Analytics Legends The knowledge platform for SAP Analytics
Guide

SAP Analytics Cloud live connection: the default and its exceptions

As of 2026-08-14

A live connection queries the source on every render — nothing is copied into SAC, row-level access and lineage flow through automatically, and there is one copy of the truth. That is why the senior default in 2026 is live to a Datasphere Analytic Model, with anything else documented in an architecture decision record. The exceptions are real, but they are exceptions with reasons, not equal alternatives.

What live inherits, and what import loses

On a live connection each render pushes queries down to the underlying Analytic Model, calculation view or InfoProvider. Data access controls flow through automatically, so users see only what their permissions allow, and lineage is tracked as a dependency in the catalogue. Latency is the model's query time plus the network, typically in the low hundreds of milliseconds at the median.

An imported model copies data into SAC's own store on a schedule. Rendering is fast because the data is local, usually under a hundred milliseconds, and the ceiling rises into the thousands of concurrent users. What is lost is not obvious from the interface: data access controls do not inherit, so every user with story access sees every imported row, and lineage breaks at the import boundary.

The concurrency ceiling nobody plans for

Live mode has a practical ceiling of roughly two hundred simultaneous ad-hoc users per Analytic Model before tail latency degrades. That is a design input, not a runtime alarm — nothing warns you as you approach it.

The failure pattern is consistent. A team sizes for a pilot's ten users, picks live because it feels more correct, and never revisits the decision as the rollout reaches the hundreds. The symptom appears as a slow dashboard and gets diagnosed as a Datasphere performance problem, when the root cause was one connection-mode decision taken months earlier for a different audience.

Above that threshold the honest options are import for the stable slice or a hybrid — live for volatile data, import for the historical baseline, joined at story level. Hybrid is the default shape for wide enterprise rollouts that still need freshness, and it is a design choice rather than a fallback.

Decision table — live, import, or hybrid

New story on a governed Datasphere model → live → the 2026 default; no exception without a written reason.

Governed reporting in a regulated domain → live, without exception → imported flat data has no row-level inheritance and no lineage, which is a compliance fact rather than a preference.

Executive dashboard read weekly by a large audience → import or hybrid → the audience values instant load over second-by-second freshness, and the staleness window is invisible to them.

More than a couple of hundred concurrent ad-hoc users with a freshness requirement → hybrid → live for the volatile slice, import for the stable history.

Federation and the one exception worth writing down

Where the source lives on-premise rather than in Datasphere, live connections route through the Cloud Connector — a reverse-proxy tunnel authenticating against the on-premise system without exposing it to the internet. Federated live queries carry the normal live latency profile plus the tunnel's own queuing under concurrent load: a story with dozens of simultaneous viewers can queue before the query reaches the source at all.

This is where a senior architect deliberately trades governance purity for a snapshot. A heavily used executive dashboard against an on-premise table, refreshed nightly, beats a live federation that times out under Monday-morning load, and pretending otherwise is dogma rather than architecture.

What we cannot assert

The concurrency and latency figures are working design budgets recorded in our corpus rather than published SAP limits, and they depend on model complexity and query shape. Use them as the point at which to re-open the connection decision, not as a guaranteed threshold.

Frequently asked

What is the difference between live and import connections in SAC?

A live connection queries the source on every render and copies nothing, so row-level access and lineage flow through automatically. An import connection copies data into SAC on a schedule, which is faster to interact with and scales further, but access controls do not inherit and lineage breaks at the import boundary.

How many concurrent users can a live connection support?

Roughly two hundred simultaneous ad-hoc users per Analytic Model before tail latency degrades. Treat it as a design input rather than a limit the product enforces: nothing warns you as a rollout approaches it, and the symptom appears later as a dashboard blamed on the platform.

Is import mode ever the right choice?

Yes, in three cases: a wide audience valuing instant load over freshness, a one-off exploration on non-governed data, and an on-premise source behind a Cloud Connector where a live federation would queue under load. Record each with its reasoning, because the reasoning tells a successor when the exception stops applying.

What this page is built on

Read next