---
title: "Scaling reads"
url: "https://docs.yugabyte.com/stable/explore/linear-scalability/scaling-reads/"
---

# Scaling reads

Scaling reads in YugabyteDB.

Read performance when scaling horizontally

Reads scale linearly in YugabyteDB as more nodes are added to the cluster. The read I/O path is very straightforward. As the mapping of a row to a tablet is deterministic, the system knows exactly where to look for a row. Let's go over how reads work and see how well they scale.

## How reads work

When an application connected to a node sends a read request for a key, say `SELECT * FROM T WHERE K=5`, YugabyteDB first identifies the location of the tablet leader containing the row with the key specified (`K=5`). After the location of the tablet leader is identified, the request is internally re-directed to the node containing the tablet leader for the requested key. The leader has the latest data and responds immediately.

A basic `select` statement involves a maximum of just 2 nodes. This redirection is completely transparent to the application.

In the following illustration, you can see that the application sent the request for `K=5` to `NODE-2`. The system identified that the key `K=5` is located in `NODE-1` and internally redirected the request to that node.

![How does a read work](/images/explore/scalability/scaling-reads-redirection.png "How does a read work")

Multiple applications can connect to any node and the reads will be redirected correctly.

## Sysbench workload

The following chart shows how reads scale horizontally in YugabyteDB using a [Sysbench](/stable/benchmark/sysbench-ysql/ "Sysbench") workload of basic selects. The cluster consisted of [m6i.4xlarge](https://www.google.com/search?btnI=1&q=aws%20m6i.4xlarge#:~:text=m6i.4xlarge) instances and had 1024 connections. All requests had a latency of less than 3ms.

![Scaling with Sysbench](/images/explore/scalability/scaling-reads-sysbench.png "Scaling with Sysbench")

You can clearly see that with an increase in the number of nodes, the number of reads scales linearly.

## Learn more

- [Sysbench benchmark](/stable/benchmark/sysbench-ysql/ "Sysbench benchmark")
- [YugabyteDB Benchmarks](/stable/benchmark "YugabyteDB Benchmarks")
- [Scaling: YugabyteDB vs Cockroach vs Aurora](https://www.yugabyte.com/blog/yugabytedb-vs-cockroachdb-vs-aurora/ "Scaling: YugabyteDB vs Cockroach vs Aurora")