Distributed Systems

Introduction to Distributed Systems

In the 1990s, if a company’s database ran out of storage space, the IT department had only one solution: Vertical Scaling (Scaling Up). They had to turn off the server, throw it away, and buy a physically larger, exponentially more expensive supercomputer from IBM or Oracle.

Eventually, companies like Google and Amazon hit a physical limit. It was impossible to build a single supercomputer large enough to index the entire internet.

The solution was the Distributed System (Horizontal Scaling or Scaling Out). A Distributed System is a network of independent, relatively cheap computers (nodes) that communicate with each other via network messages to appear to the end-user as a single, massive, unified computer.

The Core Advantages

Distributed systems form the backbone of modern cloud computing and Big Data architectures (like Apache Hadoop, Apache Spark, and Amazon S3).

1. Infinite Scalability: If a distributed database (like Apache Cassandra) reaches its storage limit, you do not buy a bigger server. You simply buy another cheap 1TB hard drive server, plug it into the network, and the software automatically distributes the load across the new node. You can scale infinitely.
2. Fault Tolerance (High Availability): In a single-server architecture, if the motherboard fries, the entire company goes offline. In a distributed system with 1,000 nodes, hardware failure is treated as a normal, expected event. The data is replicated. If Node 42 catches fire, the system instantly routes queries to Node 43, and the end-user never notices.

The Fallacies of Distributed Computing

Building distributed systems is notoriously difficult. In 1994, L. Peter Deutsch outlined the “Fallacies of Distributed Computing”—false assumptions that inexperienced software engineers make when transitioning from single-machine code to distributed code.

1. The Network is Reliable: (It is not. Cables get unplugged, routers crash. Your code must handle lost messages).
2. Latency is Zero: (It takes milliseconds for an electronic signal to travel from a server in New York to a server in Tokyo. Distributed databases must account for this physical delay).
3. Bandwidth is Infinite: (You cannot send a 1TB file between two servers instantaneously).
4. The Network is Secure: (Any node communicating over a network can be intercepted).

Distributed Coordination

The hardest problem in distributed systems is State. If Server A processes a customer’s $50 deposit, how does Server B instantly know about it? If the user immediately tries to withdraw $50 from Server B before the message arrives from Server A, what happens?

To solve this, distributed systems rely heavily on complex coordination algorithms (like Paxos or Raft) and specialized “Zookeeper” nodes to manage leader election, lock management, and ensure all nodes agree on the “State” of the universe.

Conclusion

Distributed Systems are the architectural compromise that enabled the modern internet. By sacrificing the simplicity and instantaneous communication of a single motherboard, software engineers gained the ability to scale computational power and storage across thousands of cheap, redundant machines, creating the indestructible, globe-spanning architectures of the modern Data Lakehouse.