EXASOL Shrugs Off Oracle’s In-Memory Speedup

Once dismissed as exotic toys, in-memory databases are gaining new ground in the data center. The latest in-memory believer is Oracle, which in July will ship the new in-memory option for its 12c database, which it claims will boost analytic workloads by 1,000x. That likely won’t be enough to match dedicated in-memory setups, such as EXASOL‘s.

Oracle is taking a page out of rival SAP’s HANA playbook, which says enterprises want a single hybrid database that can simultaneously process transaction and analytic workloads. It was a major turnabout for Oracle CEO Larry Ellison, who just a few years ago dismissed HANA as “wacko.”

Last week, Ellison talked up in-memory’s capability to run analytic workloads up to 1,000 times faster than before, while transactional workloads will see a more modest 2x speedup. No programming changes are required; DBAs can simply “flip a switch,” and 12c applications can immediately start using Oracle’s in-memory capabilities, the company says.

While the biggest relational database players have publicly found religion with in-memory databases, existing providers of in-memory tools quietly go about their business. Take EXASOL for example. The German company has been developing a column-oriented, in-memory analytics database since 2000, and today owns several TPC-H benchmarks that attest to its blazing speed.

This week, EXASOL unveiled the fifth generation of its technology, which promises another speedup via advanced caching and index creation techniques. “Surprise, surprise, we made it quicker,” says Graham Mossman, EXASOL’s senior solution engineer.

How much faster is new EXASolution database? That’s not precisely known yet. The company is currently working with Dell on new benchmark figures for its version 5 software, Mossman says.

But if the previous version of EXASOLultion is any indication, version 5 will be no slouch. Version 4 currently holds TPC-H records for throughput in several data warehousing categories, including the overall record of a whopping 7.1 million QphH (composite queries-per-hour) running on a Dell PowerEdge R710 cluster, which was recorded in the 3,000 GB database size range in October 2011.

No other databases have come anywhere close to that 7.1 million QphH figure in any configuration. The next-best figures are owned by a ParAccel database (now owned by Actian) running on an HP DL380 that achieved 1.3 million QphH and an Oracle database running on an HP BladeSystem c-class that achieved 1.1 million QphH; both figures are in 1,000 GB data warehouse category. In that category, EXASOL also clocked in at 4.2 QphH, 1.0 QphH, 1.6 million QphH, on different hardware. EXASOLution is also among the cheapest data warehouse platforms, coming in around $.10 to $.20 per QphH in several configurations, matched only by several Actian Vector (formerly VectorWise) setups.

The EXASolution database, which runs on the vendor’s proprietary EXACluster OS, is typically used as a data mart for analytic applications, but also occasionally in support of large data warehouses, according to Gartner’s latest Magic Quadrant on data warehouses, which listed EXASOL in the “niche players” segment. Mossman reports that the company doesn’t have any multi-hundred node deployments–usually the domain of data warehouse leaders like Teradata, Oracle, SAP, and IBM— but that it’s testing such a configuration with an American company.

EXASOL hasn’t made a big marketing splash to date, but there are indications the firm is gearing up for a big push with the version 5 software. Gartner lauds the company for its extremely fast performance, and the fact that no performance tuning is required, which keeps costs down. The company’s “positive product track record” so far hasn’t transferred into big growth numbers; it had just 45 customers, according to Gartner. A recently opened office in San Francisco could signal an increased focus on boosting U.S. sales.

Besides more speed, version 5 brings some other tricks, including a new Skyline function that aims to simplify some of the complexity that stem from traditional data mining approaches, where the “multiplicity of variables mean that sorting, filtering, and weighting lead to sub-optimal analyses.” Potential application areas of the Skyline functionality include segmentation, next best offer, fraud, and portfolio analysis, the firm says. It is also opening the database to including any statistical library packages written in R, Python, or Java.

On the marketing front, the attention that Oracle brought in-memory databases is good for everybody in the in-memory database business, including EXASOL, Mossman says. “We see this kind of movement from Oracle as being an indication that we’re on the right track all along,” he says. “The world is only getting less tolerant of waiting around for answer.”

But don’t be surprised if Oracle and its customers end up separating the analytic and OLTP databases. “It’s almost like two different worlds of physics going on there,” he says. “You have one way of organizing data for OTLP and one way for analytics. Why comprise? Why not get yourself a database for OLTP and one that’s even better for analytics?”

Related Items:

Hadoop and NoSQL Now Data Warehouse-Worthy: Gartner

NoSQL Databases RAM it Home with In-Memory Speedups

Taking NoSQL to the Next Level with In-Memory Computing