---
updatedAt: 2026-08-13T09:13:18.000Z
agentTools:
  siteIndex: https://techdocs.akamai.com/llms.txt
  projectIndex: https://techdocs.akamai.com/cloud-computing/llms.txt
---

# Choose a data center

> 📘
>
> Our **Chennai** and **Mumbai** core sites are optimized for the India market. For workloads that currently serve or intend to serve the wider Asia-Pacific region and for communication with ​Akamai​ core sites outside of India, we recommend deploying resources in our Singapore or Sydney data centers.

Deploying your Linode to a geographically advantageous region can make a big difference in connection speeds to your server. Ideally, your site or application should be served from multiple points around the world, with requests sent to the appropriate region based on client geolocation. On a smaller scale, deploying a Linode in the region nearest to you will make it easier to work with than deploying to a region in a different geographic area or continent.

There are many things that can affect network congestion, connection speeds, and throughput, so you should never interpret one reading as the sole data point. Always perform tests in multiples of three or five for an average, and on both weekends and weekdays for the most accurate information.

This page is a quick guide for choosing and speed testing a data center. Start by creating a Linode in the data center in or near your region, or several Linodes in multiple regions if you're close to more than one. From there, use the [Facilities Speedtest](https://www.akamai.com/cloud/speed-test) page for test domains to ping and files to download.

> 📘 Distributed compute regions (limited availability)
>
> Distributed compute regions support a more limited set of cloud computing services than core compute regions, but may offer greater proximity to your users. If you have a workload that requires very low latency but your users are far from core regions, consider deploying Linodes to [distributed compute regions](https://techdocs.akamai.com/cloud-computing/docs/distributed-compute-regions).

# Product availability

An important consideration when choosing a data center is the availability of specific plans, features, and services. Conversely, if you are interested in a particular plan or service, it's important to identify the closest data center where that plan or service is available.

Search by region name, filter by feature and plan, or expand a geographic area to browse locations. Click any feature or plan name to view more details. For distributed compute regions, see the [list of features and services](https://techdocs.akamai.com/cloud-computing/docs/supported-services-distributed).

<RegionMatrix />

# Network latency

The Linux [ping](https://linux.die.net/man/8/ping) tool sends IPv4 [ICMP echo requests](https://en.wikipedia.org/wiki/Ping_\(networking_utility\)#Echo_request) to a specified IP address or hostname. Pinging a server is often used to check whether the server is up and/or responding to ICMP. Because `ping` commands also return the time it takes a request's packet to reach the server, `ping` is commonly used to measure network [latency](https://en.wikipedia.org/wiki/Network_delay).

Ping a data center to test your connection's latency to that DC:

```
ping -c 5 speedtest.dallas.linode.com
```

Use [ping6](https://linux.die.net/man/8/ping6) for IPv6:

```
ping6 -c 5 speedtest.dallas.linode.com
```

> 📘
>
> Many internet connections still don't support IPv6 so don't be alarmed if `ping6` commands don't work *to* your Linode from your local machine. They will work *from* your Linode to other IPv6-capable network connections (ex. between two Linodes in different data centers).

# Download speed

Download speed will be limited most heavily first by your internet service plan speed, and second from local congestion between you and your internet service provider. For example, if your plan is capped at 60 Mbps, you won't be able to download much faster than that from any server on the internet. There are multiple terminologies to discuss download speeds with so here are a few pointers to avoid confusion:

* Residential internet connection packages are sold in speeds of mega**bits** per second (abbreviated as Mbps, Mb/s, or Mbit/s).

* One mega**bit** per second (1 Mbps or 1 Mb/s) is 0.125 mega**bytes** per second (0.125 MB/s). Desktop applications (ex: web browsers, FTP managers, Torrent clients) often display download speeds in MB/s.

* **Mebibytes** per second is also sometimes used (MiB/s). One Mbps is also equal to 0.1192 MiB/s.

To test the download speed from your data center of choice, use the `cURL` or `wget` to download the `bin` file from a data center of your choice.

For example:

```
curl -O http://speedtest.dallas.linode.com/100MB-dallas.bin
wget http://speedtest.dallas.linode.com/100MB-dallas.bin
```

Below you can see that each time `cURL` is run, a different average download speed is reported and each takes a slightly different amount of time to complete. This is to be expected, and you should analyze multiple data sets to get a real feel for how fast a certain DC will behave for you.

```text Output
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  100M  100  100M    0     0  11.4M      0  0:00:08  0:00:08 --:--:-- 12.0M
```

```text Output
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  100M  100  100M    0     0  10.8M      0  0:00:09  0:00:09 --:--:--  9.9M
```

```text Output
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  100M  100  100M    0     0  9189k      0  0:00:11  0:00:11 --:--:-- 10.0M
```

# Pricing

In general, plans and services are billed at the same flat rate across all regions. However, due to higher infrastructure costs in various global markets, pricing for some services varies from region to region. Regions with data center-specific pricing include:

* Jakarta, Indonesia
* São Paulo, Brazil

Pricing is also different in [distributed compute regions](https://techdocs.akamai.com/cloud-computing/docs/distributed-compute-regions).

The following regions have higher rates for RTX PRO 6000 Blackwell instances:

* All European data centers, located at: Frankfurt DE, London UK, Milan IT, Paris FR, Amsterdam NL, Stockholm SE, Madrid ES
* Singapore
* Jakarta

See our [Pricing](https://www.linode.com/pricing/) page for a complete list of plans and pricing.

# Sibling pages

* [Network transfer usage and costs](https://techdocs.akamai.com/cloud-computing/docs/network-transfer-usage-and-costs.md)
* [Help & support](https://techdocs.akamai.com/cloud-computing/docs/help-and-support.md)
* [Send email on Akamai Cloud](https://techdocs.akamai.com/cloud-computing/docs/send-email.md)
* [Subscribe to status updates](https://techdocs.akamai.com/cloud-computing/docs/subscribe-to-status-updates.md)
* [Grant a developer access to your services](https://techdocs.akamai.com/cloud-computing/docs/grant-a-developer-access.md)
* [Automate cloud resource deployment](https://techdocs.akamai.com/cloud-computing/docs/automate-cloud-resource-deployment.md)
* [Resell services](https://techdocs.akamai.com/cloud-computing/docs/resell-services.md)
* [Distributed compute regions (limited availability)](https://techdocs.akamai.com/cloud-computing/docs/distributed-compute-regions.md)
* [New data centers 2026](https://techdocs.akamai.com/cloud-computing/docs/new-data-centers-beta-2026.md)
* [NVIDIA RTX PRO 6000 Blackwell GPU Onboarding (limited availability)](https://techdocs.akamai.com/cloud-computing/docs/nvidia-rtx-pro-6000-blackwell-gpu-onboarding.md)