Unraveling the Mysteries of Telegram Data Centers

Originally analyzed by Coxxs

Telegram utilizes a distributed infrastructure consisting of five distinct data centers, identified in the platform's documentation and source code as DC1 through DC5.

🌐 Global Distribution of DCs

The physical locations of these servers are spread across three primary hubs:

How Account Assignment Works

When a user registers, they are permanently linked to a specific DC. This assignment is static; it does not shift based on the user's current GPS location or changes to their phone number.

Note: Connection is strict. If a client attempts to communicate with a DC not associated with that specific account, the server will reject the request and trigger an error, forcing the client to redirect to the correct DC.

---

📉 The Notorious DC5

Within the Chinese Telegram community, DC5 is a frequent topic of conversation. However, it isn't praised for its efficiency, but rather infamous for its instability.

When DC5 experiences downtime, affected users are trapped in a perpetual "reconnecting" loop. Once service is restored, these users typically migrate to group chats with members of other DCs to vent their frustrations.

---

🔍 The Case of the "Missing" DC2 and DC3

To satisfy community curiosity, a bot was developed to identify which DC a user is assigned to. The initial findings among a sample of users were as follows:

• +86 (China) numbers $\rightarrow$ Assigned to DC5 (Singapore).
• +1 (USA) numbers $\rightarrow$ Assigned to DC1 (Miami).
• European numbers $\rightarrow$ Assigned to DC4 (Amsterdam).

Using "big data" (bot logs), the distribution looked like this: $\text{DC1} = 360, \quad \text{DC4} = 50, \quad \text{DC5} = 390, \quad \text{DC2/DC3} = 0$

This led to two theories:

1. DC2 and DC3 are completely empty.
2. DC2 and DC3 act as "overflow" servers for DC4 and DC1, respectively.

---

📜 Telegram DC Allocation Rules (As of May 2022)

TL;DR: User assignment to DC1, DC2, DC4, and DC5 is determined by the country code of the registration phone number. These four centers are active and continuously accept new users.

Regarding the "missing" centers:

• DC3: While it once hosted users, it appears that around 2020, existing accounts were migrated to DC1. It currently accepts no new registrations.
• DC2: This center is very much active. For example, users with a German number (+49) are typically routed here.

Why did the bot fail to find DC2 users?

The bot's detection logic was flawed. To prove this, the author tested three different identification methods:

Detailed Method Analysis

• Method 1: The Login Attempt By using a phone number destined for DC2 but attempting to connect via DC1 using the MTProto protocol, the server returns a PHONE_MIGRATE_2 error. This explicitly tells the client to move to DC2.

  • Downside: Requires the user's phone number.

• Method 2: The File/Profile Method Using a third-party client like Plus Messenger, one can view the dc_id field within the userProfilePhoto structure.

  • Nuance: The DC is only visible after a profile picture is uploaded, as the DC is determined by where the physical file is stored.

• Method 3: The Web CDN Method The bot used this method by analyzing the source code of: https://t.me/dctest

  If a profile picture URL starts with cdn4, the bot assumes the user is on DC4. However, DC2 and DC3 borrow the CDN domains of DC4 and DC1 (their regional counterparts). Consequently, all DC2 users were misidentified as DC4 users.

---

👻 The Disappearance of DC3

While DC2 users are common globally, DC3 has become a "ghost" center. Even when using the accurate Method 2, DC3 users are nearly impossible to find.

The author identified two potential candidates: @urie** and @flowinglig**. However, deeper analysis using photos.GetUserPhotos suggested that even these users might not truly reside on DC3.

Visual Evidence from the Investigation: