Storage Area Network Used To Optimize Standard Tape Backup.

SAN-based backup must be employed to handle the heavy workload of e-commerce databases. Various forms of SAN-based backup exist. This article discusses a backup system that incorporates the following three key characteristics: high performance, high availability, and low management. This system has been operational for four months at Net2Phone, the leader in Internet telephony services.

The first problem solved was I/O throughput, the second was continuous availability, and, of course, the customer wanted these solutions at a reasonable price. The remedy to this problem is a combination of hot backup and cold backup. A database running 7/24 should utilize a hot backup. Hot backup is a backup that occurs while the server is receiving new request from users and applications. If hot backup is attempted with tape, the network performance is greatly degraded. A cold backup is backup that occurs while the server is not receiving any new request from users or applications. Cold backups eliminate problems with network availability, congestion, and, if properly implemented, the need for an allotted window of time for backup (typically described as the "backup window").

A hot backup must contain the following elements in order to meet the criteria stated above: LAN-free backup, server-free backup, the ability to integrate multiple applications, and a script-based command line to facilitate management of backup. The key to this configuration is the SAN manager-an independent device on the SAN that manages from within the SAN itself. It must be able to integrate database and backup applications and perform unattended backups.

From LAN To SAN

To eliminate bandwidth contention, backup must be moved from the LAN to the SAN. This is the same concept as a dedicated backup network. Fig I illustrates how traffic is routed through only one system rather than two systems, which ensures that backup traffic does not share the same path as the IP network.

Serverless Backup 

The SAN manager issues a "third-party copy" command, which duplicates data from one device to another on the SAN without the assistance of the server. To overcome tape's inherent limitations, the SAN manager copies data from the source disk to backup disk versus backup tape as depicted in Fig 1. Once data is copied from one disk to another, the production of a duplicate copy is called a mirrored disk drive. The duplicate disk is, then, moved offline for a cold backup to tape media with a dedicated server (Fig 2). A command line script is utilized to perform daily, weekly, or monthly backups.

Application Integration

For this configuration to work, the SAN manager must be able to incorporate heterogeneous applications. In the SAN environment, this should be a given. However, many socalled SAN devices on the market today do not provide heterogeneity. As depicted in Fig 3 and in Fig 5, the SAN manager must be able to communicate with the source (or backup) server using a database application, as well as the backup application on the dedicated backup server.

Example SAN Configuration Optimized For Backup

To obtain the best possible reliability, the SAN should employ redundant (mirror) drives, redundant servers, and redundant channels. In other words, it should have no-single-point-of-failure.

* Drive failover: In Fig 3, you will see that the, drives are mirrored to ensure data protection. Within a mirror drive, if one drive fails, the other will seamlessly pick up where the other stopped and the bad drive is, then, taken offline. The SAN manager, then, replaces the failed drive with an available spare drive. Mirror drives have an added benefit of decreasing read time. If two drives are mirrored, then the read throughput is doubled. If three drives are minored, then read throughput is tripled, etc. The disadvantage is that write is also doubled or tripled, respectively. However, databases (especially ones designed for e-commerce) receive far more reads than writes.

* Server failover: The primary server uses a standby server for redundancy as illustrated in Fig 3. If the primary server crashes, then the high availability operating system running failover software will reschedule its task to the standby server. The failing server may then be repaired and brought back to an operational state without interruption to service.

* Channel failover: Redundant channels ensure uninterrupted access while, at the same time, increasing performance. For full channel fault tolerance, each device and server must have at least two ports attached to the channel. This provides the server or device with a second data path if the first were to fail, then the channel failover software will access data through the secondary channel.

* SAN optimized for tape backup: Fig 3 depicts the basic configuration of a highly reliable fault tolerant SAN as described before. In Fig 4 or step one of the tape backup process, the SAN manager via a command line script synchronizes backup storage 1 with the source storage creating a three-way mirror drive. This is a hot backup. All new requests to the server from users and applications go unaffected because the SAN manager removes the server from managing the backup. The SAN manager, then, splits backup storage 1 from the three-way mirror drive and transports it offline to a dedicated backup server where a cold backup to tape begins for day one of the SAN backup, as seen in Fig 4.

After backup storage 1 is split from the three-way mirror drive, backup storage 2 is, then, synchronized with source storage once again, creating a three-way mirror drive. This is the end of day one. On day two, the backup process to tape is repeated again, but this time with backup storage 2. This rotation is performed daily. However, the user may customize the command-line script to whatever schedule desired. In addition, during a catastrophic event, there is always an online copy of the database from which the user can restore. This highly available and highly reliable storage system is designed to handle databases for e-commerce websites that must run continuously, 24 hours a day, seven days a week and cannot afford shutdown time for tape backups or maintenance.

Dietmar Kurpanek is the director of hardware engineering at Vicom Systems