Ezine

This article can also be found in the Premium Editorial Download "Storage magazine: Storage Products of the Year 2010."

Download it now to read this article plus other related content.

Asynchronous and synchronous replication

Asynchronous replication is the most broadly supported replication mode, supported by array-, network- and host-based replication products. Committed to the source array first, then buffered or journaled for subsequent replication to the target array, data arrives at the replication target with a delay, ranging from nearly instantaneous to minutes or even hours. Its network latency and bandwidth tolerance make it fit for long-distance replication.

Not all asynchronous replication implementations are equal, though. Key areas of differentiation are how a product deals with network outages, if it supports transaction recovery or if it simply creates a crash-consistent replica that depends on the target OS and application to resolve inconsistencies. For instance, both IBM Global Mirror for the IBM System Storage DS8000 and the Hitachi Data Systems Universal Replicator have provisions to maintain the sequence of writes. "Hitachi Universal Replicator guarantees transaction recovery by sequencing replicated data within consistency groups," explained Sarah Hamilton, Hitachi Data Systems' senior product marketing manager, data resilience and security.

Rarely supported in host-based replication products, synchronous replication is the hallmark of high-end block-based storage arrays and also supported by most network-based replication products, including the Hewlett-Packard (HP) Co. StorageWorks

    Requires Free Membership to View

    Login

SAN Virtualization Services Platform (SVSP), IBM SAN Volume Controller (SVC) and LSI Corp. StoreAge Storage Virtualization Manager (SVM). Committing data to the replication source only after committing it successfully to the replication target, synchronous replication guarantees synchronicity between the replication source and target. A reliable network and low latency are prerequisites and supported distances can't exceed 50km to 300km, depending on the array vendor. Its primary use is for high-end transactional applications that require instantaneous failover if the primary node fails. It's less relevant in network-attached storage (NAS) unless the NAS can also serve as block-based storage for high-end transactional applications. A pure NAS play, such as a BlueArc Corp. Titan or Mercury system, usually lacks synchronous replication support. "A NAS doesn't require synchronous replication," asserted Ravi Chalaka, BlueArc's senior director of solutions marketing. Conversely, NetApp filers with their support for NAS and block-based protocols, especially Fibre Channel (FC), support synchronous replication, allowing NetApp arrays to compete with very high-end block-based storage systems from EMC Corp., Hitachi Data Systems and IBM.

Replication and the network

Available bandwidth and latency between the replication source and target systems greatly impacts replication performance. For short distances under 50km, latency is negligible and available bandwidth determines replication performance. The bigger the pipe, the faster a given amount of data can be replicated. Besides buying larger connections, compression and data deduplication are critical techniques to get the best out of given bandwidth. Deduplication and compression help shorten backup and restore times, and can result in significant connection and storage savings.

For long-distance replication, latency is the determining factor. The longer the distance, the longer the communication between replication source and target takes. Less of an issue for asynchronous replication, latency is the determining factor in synchronous replication where a distance of 50km is usually the maximum acceptable range. While replication implementations make a difference (for instance, IBM Metro Mirror supports up to 300km between source and target), the impact of latency can be reduced and supported distances extended through the use of WAN accelerators, which perform TCP optimizations and implement techniques to minimize roundtrips. WAN accelerators are available from Cisco Systems Inc., F5 Networks Inc., Juniper Networks Inc., Riverbed Technology Inc. and Silver Peak Systems Inc

Array-based replication

Early on, a mechanism to replicate data from one array to another emerged as a necessity and array vendors quickly added replication to their storage systems -- to high-end arrays first, where it's standard now, and then to midrange and lower-end arrays. Dell Inc. is a perfect illustration of the trend of replication filtering down into the low end of the data storage market. Today, all of Dell's storage systems, with the exception of the lower-end PowerVault arrays, support replication, from Dell/EMC SAN Storage and Dell EqualLogic arrays to the Dell DX Object Storage Platform.

Having replication be a part of the array has many merits. For storage managers it's simply another array feature. Managed similarly as other array functions and options, it takes little effort to leverage replication. Because it's an array function, deploying it requires very little cross-departmental coordination; it's the storage group that makes it happen and supports it once deployed. Provided by the same supplier, array-based replication is supported by a single vendor, thereby eliminating a great deal of finger-pointing when problems occur. Furthermore, array-based replication is less likely to be disrupted by extraneous activities such as patching and other changes, which are more likely to pester host-based replication products, giving it a higher degree of resilience. "Application failures won't impact array-based replication because the storage system isn't impacted," said Mark Welke, director of data protection solutions at NetApp.

This was first published in February 2011

Join the conversationComment

Share
Comments

    Results

    Contribute to the conversation

    All fields are required. Comments will appear at the bottom of the article.
    Back to top