How Modern IM Systems Achieve Multi-Device Sync, Roaming, and Real-Time Search

Preface

IM (Instant Messaging) has become a core feature of many mobile and social applications. Modern IM products such as DingTalk rely on robust message systems that handle synchronization, storage, and retrieval across devices.

Core Functions of a Message System

Message synchronization : Deliver messages quickly and completely from sender to receiver, supporting online push, offline cache, and multi-device sync.

Message storage : Persist messages on the server to enable message roaming; traditional systems stored only on the client.

Message retrieval : Provide full‑text search and indexed access to stored messages.

Traditional Architecture vs Modern Architecture

Traditional architecture synchronizes messages first and stores them only for offline users. It lacks server‑side persistence, so message roaming and indexed search are unavailable.

Modern architecture stores messages before synchronizing. Two separate stores are used: a message storage repository for full history (supporting roaming) and a message sync repository for multi‑device synchronization. After persistence, online push is attempted; otherwise the receiver pulls messages from the sync store.

Basic Model

The article introduces Alibaba Cloud TableStore’s Timeline model, which abstracts messages as a logical queue with ordered SequenceIds. The model supports automatic incremental IDs, custom IDs, random access by ID, and custom field indexing.

Timeline Model

Each message has a unique sequential ID.

New messages receive a larger ID than any existing one, guaranteeing append‑only ordering.

Custom IDs can be used to satisfy application‑level ordering.

Random access by SequenceId enables forward or reverse reads.

Custom indexes allow full‑text and multi‑field queries.

Message Storage Model

Each conversation maps to an independent Timeline that stores all messages in conversation order, providing persistent storage and enabling message roaming.

Message Synchronization Model

Two synchronization strategies are described:

Read diffusion (pull) : Messages are written once to the conversation Timeline; each client pulls new messages, reducing write amplification but increasing read load.

Write diffusion (push) : Each recipient has a dedicated sync Timeline (inbox); messages are written to both the storage Timeline and each recipient’s sync Timeline, simplifying client reads at the cost of higher write volume.

IM systems typically adopt write diffusion because the read‑to‑write ratio is high (≈10:1). Hybrid approaches may be used for large groups.

Typical Architecture Design

The architecture consists of clients, stateless message servers, a message queue, processing workers, a storage and index layer, and a sync layer. New messages contain IDs, timestamps, types, and payloads; they are queued, consumed, persisted, indexed, and then pushed or pulled according to device state.

Conclusion

The article outlines the capabilities required by modern IM message systems, compares traditional and modern designs, introduces TableStore’s Timeline model for storage and synchronization, and presents a reference architecture for building scalable, searchable, multi‑device messaging services.