← Back to the journal

DataFocus · Journal

Sizing Signals: Does Your Workload Need OT Flash or OT Heavy?

DataFocus publishes a reference node and a license policy but no sizing table. This checklist turns those documented facts into measurable signals you collect yourself before choosing Flash or Heavy.

You cannot decide between a lightweight and a distributed DataFocus deployment from a vendor sizing table, because sizing depends on your workload. The download page says "Plan for your data scale and workload", and system requirements and supported platforms are listed in each edition's release notes, published with the installers on the download page and in the OT Flash and OT Heavy GitHub repositories. So the estimate is yours to make: measure your data volume (with replication), your concurrent users, how often data must refresh, and whether you need ETL or already run Hadoop. If those fit comfortably on one node and you need no pipelines, start with OT Flash. If they do not, or you already operate a Hadoop-class platform, look at OT Heavy.

What does DataFocus actually document?

Three public facts bear on sizing. Everything else in this post is a method for gathering your own numbers.

Documented factWhereWhat it does not say
Resource planning guidance reads "Plan for your data scale and workload"Download pageEdition-specific requirements, which are in each edition's release notes
Community Edition runs on a single node with unlimited users and no license fee; Enterprise subscriptions add multi-node and cluster deploymentDownload pageThe fee for Enterprise, and any capacity numbers
Policy DATAFOCUS_FREE_1DNU_GROWTH_V1: 1 active DataFocus Node Unit per node under a local hard limitreleases.jsonWhat a DNU measures beyond "per node"

For both editions, system requirements and supported platforms are listed in the release notes. The editions page tells you to take license, service scope and node definitions from the documentation included with the edition and to plan around data volume, concurrency and model requirements.

Why a single node is the starting assumption for Flash

OT Flash uses Apache Doris as both the in-memory compute engine and the data warehouse, and DataFocus positions it for teams that want one business domain running quickly with data that can sit in a single analytical engine. The Community Edition license is a single-node license. Together with the 1-DNU-per-node policy, the documented free path is one engine on one node.

Read the stored size carefully: compression, replication, document indexes and snapshots all change how much source data fits on a node, so do not treat any disk size as a data limit.

OT Heavy, by contrast, is described as continuing the Hadoop, Hive, Trino and Iceberg architecture with ETL and AutoML built in, and is suited to enterprises with large data volumes, an existing big-data platform and a need for pipelines and modelling.

Which signals should you measure yourself?

Collect these before you pick an edition. None of them comes with a DataFocus threshold, so the decision rule is "does it fit one node with headroom, as shown by a pilot".

  1. Stored size, not raw size. Count the tables you would import for the first business domain, then add the documents you would index. A hypothetical example, illustrative and not run: a domain with 400 GB of source tables and 50 GB of documents is 450 GB before any engine overhead. You would then need to test compression on a sample, and check that figure against the disk you plan to provision, before concluding anything.
  2. Replication factor. Distributed storage multiplies stored bytes. Apache's HDFS design page says the replication factor "can be specified at file creation time and can be changed later" and refers to "the common case, when the replication factor is three" (Apache Hadoop, HDFS Architecture). The Apache Doris overview likewise describes data partitioned into shards stored with multiple replicas across backend nodes (Apache Doris, What is Apache Doris). Ask how many replicas your target layout will keep.
  3. Growth rate. The policy name contains "GROWTH", but the sources do not describe a growth path from one node to a cluster. Estimate monthly growth yourself and decide in advance at what size you would move.
  4. Concurrent human users. The policy allows unlimited human users. It also specifies a one-time, non-blocking prompt to the system administrator when concurrent human users exceed 3 and stay idle for 900 seconds. The documents do not say what that prompt is meant to indicate, so log your own peak concurrency from the pilot instead of guessing.
  5. Agent and model traffic. The product uses a large language model service configured by the administrator. Count the questions per day you expect and where the model runs; the model is a cost and capacity line separate from the data node.
  6. Refresh frequency. Plan how data reaches Doris and how often it refreshes, as the edition notes advise. A nightly CSV import and a continuous sync have very different load profiles.
  7. Pipeline need. ETL and AutoML are listed for Heavy; the product brief's edition table shows "—" for Flash. If you need DataSpring ETL, that settles the edition question before any capacity arithmetic.
  8. Existing platform. If you already run Hadoop, Hive, Trino or Iceberg and have operators for them, Heavy builds on skills you hold. If you do not, a Hadoop-class cluster is a new operations burden.

A decision rule that does not invent thresholds

Use the signals in this order.

  1. Need ETL or AutoML, or already run a Hadoop-class platform you want to analyse in place? Look at Heavy.
  2. Otherwise, size the stored data with your planned replication and index overhead. Does it fit one node with room to grow?
  3. Does the pilot show acceptable response times for your real questions at your real peak concurrency? Record your own measurements; this post has none.
  4. If the answer to 2 or 3 is no, either provision a larger single node within the license terms, or move to a subscription that covers multi-node deployment.

Step 4 has a licensing side. The Community Edition license "does not cover servers, model calls or operations", so infrastructure and operations cost is yours in both editions. The Community Edition is single-node; multi-node and cluster deployment come with Enterprise subscriptions, and DataFocus does not publish their price.

When this checklist is not enough

  • Getting the editions. Both editions are distributed free from the download page and the OT Flash and OT Heavy GitHub repositories, which also publish the release notes with system requirements. Pilot either edition with your own data before you commit.
  • Unpublished numbers. Latency, throughput and concurrency limits are not published. Any figure you read elsewhere should be treated as unverified until you reproduce it.
  • Workload shape. A few heavy analysts and many light dashboard viewers load a system differently. The signals above do not capture that; a pilot does.

Next step

Read the edition comparison and download page and the deployment documentation for the preparation checklist (usage scope, OS and architecture, capacity, data authorization, model choice, data egress and backup ownership), then run a pilot on a single node with your own data.