Vision Lays Out HA and DR Options in Well-Written White Paper

Published: September 22, 2009

by Alex Woodie

Remote journaling. XSM. Logical replication. iASPs. Disk-based mirroring. If one thing is certain, it's that there is no shortage of options for disaster recovery (DR) and high availability (HA) on the IBM Power Systems server. Thankfully, there are experts that can guide you through this technological maze, including the folks at Vision Solutions, who recently published a well-written white paper that takes an in-depth look at the wide spectrum of DR HA options for i OS and AIX environments.

In the new white paper, titled "Next-Generation Disaster Recovery and Availability Technologies for IBM Power Systems," Vision Solutions takes on the big challenge of comparing and contrasting the various technologies and approaches to DR and HA, including the advantages and limitations of each approach and its impact on recovery point objective (RPO) and recovery time objective (RTO), the two most important and useful metrics in the HA and DR business.

The hard work paid off for Vision, which filled the report with clear descriptions and objective analysis of nearly every DR and HA strategy imaginable. While the company could have been a little more consistent with its choice of terms (which is always a challenge in this field, where everybody seems to have their own ways of describing something), the overall affect achieved by Vision is to provide greater clarity of this important class of products. And Vision's intelligent use of graphics depicting RPO and RTO more than makes up for any shortcoming in the choice of words.

The result is that this report is a must-read for i OS and AIX professionals looking to improve their DR or HA strategies. And according to Vision director of marketing Bill Rice, it's also an industry first.

Here's a quick summary of the paper, in which Vision splits the spectrum of DR and HA in two categories: single and multi-system approaches. Single-system DR technologies and strategies covered in the report include the following:

• Tape-Based Backups--The most basic and primitive form of DR protection delivers RPOs and RTOs measured in hours or, in some cases, days. Installing tape libraries with multiple tape drives can speed RTO somewhat, as can a software product that keeps close track of tapes and guides the restore process. It's often used in conjunction with third-party hot site and cold site DR services.
• Tape-Based Backups Plus Journaling--Delivers better RPO than tape alone (journaling saves transactions that occurred just before the outage), but RTO is higher due to greater complexity and skills needed to recover the journals. On AIX, it's called Journal File System (JSF), but there is no equivalent to i OS remote journaling on AIX.
• Single-System Disk Mirroring and RAID--Protects from single- and multi-disk failures. Delivers RPO and RTO measured in minutes to hours.Requires 10 to 25 percent more disk capacity (for RAID), and at least twice the amount of internal DASD for full single-system disk mirroring.

Multi-System DR and HA strategies and technologies covered in the report include:

• Logical Data Replication--The time-tested standard for replicating data and objects between multiple i OS or AIX servers or LPARs. It takes many forms, including third-party proprietary solutions or as an integrated part of i OS through remote journaling (see below). Capable of delivering near zero RPO in synchronous replication mode (at the cost of slower application response time). Also available in asynchronous mode that doesn't hurt performance, at the cost of lower RPO. RTO is determined by the technology or technique used to recover systems. The big downside is tendency of source and target machines to get out of synch. The big advantage is it frees up hardware resources on primary machine by offloading things like tape saves and query workloads to target machine.
• Remote Journaling--Not an HA strategy per se, but one of the newer IBM i technologies enabling logical data replication, either as part of a third-party product or a homegrown HA/DR solution. It offloads most of the replication workload to the target system, which can also be used for performing tape saves or running query workloads.
• Logical Data Replication Plus Switchover/Failover--the capability to execute a planned "switchover" or an unplanned "failover" is what separates this class of HA technology from basic logical data replication discussed above. Vision classifies the technology that enables switchovers or failovers as a form of clustering. It also throws into this product category the array of "autonomic" capabilities, such as automated synch checks and the capability to "heal" problems that are found.
• Disk-Level Replication--So-called "hardware-based replication," or disk mirroring, includes IBM's Metro Mirror, Global Mirror, and Global Logical Volume Manager (GLVM) on AIX, and SRDF from EMC, and generic SAN copy services. It is newer and generally less understood than logical replication, and requires multiple expensive SAN arrays from IBM or EMC. Offered in synchronous and asynchronous modes, with slightly worse RPOs than logical data replication. Synchronous disk mirroring can only be conducted with systems separated by no more 168 miles. RTO objectives can be hampered because storage resources are not "switchable" and the need to conduct abnormal IPL. Users must log off systems before "flash" copy services can be utilized. Target systems typically can't be accessed for uses like backup or performing queries.
• Continuous Data Protection and VTL--Vision does not cover CDP technology in this white paper (despite claims to the contrary) because, it says, CDP doesn't offer the level of DR capabilities required by customers. Vision also didn't cover disk-to-disk, or virtual tape library (VTL) offerings, which can improve RTO and RPO compared to standard tape backups, and, when equipped with data replication and a second system, can deliver many elements of a successful DR strategy.
• Managed Journal/Log Recovery--This class of products can provide better protection than tape-plus-journaling by automating much of the work surrounding the management of journaled data. It offers a very good RPO. RTO is dependent on how quickly a customer can repair or replace a failed server. Products can run on Windows, Unix, or Linux, not just i OS. Vision doesn't mention it, but this class of product is increasingly being offered as a SaaS service over the Internet.
• Clustering--The highest form of HA is achieved through clustering, in which multiple i OS or AIX servers are hooked together and communicate intelligently as integrated nodes. Benefits include the elimination of data synchronization problems of logical data replication solutions (because there is only one copy of data in an iASP). Overcomes the RTO limitations of disk-level data replication (or hardware mirroring) by making iASP storage resources "switchable" among different nodes without requiring a time-consuming IPL. However, iASPs present a single point of failure, unless RAID or some form of mirroring, such as cross site scripting (XSM) or even disk- or logical-replication, is implemented in parallel. iASPs can't be shared among different servers for backup or query workloads. iASPs don't support a host of i OS object types, including user profiles and job descriptions, thereby requiring logical replication or restores from tape to restore a system.
• Cluster Enabled Application--The "holy grail" of HA delivers sub-second automatic failovers, basically eliminating the possibility of lost data; similarly, RTO is next to zero because there is no "recovery." The application is always online no matter what happens to one of the nodes. The big downside is: Does it even exist? Few vendors have invested in IBM's cluster-enabled API, relegating this class of technology to pie-in-the-sky dreaming.

The entire white paper can be downloaded (after a short registration process) at Vision's Web site at www.visionsolutions.com.

                     Post this story to del.icio.us
               Post this story to Digg
    Post this story to Slashdot