On the northern outskirts of the ancient capital stood the nine-story Stargazing Tower. At its pinnacle, within the central pavilion, hung the most sacred possession of the Astronomical Bureau: a colossal star chart painted on cerulean silk.

Every day, hundreds of seafarers, court astronomers, imperial students, and roaming caravaneers ascended the tower. They gathered beneath the great silk canopy, copying coordinate degrees and verifying navigation tracks. There was no guard at the entrance and no register to sign. Whoever arrived stepped forward, dipped a brush in ink, and began calculating at once—no gatekeeper, no queue, and no contention.

Yet the sky is never frozen.

Equinoxes shift, shooting stars blaze through the void, and wandering planets periodically retrograde. Every few days, the imperial astronomers had to re-verify celestial alignments and correct dozens of coordinate lines upon the chart.

Years ago, a newly appointed junior minister attempted to handle revisions by the old administrative manual. Whenever the chart required correction, he ordered guards to strike a bronze gong, shouting, “Clear the hall! The tower is barred for revision!” They expelled dozens of calculating scholars into the courtyard and latched the great vermillion doors shut for four continuous hours.

Disaster struck swiftly. A merchant fleet laden with imperial silk ran aground on shallow shoals because their navigators could not obtain the dusk celestial bearings; imperial students locked outside miscalculated the solar terms for an entire agricultural season. The Chief Astrologer was incandescent with fury. He took the great iron padlock and cast it down the cliffside: “The star charts belong to the realm! Those who read them must never be halted for a heartbeat! Who gave you the audacity to bar the doors?!”

The elder archivist who succeeded him never locked a door, nor did he ever raise his voice within the hall.

His manner of updating the chart was fascinating to behold.

Whenever the heavens shifted, the elder neither scraped the old silk in front of the scholars nor defaced it with crimson blotches. Instead, he retired quietly to the inner study, rolled out a pristine scroll of fresh cerulean silk, and meticulously traced the constellations from the old draft, amending only the coordinate lines for the shifted bodies. This was known as crafting anew in stillness.

When the new scroll was complete and stamped with the bureau seal, the elder carried the roll into the great rotunda. Seizing a moment when two seated scholars dipped their brushes into inkstones, he lightly stepped onto the central dais. With his left hand he unhooked the brass latch of the old chart, and with his right hand slid the new chart into the overhead timber groove. The entire exchange took less than a breath—silent as drifting willow catkins.

In the very next instant, scholars stepping through the courtyard archway looked up and immediately beheld the freshly calibrated constellations. Their subsequent calculations were accurate to the minute.

Yet within the hall still sat eight scholars who had taken their seats before the elder swapped the scrolls. Some were verifying the trajectory of a guest star observed the night before; others were deeply absorbed in calculations based on the half-copied notes in their hands. The elder archivist neither urged them to look up nor snatched the old scroll from their sight to cast it into the brazier.

He simply leaned against a wooden pillar with a steaming cup of tea in his palms, quietly watching the room.

What he watched for was each scholar’s departure.

Half an hour later, a mathematician at the eastern table pushed his inkstone aside, let out a satisfied breath, gathered his bamboo slips, and walked out the door—his inquiry was finished. A quarter-hour later, a young apprentice who had fallen asleep at the western desk rubbed his bleary eyes, yawned, and crossed the threshold to wash his face in the courtyard fountain—he too had stepped away. By late afternoon, the last weathered sea-captain folded his compass parchment, bowed three times toward the canopy, and descended the tower stairs with his walking staff.

Every single scholar who had been in the hall when the elder swapped the charts had now passed through the courtyard gate. Even though waves of new visitors had entered in the meantime, the elder knew with absolute clarity: every set of eyes that had ever beheld the old chart was now completely disengaged from it.

Only then did the elder set down his teacup, step behind the partition, roll up the retired silk scroll, and consign it to the charcoal hearth.

From beginning to end, not a single person had waited at a locked gate, not a single calculation had been interrupted, and the constellations transitioned as seamlessly as the turning of the seasons.

By now, you have probably recognized the elder’s wisdom: this is Read-Copy-Update (RCU) in the Linux kernel.

What it is

In modern operating systems and concurrent architecture, Read-Copy-Update (RCU) is a high-performance synchronization mechanism specifically designed for read-heavy, write-rare data structures.

Traditional reader-writer locks (rwlocks) permit multiple concurrent readers, but readers must still execute atomic operations to increment and decrement lock reference counters. Under massive CPU core counts, this triggers brutal cacheline bouncing across processor sockets. Furthermore, when a writer arrives, it must either block all incoming readers or wait for active readers to clear, causing severe latency spikes.

RCU fundamentally upends this design by splitting updates into three clean phases:

1. Lock-free Reader Critical Sections: Readers entering an RCU critical section do not acquire locks and execute zero atomic bus instructions (in non-preemptible kernels, it merely requires disabling preemption). They traverse shared data structures via pointers with zero wait time and minimal overhead.
2. Copy and Publish: When a writer wishes to modify a data structure (such as updating a node in a linked list), it first allocates a private copy, applies the modifications to the copy, and then atomically overwrites the global pointer (e.g., via rcu_assign_pointer). From that exact microsecond forward, all newly arriving readers observe the new version, while existing readers safely continue traversing the old version undisturbed.
3. Grace Period and Deferred Reclamation: The writer cannot immediately reclaim or free the old memory object because concurrent readers may still hold references to it. The writer enters a Grace Period. During this grace period, every CPU or worker thread must pass through at least one Quiescent State—such as a context switch or idle loop, which guarantees that any pre-existing reader critical section has concluded. Once all CPUs have reported a quiescent state, the writer knows that no thread holds a reference to the old memory, and safely reclaims it (via synchronize_rcu or asynchronous call_rcu).

Why it matters

In operating system kernels, routing engines, and high-throughput microservice caches, read-to-write ratios frequently exceed 100,000 to 1:

• Linux Kernel IP Routing & Netfilter Connection Tracking: Packets arrive at millions of frames per second, each requiring a routing table lookup. The routing table itself changes only when network topology shifts. RCU allows packet forwarding to scale linearly with CPU core counts without encountering lock contention.
• Virtual Filesystem (VFS) Dentry Cache: High-concurrency directory path resolution relies on RCU-walk mode to traverse file trees without locking inodes, giving Linux unmatched filesystem lookup throughput.
• Service Discovery & Config Routing (Local Immutability): Local microservice proxies look up upstream service pools millions of times a minute. Background threads fetch endpoint updates, publish new immutable maps, and defer reclamation of old routing tables until in-flight requests drain.

The genius of RCU is trading a delay in memory reclamation for uncompromised, wait-free read performance:

• It tolerates a brief transitional window where concurrent readers may observe either the older or newer version;
• It eliminates reader-side lock contention and cache-invalidation penalties;
• It achieves near-ideal linear scalability across modern multi-core computing architectures.

Metaphor mapping

• The grand cerulean star chart hung in the central rotunda → The shared data structure referenced by a global pointer
• Scholars consulting and copying the chart freely without locks → Lockless read-side critical sections
• Barring the tower doors and halting trade fleets → Reader-writer lock contention and Stop-the-World latency spikes
• Painting the revised chart on fresh silk in the back room → The Copy phase (modifying an isolated duplicate)
• Unhooking the old chart and hanging the new chart in one breath → Atomic pointer replacement and publication (rcu_assign_pointer)
• Newly arriving scholars immediately calculating from the new chart → New readers observing the published pointer
• Scholars who were already calculating from the old chart before the swap → In-flight readers traversing the pre-existing version
• Scholars finishing their notes and stepping through the courtyard gate → CPUs passing through a Quiescent State (context switch)
• Every pre-existing scholar having departed the room → Completion of the Grace Period
• Consigning the old silk scroll to the charcoal brazier → Deferred memory reclamation (synchronize_rcu / kfree)