21 years and counting of 'eight fallacies of distributed computing' (2025)
The 'Eight Fallacies of Distributed Computing' remain strikingly relevant 21 years after their inception, serving as critical reminders for anyone building networked systems. This piece revisits these foundational assumptions, originally conceived at Sun Microsystems, detailing why common beliefs about network reliability, latency, and security are often dangerously optimistic. It's a timeless technical deep dive that resonates with developers and network operators striving for robust distributed architectures.
The Lowdown
This article reflects on the enduring significance of the 'Eight Fallacies of Distributed Computing,' a set of mistaken assumptions about networks that continue to challenge developers and administrators despite decades of experience. Originating from insights by Bill Joy, Tom Lyon, L. Peter Deutsch, and James Gosling at Sun Microsystems, these fallacies highlight the inherent complexities of distributed systems.
- Origin of the Fallacies: The list began with four core fallacies from Bill Joy and Tom Lyon, later expanded by L. Peter Deutsch and James Gosling, all key figures at Sun Microsystems, a company instrumental in the early internet and technologies like ZFS, NFS, and Java.
- The Network is Reliable: Challenges the notion that data transmission is guaranteed, emphasizing the need for protocols (like TCP and QUIC) to handle packet loss.
- Latency is Zero: Dispels the idea of instantaneous communication, explaining how physical distance, signal conversions, and jitter contribute to unavoidable delays.
- Bandwidth is Infinite: Highlights the finite nature of network capacity, leading to queuing, delay, jitter, and packet loss, particularly relevant for home connections and even within CDNs.
- The Network is Secure: Argues against inherent network security, stressing the necessity of strong encryption (HTTPS/TLS) and cautioning that even encrypted traffic can reveal patterns through traffic analysis.
- Topology Doesn't Change: Explains that network paths are dynamic due to mobile connections, provider optimizations, and routing protocols, which introduce loss, delay, and jitter.
- There is One Administrator: Points out the reality of multiple, often uncoordinated, administrators and processes even within a single network operation center, complicating management.
- Transport Cost is Zero: Reveals that data transport always incurs costs (electricity, hardware, support), even if not directly exposed in protocols, and influences network design and pricing models like those for cloud storage.
- The Network is Homogeneous: Dispels the illusion of uniform network performance, noting significant differences between various links, devices, and protocols that higher layers must manage.
The article underscores that understanding these fallacies is crucial for designing resilient network software and protocols, urging practitioners to account for the unpredictable realities of distributed environments rather than making naive assumptions.