Exadata Smart Features on OCI

Exadata's value is concentrated in a handful of features that ordinary database platforms do not have, and understanding them is the key to deciding whether Exadata is worth its cost and to running it well once you have it. These are the smart features: Smart Scan, storage indexes, Hybrid Columnar Compression, the flash and persistent memory tiers, and Smart Flash Logging. This article explains what each does, which workloads exercise it, and why they collectively make Exadata a different kind of platform rather than just a faster server.

For the wider picture of when Exadata fits, see our complete guide to Exadata Cloud Service, and for how these features translate into measurable speed, Exadata Cloud Service performance.

Smart Scan: offloading work to storage

Smart Scan is the defining Exadata feature. In a conventional database, a query that scans a large table pulls every block of that table across the network to the database server, which then filters out the rows and columns it does not need. Exadata inverts this. Its intelligent storage servers run database aware software that performs the filtering and column projection at the storage layer, so only the relevant rows and columns travel back to the database nodes. On a query that selects a few columns from a tiny fraction of a huge table, this can reduce the data moved by an order of magnitude, and that reduction is where much of Exadata's speed comes from. Smart Scan applies to full scans of large objects, so it benefits analytic and reporting workloads enormously and transactional single row lookups not at all.

Storage indexes: skipping reads entirely

Storage indexes are an automatic, in memory structure on the storage servers that tracks the minimum and maximum values of columns within each region of storage. When a query filters on a column, the storage server can consult the storage index and skip entire regions that cannot possibly contain matching values, avoiding the read altogether. This is not an index you create or manage; it is maintained automatically and works in concert with Smart Scan to reduce IO. For well ordered data, storage indexes can eliminate a large fraction of the reads a query would otherwise perform, and because they require no administration, they deliver their benefit silently.

Hybrid Columnar Compression: smaller and faster

Hybrid Columnar Compression stores data in a format organised by column rather than by row, grouping similar values together so they compress far more effectively than row organised data. The benefit is twofold. The footprint shrinks, often dramatically for analytic and archival data, which reduces storage cost and the volume of data to scan. And because the data is smaller and column organised, the scans that Smart Scan performs run faster. The compression has different levels suited to different access patterns, with more aggressive compression for data that is queried but rarely updated. Applied deliberately to the right data, it is one of the most effective levers for both performance and cost on Exadata.

The flash and persistent memory tiers

Exadata places flash and persistent memory in front of disk so that hot data is served from the fastest possible medium. The flash cache holds frequently accessed blocks, and persistent memory on current generations provides an even faster tier for the hottest data. For transactional workloads with high concurrency, this is what keeps latency flat as load climbs, because the data those workloads touch repeatedly is served from memory speed storage rather than disk. Smart Flash Logging accelerates the redo log writes that every transaction depends on, smoothing the write path that can otherwise become a bottleneck for write intensive systems. Together these tiers are why Exadata suits demanding transactional cores, not just analytic warehouses.

Why the features matter for the buying decision

The reason to understand these features before buying Exadata is that they define exactly which workloads the platform serves well. A workload dominated by large scans, heavy analytics, or high concurrency transactions with hot data will exercise these features and get performance that ordinary infrastructure cannot match. A workload of small, scattered single row lookups on tiny tables will exercise almost none of them, and for that workload Exadata is an expensive way to run something a modest virtual machine would handle. The honest buying question is not whether Exadata is powerful, because it plainly is, but whether your workload triggers the features that justify the cost. We frame this decision in the complete guide and the consolidation case in consolidating databases on ExaCS.

Making sure your workload uses them

Owning Exadata is not the same as benefiting from it, because the features deliver only to workloads configured to trigger them. Stale optimiser statistics can lead the database to execution plans that bypass Smart Scan. Tables left uncompressed miss the footprint and scan benefits. SQL written in ways that defeat offload runs without it. The work of making sure a workload actually exercises the smart features, by keeping statistics current, applying compression deliberately, and tuning the queries that should offload, is what turns a capable platform into a fast one. This is covered practically in Exadata Cloud Service performance, and it is the kind of ongoing tuning that our managed services practice provides.

If you are evaluating Exadata and want to know whether your workload would actually exploit these features, or you suspect an existing Exadata is not using them, an assessment that profiles the workload against the platform's capabilities is the right place to start. The features are what you pay for, so it is worth confirming you are using them.