This article can also be found in the Premium Editorial Download "Storage magazine: Storage managers give thumbs up to IP storage."
Download it now to read this article plus other related content.
Disk backup options
All major backup application software vendors now have disk backup options that take full advantage of the unique properties of disk as compared to tape: random access, speed and relative reliability. Instead of laying down data in tape format, the backup application takes advantage of the random access nature of disk to lay down data consecutively for more rapid restores. In addition, backup applications now support concurrent read/write operations reducing the time required for restore, staging and cloning operations.
Legato NetWorker has introduced an advanced file type device for this purpose. Products like Veritas NetBackup 5.0 enable users to create a synthetic full backup from a full backup and a series of incrementals, dramatically reducing the need to perform a full backup on a regular basis.
To take advantage of the random access capabilities of disk, for example, backup applications multiplex streams of data from several application servers (see "
Data reduction is a great way to reduce secondary storage. On average, the ratio of backed up data to primary data is 10:1. What if the same amount of information that is buried in 10TB can be squeezed into 1.2TB of secondary disk? This is now possible with CAS technology, which is sometimes called object-based storage. Although CAS technology was first commercialized by EMC Corp. with its Centera and targeted at the archival market; other vendors, notably Avamar Technologies, have applied CAS to create an innovative backup and restore solution that is focused on reducing backed up data stored on secondary media. Data Domain, Permabit Inc. and a number of other newcomers have also entered this market.
Here's how Avamar, for example, cuts down the amount of stored data. The first time the product is used, each file is broken into variable sized chunks of data (not to be confused with blocks). Each chunk is usually stored only once--or a few times, depending upon one's requirements for additional protection--and a file is re-created on the fly from chunks, independent of their location. The agents on the application servers (clients to the backup application) determine if a chunk already exists on the secondary storage. If it does, it isn't transmitted across the network.
Given the high level of duplication in full backups and even incrementals (when a single bit is changed many products transmit the entire file over again), the reduction of data is massive. Avamar claims a 99% reduction in network traffic when compared to traditional backup methods. This powerful data reduction capability makes possible a variety of new data protection schemes, including the cost-effective replication of data to secondary disk storage at a disaster recovery site over IP. In some enterprises, the technology has the potential to eliminate tape completely as a backup medium.
Although CAS promises to drastically reduce the ratio of primary to secondary storage there are downsides. Avamar, for example, requires that users replace their existing backup application with Avamar agents, and all backed up data is stored in a format that only Avamar can decode.
These data protection approaches--when used in conjunction with snapshot technologies--only provide data restorability to specific points in time. For example, if snapshots are taken hourly, an average of 30 minutes of data may be lost. Unfortunately, the only viable alternative to date had been to increase the frequency of snapshots. But now, companies such as Revivio Inc., Timespring Software Corp., Vyant (now Mendocino), XOsoft and others are delivering solutions that provide data protection at any point in time.
Revivio, for example, focuses on critical application environments that have traditionally relied on expensive array-based snapshot technologies like EMC's TimeFinder to provide point-in-time copies of application data. To minimize the risk of data loss, administrators must create multiple concurrent business continuance volumes (BCVs) on their Symmetrix arrays, which is very costly. And if the primary volume gets corrupted, the recovered volume could substantially lag behind the primary volume and data will be lost.
The Revivio solution deals with this problem by attaching its appliance to the SAN and creating a third mirror of the live volume. With this technology, Revivio claims that there's no need to create BCVs or maintain mirrors of BCVs. The result? Your storage needs just dropped dramatically. Every write to the primary storage is captured (no agents are required to be placed on application servers), time stamped and redirected to the Revivio appliance (see "A typical Revivio setup").
If the primary volume gets corrupted--say at 2:30 p.m.--the IT administrator can ask for the volume as of 2:29 p.m., and the Revivio appliance will re-create it from the combination of the third mirror and its TimeStore software by essentially deleting all writes that occurred between 2:29 and 2:30 p.m. The volume as of 2:29 p.m. is mountable directly from the appliance as soon as it is ready (a few minutes) and the application can be restarted. Now, the 2:29 p.m. volume is fed into the primary storage, and as a second step brought into complete harmony with the current state. At this time, the triple mirror and TimeStore are returned to their regular responsibility of data protection and the primary volume takes over. Other products apply variations of this theme, focusing on Exchange and SQL Server environments.
These solutions aren't for everyone. Only the most critical applications--especially those where downtime costs are astronomical--need this level of protection. Another fact worth noting is that while these applications will reinstate a volume at any point in time in the past, its consistency is only guaranteed as crash consistent. This means that the volume is presented as it was at that moment in time. Because application consistency is what one ultimately cares about, a crash-consistent image takes you a step closer, but it's not a full solution. The administrator may still need to apply logs in order to create the right total picture for a database application. The only way to get application-level consistency is to take point-in-time images. And then you are back to where you started.
Revivio only comes into play when volumes need to be recovered, and later resynchronized. The solution is used for volume--not file recovery--and is most effective for database applications. Other vendors--XOsoft, for example--are claiming application-level consistency and file-level recovery, so it's crucial to drill down and understand these critical differences.
This was first published in April 2004