> For the complete documentation index, see [llms.txt](https://wiki.datadike.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://wiki.datadike.com/pam/deploy-operate/requirements.md).

# System requirements

Deployment models, machine sizing, storage and platform dependencies for a DataDike PAM instance, and the Windows host that runs published applications and web sessions.

## What this page covers

What one instance of the solution needs in order to run: where it can be deployed, how large the machine has to be, how much disk the recordings will consume, and which platform services it depends on.

For the connectivity it needs, see [Network and connectivity requirements](https://wiki.datadike.com/pam/deploy-operate/network-requirements.md). For the ports used to reach *assets*, see [Ports and protocols](https://wiki.datadike.com/pam/reference/ports.md).

{% hint style="warning" %}
The sizing figures below are the ones published in DataDike's datasheet. They are a starting point, not a capacity model: storage in particular depends on your own retention policy and session mix. Confirm the sizing for your deployment with DataDike before you commit hardware, and confirm it against the version you are actually deploying.
{% endhint %}

## Deployment models

The solution runs on standard **x86\_64** infrastructure, on physical servers or virtual machines.

| Model             | What it means                                     |
| ----------------- | ------------------------------------------------- |
| **On-premises**   | Deployed in the customer's own data centre        |
| **Private cloud** | VMware, Hyper-V, KVM or OpenStack                 |
| **Public cloud**  | AWS, Azure, Google Cloud or Oracle Cloud          |
| **Hybrid**        | Central instance on-premises, connectors in cloud |

The DataDike OS documentation adds **Nutanix** and **KVM/Proxmox** to the list of hypervisors, and recommends installing directly on bare metal, without virtualisation, for best performance.

## Machine sizing

### Minimum — labs, testing and proof of concept

| Resource         | Minimum                                                   |
| ---------------- | --------------------------------------------------------- |
| CPU              | 4 vCPU                                                    |
| Memory           | 8 GB RAM                                                  |
| Storage          | 100 GB SSD                                                |
| Network          | 1 Gbps                                                    |
| Architecture     | x86\_64                                                   |
| Operating system | Oracle Linux, Ubuntu Server, Red Hat, Debian, DataDike OS |

### Recommended for production

| Resource  | Recommended   |
| --------- | ------------- |
| CPU       | 8 vCPU        |
| Memory    | 8 – 16 GB RAM |
| Storage   | 500 GB SSD    |
| Disk type | SSD or NVMe   |
| Network   | 1 Gbps        |

### What drives those numbers up

Storage, above all. The datasheet names four factors:

* the retention period for **session recordings**;
* the volume of **audit logs**;
* the number of **managed assets**;
* the number of **concurrent sessions**.

## Storage for session recordings

Recordings are compressed and optimised before they are stored, in local storage and in external object storage alike. The datasheet publishes this average:

| Sessions per day | Storage per month |
| ---------------- | ----------------- |
| 100              | \~50 GB           |
| 500              | \~250 GB          |
| 1000             | \~500 GB          |

{% hint style="warning" %}
The source does not state the session length, protocol mix or compression setting behind those figures, and graphical sessions (RDP and VNC) consume far more space than terminal sessions. Treat the table as an order of magnitude only, and measure your own consumption over the first weeks of operation.
{% endhint %}

Two operational points that follow from it:

* The recording backend installed by default (**default**, type **Server**) is local disk on the node. It has no redundancy, it consumes the node's own disk, and it does not scale — move to external object storage for production.
* The retention setting in the console controls the **metadata in the database**, not the files. The real lifecycle of recording files is configured in the bucket or container that holds them.

## Platform dependencies

| Dependency                    | What the sources say                                                                                                                                               |
| ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Database                      | Integrated into the appliance, internal to the solution. Managed database services (Amazon RDS, Cloud SQL) appear in the datasheet as optional integrable services |
| Redis                         | Present in a deployed instance — the health endpoint reports its status and response time alongside the database's                                                 |
| Object storage for recordings | S3, Ceph, Swift, OSS, Azure, OBS, COS, SFTP, local (**Server**) or discard (**Null**)                                                                              |
| Command storage               | Stored in the solution's own database by default; an external index such as Elasticsearch can be configured instead                                                |
| Time source                   | NTP, or the host operating system's own date and time service                                                                                                      |

{% hint style="danger" %}
The clock is not a detail. TOTP codes are validated to the second, so a node whose clock has drifted rejects every one-time password with *OTP code invalid, or server time error* — an outage that looks like an authentication bug. Synchronise the clock and set the correct timezone before you hand the instance over.
{% endhint %}

## Browsers

Console access is supported on Mozilla Firefox, Google Chrome, Microsoft Edge, Opera and Safari.

## Windows publishing host

Published applications (RemoteApp) and web sessions run on a separate Windows machine, never on the PAM server itself. Plan it as part of the deployment when either feature is in scope.

| Requirement             | Detail                                                                                                                                                                                                                                                                                                                                                                                    |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Operating system        | Windows Server. Web sessions require Windows Server 2022 or 2025                                                                                                                                                                                                                                                                                                                          |
| Remote Desktop Services | The Remote Desktop Session Host role, installed by the solution when the host is deployed, and licensed by Microsoft through RDS CALs and a Remote Desktop license server. Without a license server, Windows allows a 120-day grace period, after which the host refuses sessions. See [Remote Desktop Services licensing](https://wiki.datadike.com/pam/deploy-operate/rds-licensing.md) |
| OpenSSH                 | Installed and running                                                                                                                                                                                                                                                                                                                                                                     |
| Network                 | A route from the host to the console address, which the host calls during deployment and while it runs, and to the applications its sessions open                                                                                                                                                                                                                                         |

## Public cloud instance examples

These are the example shapes the datasheet publishes for each provider. They are examples, not a support matrix.

|                | AWS                           | Azure               | GCP                            | OCI               |
| -------------- | ----------------------------- | ------------------- | ------------------------------ | ----------------- |
| Instance       | EC2 t3.xlarge or m6i.xlarge   | D4s\_v5 or D8s\_v5  | n2-standard-4 or n2-standard-8 | VM.Standard3.Flex |
| CPU            | 4–8 vCPU                      | 4–8 vCPU            | 4–8 vCPU                       | 2–4 OCPU          |
| Memory         | 8–16 GB                       | 8–16 GB             | 8–16 GB                        | 8–16 GB           |
| Storage        | EBS SSD (gp3)                 | Premium SSD         | Persistent SSD                 | Block Volume SSD  |
| Load balancing | AWS Application Load Balancer | Azure Load Balancer | Google Cloud Load Balancer     | OCI Load Balancer |

Services the datasheet lists as optionally integrable, per provider:

| Provider | Services                                            |
| -------- | --------------------------------------------------- |
| AWS      | Amazon RDS, Amazon S3, AWS IAM, AWS Secrets Manager |
| Azure    | Azure AD, Azure Key Vault, Azure Storage            |
| GCP      | Google IAM, Cloud Storage, Cloud SQL                |
| OCI      | OCI Object Storage, OCI Vault                       |

## Scaling and high availability

The solution scales horizontally. The datasheet names four strategies: multiple nodes, distributed session proxies, an external database, and dedicated storage for recordings.

High availability is offered as **active-active**, **active-passive** or a **DR replica** cluster: several nodes behind a virtual address (keepalived), with data replicated between the nodes. Access survives the loss of a node, and the session load is distributed or redistributed across the remaining ones. [Availability, SLA and updates](https://wiki.datadike.com/pam/deploy-operate/availability-and-sla.md) states design objectives of **RTO ≤ 1 hour** and **RPO ≤ 15 minutes** for the active/active architecture, and redundancy of the vault, session-brokering and administration console components so that none of them is a single point of failure.

See [Install a high-availability cluster](https://wiki.datadike.com/pam/deploy-operate/install-cluster.md) for what a cluster deployment involves.

## How to check a running instance against these requirements

* Poll the unauthenticated health endpoint `/api/health/`. It returns JSON with `status`, `db_status`, `db_time`, `redis_status`, `redis_time` and `time` — which proves the database and Redis are reachable, not merely that the web server answered. This is the right thing for an external monitor to watch.
* Open **System settings › Components** and read the **Stat** column, which shows CPU and memory use per component, and **Load status**, which reads **Normal**, **High load** or **Offline**. Components sitting at **High load** are the signal to add capacity, either by growing the node or by adding instances of that component type.
* Check the recording backend before disk pressure finds you: an instance left on the local **default** backend fills the node's own disk.

## Related pages

* [Network and connectivity requirements](https://wiki.datadike.com/pam/deploy-operate/network-requirements.md)
* [Install a single node](https://wiki.datadike.com/pam/deploy-operate/install-single-node.md)
* [Install a high-availability cluster](https://wiki.datadike.com/pam/deploy-operate/install-cluster.md)
* [Ports and protocols](https://wiki.datadike.com/pam/reference/ports.md)
* [Core concepts](https://wiki.datadike.com/pam/get-started/core-concepts.md)
