This article investigates how the newly-available In-memory Database (IMDB) and Relaxed Durability Database (RDDB) in Sybase ASE 15.5 offer new techniques for supporting performance-intensive applications while not mandating any deviations from traditional Sybase development practices.
Introduction
With the release of version 15.5 of its Adaptive Server
Enterprise (ASE) database platform, Sybase offers new database technology for
database architects, application developers, and system administrators. In this
article, we investigate how the newly-available In-memory Database (IMDB) and Relaxed
Durability Database (RDDB) offer new techniques for supporting
performance-intensive applications while not mandating any deviations from
traditional Sybase development practices.
We begin by examining the special performance requirements of
high-volume modern database-driven applications, and how memory-based data
management is an innovative approach that helps deliver the throughput
necessary to support these mandates. Next, we delve into the new IMDB and RDDB
options offered by ASE 15.5, including a brief overview of how they work. Once that
topic has been covered, we move on to a series of inter-related scenarios that chronicle
cases where each option is appropriate.
How in-memory databases help modern applications
Given their role as foundational technology for
mission-critical applications, traditional relational database engines are
designed to support the well-known ACID principles (Atomicity, Consistency,
Isolation, and Durability). In particular, they typically implement the
Durability principle by using logs that are written to disk whenever a
transaction is committed by a user or application. Combined with data page
writes, this approach delivers full durability: transactions that have been
committed will be permanently registered on physical media, even in the event
of a system failure. This architecture has delivered decades of success for
relational databases.
However, this protection comes with a rather high price tag,
namely the expensive overhead incurred by all of the disk activity necessary to
preserve data integrity. Since disk operations are typically 10 times slower
than memory-resident activities, these costs can quickly add up.
While many applications require this level of insurance,
there are numerous situations where streamlined coverage will suffice. For
example, increasing numbers of applications supporting industries as diverse as
finance, government, and telecom all generate tremendous amounts of relatively
short-lived data. Although each of these situations is different, they all
require higher speed than often can be provided by traditional disk-based data
persistence.
In response to these trends, ASE 15.5 offers new data processing
options that operate entirely in memory, thereby bypassing the
durability-focused burden found in conventional disk-based data management. In
addition, Sybase has made it easy for developers and administrators to
seamlessly take advantage of these new alternatives without needing to rewrite
applications or make any other changes.
Understanding new database options
Before we look at some scenarios where in-memory and relaxed durability
databases can help, lets briefly explore their architecture as well as how to work
with them.
In-memory database (IMDB)
As we describe earlier, traditional disk-based relational
databases continually synchronize their memory (data and transaction logs) with
disk drives. However, in the case of an in-memory database such as offered by
ASE 15.5, neither log nor data is ever written to disk. Instead, both remain in
memory at all times. Transaction logging is still performed, but the log
remains in memory. As long as no unexpected events (such as a server, database,
or application failure) occur, the information contained in this database will
remain safeguarded and usable. Furthermore, you’re always free to copy
information from the in-memory database onto a disk-based database, thus
preserving your information beyond the lifespan of a single server instance.
To create an ASE in-memory database, you first run a series
of stored procedures that allocate an in-memory storage cache and generate a
logical disk device within this cache. Once that’s been done, you create the
database using the traditional ASE syntax, with an added directive to use the
in-memory option. From this point onward, the in-memory database appears (to
users and applications alike) just like a traditional database, except that it delivers
better performance and diminished durability.
Once you shut down ASE (or an unplanned event occurs), the
IMDB’s contents are lost. However, note that an empty copy of the database will
be re-created upon server re-boot. In fact, you can even create a template as a
boot-time blueprint. This template can contain metadata (such as table
definitions), data, users, and permissions. This is a great way to bootstrap
your IMDB and eliminate manual preparation.
Relaxed durability database (RDDB)
As we described earlier, traditional disk-based databases
provide durability by regular flushing of data pages and the transaction log to
disk. To deliver their enhanced performance, in-memory databases skip these
time-consuming, expensive responsibilities. In an effort to provide a degree of
compromise between these two extremes, ASE 15.5 offers another option: relaxed
durability databases.
In particular, there are two options at your disposal. The
first option delivers similar durability as an in-memory database: when the
database server shuts down (whether normally or via an unplanned event such as
a crash) your in-memory data is lost. On the other hand, selecting the second
option provides enhanced durability if you perform a proper, normal database
server shutdown. However, note that if an unplanned shutdown occurs, the
data will still be lost.
Relaxed durability databases are fully integrated with ASE,
just like in-memory databases. This provides the same look-and-feel, user
experience, and language compatibility.
Database usage scenarios
Now that we’ve outlined the new choices offered in ASE 15.5,
let’s see how these options might be used in support of a series of real-world scenarios.
For the purposes of this illustration, imagine that you’ve been asked to create
a database architecture for a new e-commerce web site. Your design will need to
support a collection of important capabilities, including:
- An online shopping cart
- Customer feedback and comments
- Order and inventory management
To support these requirements, we’ll look at three separate use
cases. As you’ll see, each situation has its own unique necessities that can be
addressed by ASE 15.5.
Requirement 1: An online shopping cart
As you begin your design efforts, you face the sobering fact
that your new e-commerce site will confront a formidable collection of
competitors. To make matters worse, you’re well aware that your customers are a
highly demanding, fickle lot; they expect very fast responsiveness, and won’t
hesitate to abandon a shopping cart and go elsewhere if your site isn’t
available, even for a moment. You also know that once you’ve lost a customer,
they’re unlikely to ever return.
Thus, to help support this essential responsibility and
deliver an optimal customer experience, your job is to apply the right mixture
of hardware and software technologies. From the hardware perspective, you
employ a collection of dedicated, powerful servers in a highly controlled
environment. To minimize the possibility of site outages, you employ
sophisticated server failover capabilities, along with network redundancy and
other techniques to ensure availability.
Given that your processing environment is so robust, and that
server outages are highly unlikely, the optimal information management software
choice would be to select the ASE in-memory database option. By avoiding disk
interaction whenever possible, it will deliver the fastest possible database
performance. In addition, by creating a boot-time template with pre-defined
tables and data, you minimize the amount of preparatory work necessary to stage
your site to accept orders.
Requirement 2: Customer feedback and comments
Product reviews, experiences, and other customer-driven
commentary are essential components of any modern e-commerce shopping
experience. To support this key capability, you plan to create a
well-integrated forum where customers can share their thoughts on your
products. Since this intelligence will be a major component of the buying
decision, you need to ensure that both performance and data integrity are present
in your solution: the data must be available quickly, and newly created
comments need to be safely stored on your servers.
By analyzing the various ASE data management options, you
decide that an in-memory approach will deliver the performance you need to
complete these tasks within the specified response window. However, you’re
concerned about a potential loss of open comments should a planned server
shutdown occur in the middle of the editing process. While the likelihood of a
server shutdown affecting many comments is low, you still want to avoid the
prospect of a customer carefully entering a review only to see it vanish into
the ether. This argues against the pure in-memory database technique, since
this option provides no durability in this type of situation.
In this case, the relaxed durability option (specified when
creating the database) offers a suitable compromise: it delivers the enhanced
performance of an in-memory database while preserving data durability if a
planned shutdown takes place while processing is underway. Note that the
relaxed durability option does not provide any durability protection if an unplanned
server shutdown or application failure occurs.
Requirement 3: Order and inventory management
The amount of time it takes to receive their order, and the
accuracy of the shipment, will form the lasting impression that a customer will
take away from your site. Thus, an always-on, lightning-fast website is of
little use if the order fulfillment process is subpar. However, since budgets
are constrained in every organization, it’s quite likely that the fastest and
most available servers will be dedicated to the shopping cart and customer
forum, leaving somewhat less-capable hardware to support the back office. While
outages and unplanned downtime are still relatively unlikely, they are more
prone to occur in this type of environment.
Since data durability outweighs performance requirements for
this type of processing, this is a situation where traditional disk-based persistence
is the optimal choice. As an added benefit, if additional resources become
available to support the order fulfillment process, you’re free to revisit your
decision: switching to an in-memory database won’t require any application
changes, which illustrates the beauty of Sybase’s holistic architectural
philosophy.
Conclusion
The new database options provided in ASE 15.5 are worth a
look, especially if you’re charged with supporting high-volume applications,
situations that don’t require perfect data durability, or are lucky enough to
have highly reliable hardware at your disposal.
What’s especially intriguing about these fresh innovations is
that they can be used transparently; you don’t need to re-code your applications
or make any other time-consuming alterations to take advantage of these new alternatives.
Finally, it’s important to apply these new selections
properly: not every situation requires an in-memory or relaxed durability
database; your data integrity requirements may mandate a traditional database
instead. However, for those occasions where they make sense, they offer an
ideal package of performance and flexibility.
Additional Resources
Sybase Getting Started with In-Memory Databases in Adaptive Server® Enterprise 15.5
Overview of Sybase ASE In-Memory Database Feature
Sybase Sybase ASE In-Memory Databases
About the author
Robert Schneider is a Silicon Valley-based technology
consultant. He has written six books and numerous articles on advanced
technical topics such as Cloud Computing, Service Oriented Architecture (SOA),
and high-performance relational database design and optimization. He may be
reached at robert.Schneider@think88.com
