|
http://www.washingtontechnology.com/print/22_13/31083-1.html
07/23/07; Vol. 22 No. 13
Not
just for disasters anymore
Backup and failover databases find new uses beyond recovery
Later this summer, the computer system for the 311 dispatch
center in
One potential source of relief is the failover database
system. The city is considering running reports using data in the failover
system rather than the primary database system, said Wesley Jackson, manager of
business applications at the city’s communication and technology
management office.
The idea of using the failover system or other backup
databases for duties in addition to disaster recovery is not new. It’s a
feature users have been requesting for a while, said Juan Loaiza,
senior vice president of database development at Oracle Corp. “In order
to get protection from site failures, they have to set up an entire other data
center — this entire other computer system, this entire other database
— and feed it this information, but they’ll never use it except in
case of a disaster,” he said. “Nobody wants to do that, because
site disasters don’t happen every day.”
Yet, with recent replication and failover technologies
improving, accessing and using that data is becoming easier. And organizations
are finding uses for that data — and those databases. “The data
that resides on a failover database has some value,” said Bill Cooper,
vice president of data warehouse system provider Teradata,
a division of NCR.
Like all cities and towns,
Such reports are based on the service requests filled out by
the call center operators. When a new call comes into the center, employees log
the information, which the database then stores. When reports are required,
they are easily generated by tapping into the database. The database servers,
running instances of Oracle 9i databases, are located at the call center.
Shared load
This summer, the
Traffic coming into the call center is expected to double or
even triple. As a result, the city’s IT department is rethinking the idea
of running the reports from the primary database. With the increased traffic,
the city can’t afford to let the database be slowed by the generation of reports.
“We don’t want to have a major incident where we
have to pull reports for a 311 call center [and] at the same time have our
service reps take in [a surge of] support calls,” Jackson said. “We
had this failover database at this other location, so we thought, ‘Why
not point to that failover database and run our reports from there?’
”
Initially looking into this possibility, however,
Others are kept in a failover status, meaning they can be
used as a primary database in an emergency should the primary one fail. In
fact, current software, such as that offered by SunGard or Oracle, allow systems to switch between the primary and backup
databases so quickly that users don’t notice the switch.
Loaiza said Oracle has made great strides
in making failover or backup databases more readily available for additional
purposes. This is not easy, given the trickiness of updating and reading
databases.
In the most basic configuration, the primary database makes
a log file of the changes, which then can be shipped to the secondary database
and applied to make the two databases identical. Oracle has long offered this
approach, which it calls Critical Standby. Users could query Oracle’s
Critical Standby database when it wasn’t being updated with new material.
However, both actions couldn’t be done at the same time.
“So what people would do is switch back and
forth,” Loaiza said. “Maybe at day,
they’d read the database, and at night, they’d apply the
changes.”
This proved to be unsatisfactory to some users, so about
seven years ago, the company introduced something called the Logical Standby
database, which allowed reads and writes at the same time.
“It’s more flexible,” Loaiza
said. But the downside to Logical Standby, he said, was that it wasn’t as
fast because a lot more conversion of data was needed. It was also more
complicated for administrators.
With the upcoming release of Oracle 11g, a new feature called
Readable Physical with Real Time Queries will offer the best of both Logical
Standby and Critical Standby, Loaiza said.
The agility of new tools such as Readable Physical with Real
Time Queries can give organizations a lot more flexibility. For example,
organizations can bring down the primary system to apply patches and updates
while running the failover system in its place. Then when the updates are
finished, they can bring the secondary system down for its upgrades.
“Organizations like Amazon can’t be down at all,
so this is a technique that they use,” said
In addition to patching, organizations could also reuse
failover databases to take backup chores away from the primary database server,
Kaplan said. When a system is being backed up, it can slow service by 10
percent to 40 percent. Backing up the failover server — rather than the
primary server — removes any lag felt by users of the primary system.
The failover system could also be used for testing, Loaiza said. An organization may want to make a change in
the production environment but worries that the changes will cause havoc. The
change could be made to the failover server first and only applied to the
production database if the change is successful.
Report cards
Increasingly, though, organizations are generating
reports against these secondary databases.
In many cases, the data organizations’ need to report
exists only in the database in which it was created, Kaplan said. Yet such
databases are optimized for transactions, so running reports against these
databases slows the system for users. “Organizations are stuck in that
mixed, dual-purpose environment,” Kaplan said. “If you have
everything in one environment, it hurts performance.”
Because failover databases don’t have the performance
requirements that the primary ones do, they might offer quicker responses to
queries by business intelligence software. “You can do the heavy
reporting in one system and the updates on the original,” Kaplan said.
One thing to keep in mind with this approach is that the
reporting database is a little less current than the production database,
Kaplan said. Unless the two database servers are fully synchronous —
meaning the two are updated at the same time — the secondary server may
not be as complete as the primary one. “That might be OK if you’re
doing a report about what happened last month,” he said. In some cases,
however, it may not be acceptable to have differing versions of the data.
Joab Jackson is
chief technology editor for Government Computer News, an 1105 Government
Information Group publication.