Data Protection: Tape, Drives, And Beyond

 April 5, 2011, Storage Network World, Santa Clara, CA—Michael Fishman of EMC Corporation presented concepts of data protection in an era of shrinking time windows for data safety and recovery. Data protection is related to data availability and the ability to recover those data under any circumstances.

Starting at the end, recovery is the ultimate goal for any data protection process. The first step in recovering data is to detect a data fallout and diagnose and decide on restoration actions. To recover data at a point in time, you first need to determine when the data was last valid. Unfortunately, you have to redo the logs and all work done after the back-up point. To do a recovery quickly and accurately, you have to practice, test and verify the data and procedures.

Traditional recovery methods get to a last known good image and look at the change log to detect the failure. Then you restore everything. Databases are hard to recover, and files are easy. Protection and recovery are on a sliding scale that compares risk and costs, there is a range of solutions to pick from. The tradeoffs in cost, level of protection, and risk are determined by how the time off-line affects the business. The recovery point determines the amount of backup at risk.

The topology of the network components are defined by the entities and their roles. The backup server is an auditor. The storage node or media server is the mover, which requires an agent to determine what to save and the characteristics of the applications server. All of these must communicate with the secondary storage systems in real time.The networking configuration is one of the keys to failure recovery. A SAN (storage attached network) has high performance and high bandwidth for application hosts and backup hosts. A NAS (network attached storage) is direct attached to the network and host machine. Newer configurations are becoming cloud based and provide spatial diversity.

The intent for the backup is to protect against physical and logical failures in the system. A LAN-free local data mover has the applications server direct the data to the backup system. This puts a high load on the rest of the network and the server as the metadata must go to the catalogue in a separate store and data is routed to the backup secondary storage systems.

Using the LAN with a separate agent like a media server frees up resources on the applications server, but increases LAN traffic. This is a good configuration for mirror or snapshot backups. The applications server forces a backup and traffic remains local in the SAN and not on the LAN. A server-less backup goes through the data mover and the SCSI xcopy function moves the metadata to the catalogue.

A cloud backup moves the data in the same way, but moves the traffic to the WAN, requiring more activity in the host. Some of the host activities include de-duping the data to reduce data volume before backup. One control issue is data protection on the wire and at the cloud site. Multi-tenancy and other security issues are paramount. In addition, the WAN network performance is a consideration.

Cloud storage can lower capital expenditures and offer automatic off-site protection. This protection is enhanced if just the changes are sent over and not have a full backup all the time. For incremental backups, the metadata is critical to performance and recovery. The problem is that it may take a number of incremental snapshots to recover the full data set.

A full backup copies everything. This copy can be performed hot, with data still active, or cold, with data at rest. An incremental backup only saves the changed data, so recovery time can be longer due to the need to find all the files and their metadata. A differential backup saves all data that changed since the last backup. This method uses more data storage than incremental but takes less time to restore.

A synthetic full backup uses a series of incremental backups to regenerate the full backup. A variation is incrementals forever, which consolidates incremental saves. This method is complex and has a higher risk due to the long times between full backups.

Understanding the various backup methods is only part of the equation. It is just as important to know what to backup. Backups can be at the file level, block level, and a new trend is to consider client-side data. The client-side backup moves the backup agent to the applications server to offload work from the net.

Storage media are another facet of backup. Disks and SSDs have lifetime issues, so the tradeoff is speed versus lifetime. Tape is slower, since it is sequential, but is the lowest cost, long-term medium. Tape now includes integrated wear integrator and is self correcting. The downsides are that tape needs complex robotics, barcode scanners, and media management tools.

For tape to be viable, the user needs to know where the bottlenecks are, and needs to have a well managed architecture for storage. Encryption provides security for the data. Backup to disk, on the other hand, allows random access and RAID configurations provide more data protection. The issues to consider are the higher costs and complexity of the drives and the need to use different protocols for backup storage. A virtual tape uses tape methodologies on disk arrays. This allows the retention of existing software and eliminated any additional training for the operators.

Continuous data protection requires a complete plan that addresses the differences between a full backup and snapshots or continuous background saves. the user has to define appropriate granularity for the data; file, block, or application-based structures. Capturing all change as it occurs is expensive and complex, but enables zero lost data in a zero time backup window.

Backup is less of an issue if the data is de-duped before storage to reduce redundancy, but you have to know the data use case. Reducing bandwidth for backup gains about a 5:1 improvement for de-dupe and an additional 2:1 for compression resulting in a 10:1 reduction in data for backup. The level of protection has to balance risk and cost while matching optimal performance and throughput goals with the technologies.