Dell MD3620F, MD3600f manual Asynchronous Replication with Write Order Consistency

Dell PowerVault MD3600f/MD3620f Remote Replication Functional Guide

NOTE: The write completion signal is sent in Step 3 as opposed to Step 7 in synchronous mode as shown in Figure 3.

Asynchronous replication is usually required when the remote site distances are greater than 10KM and cannot tolerate slowing the host application I/Os by waiting for remote write completions.

The number of write I/Os supported by asynchronous replication is distance determined. The usual rule of thumb is to allow 1ms for each 100KM in distance. Therefore a 1000KM (621 miles) distance will require 10ms for a round-trip RR I/O and will, therefore, limit the numbers of IOPs to 100. Longer than 1000KM distances will proportionally reduce this IOPs limit. It is imperative that the distance between MD36X0f storage arrays be a critical design factor that is surfaced early in the design and accounted for in the implementation of remote replication.

There are several reasons for highlighting the replication latencies. First, peak period writes (in IOPs) will determine whether RR can keep its replication queue from exceeding the queue limits. If the communications link cannot handle the peak period IOPs, and if the queues are then exceeded, then RR will suspend replication and go into suspend mode. We will take a deeper look into this later.

Asynchronous Replication with Write Order Consistency

For some applications such as database applications, out of order write I/Os will cause database recovery to fail and lead to data inconsistency and data loss. Databases operate under a strict sequencing protocol of writes to tablespaces and to logs. The logs are used to ensure the consistency or correctness of the database transactions as they occur and especially when they complete or commit. The order of the writes to each of the database tablespace objects that reside on virtual disks must also be the same order the remote writes are applied at the remote MD36X0f. This includes the same relative order within the same virtual disk and between the tablespace virtual disks and the log files.

Replication write activity in a synchronous mode guarantees the same write completion order on the remote array, providing the best chance of data recovery from the remote site. With asynchronous write mode, replication write requests are issued in parallel, and therefore not guaranteed to be sent and/or completed in the same order as received by the primary array.

In order to ensure that all writes to the secondary virtual disks are in the same exact order as they were applied to the local primary virtual disks, RR provides an Asynchronous Write Order Consistency Replication (AWOC) mode. The collection of virtual disks configured for write consistency is often referred to as a consistency group. AWOC provides a consistency group in which all associated virtual disks that need remote writes to be completed in the same order as the primary writes occurred are included. RR then ensures that all the writes for this consistency group’s virtual disks are sent and written in absolute in-order sequence. Only one write consistency group is allowed per array.

In Figure 4, when host issues writes X, Y and Z to the primary site array, preserved write order consistency ensures write requests are issued to the remote array in the same order as on the local array. So, the remote array receives the write request in the same X, Y and Z order as the primary system did.

Page 8