Why Is Redis Fast? Analysis of Single-Threading, IO Multiplexing, and Data Structures

Redis is single-threaded yet can handle 100,000 QPS. This article explains the design behind it.

1. Why Single-Threaded?

Redis chose single-threading not because it can't write multi-threaded code, but because memory operations are so fast that multi-threading's context switching and lock contention actually slow things down.

The execution path of a Redis command:

Parse command → Look up key → Execute operation → Return result

All are pure in-memory operations, taking microseconds. In comparison, thread switching takes 1-10 microseconds, and lock contention is even more expensive.

However, Redis 6.0 introduced IO multi-threading — network reads/writes can be multi-threaded, but command execution remains single-threaded.

2. IO Multiplexing: epoll

How does a single thread handle tens of thousands of client connections? The answer is epoll.

# Pseudocode: epoll workflow
epoll_fd = epoll_create()
for client in clients:
    epoll_ctl(epoll_fd, ADD, client.fd, EPOLLIN)

while True:
    events = epoll_wait(epoll_fd, timeout=10)  # Blocking wait
    for fd, event in events:
        handle_client(fd)  # Only handle ready connections

Comparison: - select: Iterates over all fds → O(n), max 1024 - poll: Iterates over all fds → O(n), no upper limit but not efficient - epoll: Only returns ready fds → O(1), based on red-black tree + ready list

3. Efficient Data Structures

Redis is fast not just because of memory, but also because of cleverly designed data structures.

SDS (Simple Dynamic String):

struct sdshdr {
    int len;     // Used length
    int free;    // Remaining space
    char buf[];  // Actual data
};

• O(1) string length retrieval (C strings require O(n))
• Pre-allocated space reduces realloc calls
• Binary safe

ziplist: Contiguous memory storage saves pointer overhead. 5-10x more compact than linked lists for small data volumes.

skiplist: The underlying structure for ZSet, average O(log n), simpler to implement than balanced trees.

4. Performance Data

Redis's speed is the result of "correct architectural choices × extreme data structure optimization," not something brute-force hardware scaling can match.