Backup and Restore Benchmarks for Very Large Database Applications

These benchmarks were performed in order to demonstrate the feasibility of storage management solutions for VLDB applications. This study establishes a range of benchmark results for backup and restore of large Oracle7 installations running on a DIGITAL AlphaServer 8400 system, in conjunction with StorageTek SD-3 Redwood tape subsystem, Oracle's Enterprise Backup Utility and DIGITAL NetWorker Save and Restore. These results were achieved using a new high performance version of NetWorker that is currently under joint development by DIGITAL and Legato.

Three benchmark scenarios were performed to represent real-life backup and restore situations. These benchmark scenarios consisted of the backup of an Oracle database that was off-line (cold backup), the backup of an Oracle database that was on-line without a transaction load (hot backup), and the restore from backup of an Oracle database.

The following results were achieved:

• 367 gigabytes per hour for a cold backup
• 420 gigabytes per hour for a hot backup
• 136 gigabytes per hour for a restore

Benchmark installation

Introduction

Managing very large database applications

While it is now common for RDBMS applications to range from 10 to 100 gigabytes (GB) in size, the last several years have seen the emergence of very large databases, or VLDB for short. The definition of a VLDB is a database ranging from hundreds of gigabytes to terabytes in size. The dramatic decline in the cost of disk storage, coupled with the growing popularity of data warehousing applications has also fueled the amount of RDBMS data that must be managed. According to Gartner Group, the average database will grow 1000% between 1995 and 2000, or an average of 200% per year.

The impact of all this growth in data being used by RDBMS applications is a growing concern of database administrators (DBA) and IS management as to how this data can be protected in case of application error, operations error, hardware failure or disaster. For example, many installations have service level agreements that require RDBMS applications to be available for the majority of the day or up to 24 hours a day, 7 days a week. The loss of even one hour's availability for some applications can cost an organization thousands or hundreds of thousands of dollars in lost revenue and productivity. All of this compounds the workload of the DBA who must perform many duties other than backup and restore.

Solutions for Oracle7 Data Protection

Hot backup: A backup performed while the database is online, available for read/write.
Cold backup: A backup performed while the database is off-line

Prior to EBU, Oracle7 users relied upon export and import along with physical backup. Export and import has the limitation of not being able to work with the redo logs (a.k.a. transaction logs) while the physical backups do not provide the performance needed to backup and restore large databases. Both suffered from the inability to utilize high speed tape devices and libraries and lack features such as scheduling, monitoring and media management.

While high speed backup and restore is a critical element, it is not the only dimension of a comprehensive VLDB data protection strategy. Installations looking for a RDBMS backup and restore solution should look for ease of use (scheduling, monitoring, tracking) along with functionality (disaster recovery, database and filesystem data support, media management). Also of importance is heterogeneous support, allowing an organization to standardize on one product that has components that support multiple RDBMSs, operating systems and applications.

What use are benchmarks?

While it is true that "power" benchmarks using state of the art hardware and software with databases in the hundreds of gigabytes to terabyte size seem to be more prevalent, they serve a purpose. They demonstrate to installations that currently have or who are considering VLDB applications that viable data protection solutions are available.

For installations that do not or will not have VLDB applications in the near future, the "power" benchmarks give them the security that a product they invest in today can provide the scalability they may need in the future. Lastly, VLDB benchmarks highlight the fact that the leading RDBMS and storage management vendors are architecting their software to make use of the latest hardware technology.

Benchmark Results

Benchmark 1

Cold backup - 367 GB/Hour

Benchmark 2

Hot backup without a transaction load - 420 GB/Hour

Benchmark 3

Restore - 136 GB/Hour

Benchmark Environment

Explanation of the benchmark process

• Data compression was not used for any of the tests. Data compression can produce misleading results, which depend more on the exact data being saved than on the software doing the saves. It was felt that avoiding compression altogether would provide results that were not controversial. With compression off, the transfer rate of the SD-3 tape drives is near 11MB per second. With an average transfer rate of 4.5 MB per second per disk drive, it was determined that 4 disks should always be multiplexed to one tape drive. The factor of 4-to-1 was maintained as the number of tape drives was increased. Data was recorded for 3, 6, 9, 12 and 15 tape drives. (During the run of 15 tapes, one drive had only 3 disks multiplexed due to physical limits).

• Read buffer size was set to 2MB for both hot and cold runs. The tape write buffer size was set to 256K, due to limitations of the SD-3 drives.

• Cold tests were run using the entire disk while hot tests used only the data partition, leading to the difference in the amount of data backed up between the two types of runs. During hot backups, archiving was set to "none" within EBU. This type of archiving is not needed for consistency during a restore operation. Database control files were transferred to tape but not included in the calculations for sustained transfer rate. Database control file backup did factor into the elapsed time calculation, adding less than one minute.

• For the 15 tape test cases, the entire database was backed up. Generally, the contents of four disks were used for each tape drive used in the 3, 6, 9 and 12 tape drive test cases.

• Elapsed run times were included in the data, including tape mounting/labeling, database table altering, database instance manager startup, and ramp down times. No physical mounting of the tape media was required during the test runs and thus it was not included in the elapsed run time.

• Sustained transfer rates were measured using the following method: NetWorker provides data on the rate of data being written to each tape drive, averaged over a tunable period. Internally, this is accomplished by keeping track of the number of bytes written to each drive during each polling interval, and then dividing by the time elapsed. For each run, the rates for all drives were captured every 30 seconds for the duration of the run. We then went back and summed the rates for all drives at each snapshot, and averaged the samples which represented the sustained run. Samples discarded from this calculation included ramp-up time, unrealistically high results caused by filling the onboard cache of the tape drives, and the ramp-down times.

• The "Wall Rate" is the calculation of the total number of bytes moved to tape, divided by the elapsed time.

• CPU utilization was calculated using the command, "iostat 30" and cpuinfo, which graphically shows the differential usage of the 8 CPUs. The results over the run were averaged together to determine the CPU utilization. Hot elapsed times were recorded by OEBU trace logs. All other calculation of performance were recorded with the logs and monitoring utilities provided by NetWorker. The information was validated with the DIGITAL utility "monitor," which would display the CPU load by processor and total kilobytes transferred per second.

Benchmark installation

Hardware configuration

Software configuration

Database configuration

Table data structure
Tables	59, one per disk, each 3.1 GB in size
Index	59, one per disk, each 0.7 GB in size
Rows	26,000,000 per table; 1.534 billion total
Total allocated size	224.2 GB
Total used size	approximately 172 GB

Table data structure
Field Name	Data type
L_ORDERKEY	NUMBER(10)
L_PARTKEY	NUMBER(10)
L_SUPPKEY	NUMBER(10)
L_LINENUMBER	NUMBER(10)
L_QUANTITY	NUMBER(10)
L_EXTENDEDPRICE	NUMBER
L_DISCOUNT	NUMBER(10)
L_TAX	NUMBER(10)
L_RETURNFLAG	CHAR(1)
L_LINESTATUS	CHAR(1)
L_SHIPDATE	DATE
L_COMMITDATE	DATE
L_RECEIPTDATE	DATE
L_SHIPINSTRUCT	VARCHAR2(25)
L_SHIPMODE	VARCHAR2(10)
L_COMMENT	VARCHAR2(27)__

Future Direction and Summary

The future of Oracle database backup and restore

NetWorker® BusinesSuite™ Module for Oracle®

The BusinesSuite Module for Oracle exploits EBU by integrating the database aware and parallel processing performance into NetWorker's parallel, high speed engine that supports a broad range of high performance tape devices. The BusinesSuite Module also provides administration, scheduling, monitoring and media management. This combination provides the safest, fastest and easiest to use solution available for Oracle data protection.

Oracle7 and EBU

Oracle8 Recovery Manager

Key features of RMAN include the ability to perform incremental backups, archive log management and a new recovery catalog. The option of performing incremental backups is significant because this gives VLDB users the ability to backup only the data that has been modified. This will result in greater database availability along with reduced resource requirements such as tape and cpu. RMAN will automatically utilize the incremental backups along with full database backups that have been taken during a specific time (e.g. weekly).

While EBU is not planned to be supported against Oracle8, all existing EBU processes, products and scripts can still be used against Oracle8. Legato will support and exploit both EBU and RMAN in addition to internally developed solutions.

Summary

NetWorker® BusinesSuite™

The combination of BusinesSuite and NetWorker provides a network-wide storage management software solution that delivers a rich set of advanced features to support enterprise applications, such as autochanger support, automated media labeling and tracking, and media cloning for local and off-site storage strategies, as well as support for state-of-the-art high-speed devices and mainframe-class tape silos. NetWorker, in concert with BusinesSuite, is the only comprehensive storage management solution that addresses the need for cross-platform support of enterprise applications running on Microsoft Windows NT and UNIX.

Digital Equipment Corporation AlphaServer 8400 5/300

Whatever applications you're currently running on your mainframe, the AlphaServer 8400 system - powered by the breakthrough 64-bit 21164 Alpha chip - can run them faster than anyone. You can equip the AlphaServer 8400 system with up to twelve 440 MHz processors. That gives you a competitive advantage, with the industry's fastest enterprise server performance - running at over 3,000 transactions per second (TPS) with symmetric multiprocessing capabilities. High-speed processing is only part of the story. The AlphaServer 8400 specs show this system to be a veritable client/server juggernaut. Only on a 64-bit system can you put gigabytes of your database in memory, and get a performance increase far beyond the 2 GB limit of any 32-bit system. Internal memory starts at 256 MB and can be expanded to 14 GB as your need for additional and more complex applications increases. Storage capacity tells an even bigger story: it is capable of growing beyond 10 TB.

Consider the cost-efficient mainframe alternative or super-computer replacement - the AlphaServer 8400 - a system that can fulfill your open client/server needs today and product your investment tomorrow.

Oracle7 and Oracle^® Universal Server^®

Oracle Universal Server allows you to store, manipulate, retrieve, and share any type of data - including business records, documents, messages, images, audio, and video - in any type of application, such as transaction processing for improving the bottom line, data warehouses for making better decisions, network-centric systems for streamlining the way your business communicates, and electronic commerce for building new markets and automating more processes.

Oracle Universal Server and Oracle7 allow you to securely and reliably deliver these powerful applications to any number of users, over any type of network, no matter what hardware and software they use.

Oracle Enterprise Backup Utility (EBU)

STK Redwood™ SD-3 Helical Cartridge Subsystem

The Redwood SD-3 revolution is the result of the collaborative efforts of three industry leaders - Storage Technology Corporation, 3M and Matsushita Electric Industrial Co., Ltd. of Japan. This collaboration has resulted in the development and implementation of a unique storage strategy, one that fulfills all of the immediate and long-term storage capacity requirements of a broad spectrum of industries. Global satellite telemetry data capture, seismic oil exploration, financial institutions, multi-national insurance companies, television broadcasting - these are but a few of the data-intensive, technology hungry industries that will realize immediate benefits from Redwood's capacity and performance.

Contacts

Digital Equipment Corporation
334 South Street
Shrewsbury, MA 01545
www.digital.com

Legato Systems, Inc.
3210 Porter Drive
Palo Alto, CA 94304
415-812-6000
www.legato.com

Oracle Corporation
500 Oracle Parkway
Redwood Shores, CA 94065
415-506-7000
www.oracle.com

StorageTek
2270 South 88^th Street
Louisville, CO 80028
1-800-STORTEK
www.stortek.com