes active replication of the primary virtual disk.
It is important to note that the delta log reads will be larger than the original write I/O due to the fact
that the delta log is composed of data ranges that are often larger than I/Os. Because the delta log is
fixed in size, larger virtual disks will have larger data ranges. This means the resumption time may be
longer than it would have taken if RR did not suspend.
If the communications link is sufficiently undersized (for example, due to peak period traffic growth),
there is the possibility that the RR resume operation may never complete. The assumption for the
resume operation completing is that there is sufficient time and bandwidth to process all missing
writes in the delta log before the next peak period occurs. This may not be the case if the peak period
traffic continues or if the non-peak traffic prevents emptying the delta log.
It is important to size the communications link to handle current workloads, future growth, and RR
resume operations.
Page 10