Single Server Setup

First, you might try to put everything on a single server.

Breaking that down, first each client makes a DNS request to the DNS server to get the IP address connected to www.mysite.com. Then requests are made to our IP address, which returns HTML pages/JSON for clients.

Split DB from Requests

To handle more users, we’ll split out the db so it can scale independently.

Vertical Scaling vs Horizontal Scaling

Vertical Scaling means adding more power to your servers. Horizontal Scaling means adding more servers to your pool of servers.

Pros of Vertical Scaling:

• Simplicity (fewer machines to manage)
• Can be stateful (horizontal architectures are mostly stateless).

Cons of Vertical Scaling:

• No failover/redundancy
• Hard limit (you can’t add RAM and processors to one server)
• Expensive (adding to one server continuously costs too much).

Load Balancing

With a load balancing setup, clients now query the load balancer, which routes requests to N number of servers in a private IP. The load balancer communicates with web servers through private IPs and returns the requests. Now we have graceful failover for requests made to the web.

Database Replication

To achieve failover for the database tier, one way is master slave partitioning.

Advantages:

• Better Performance: More queries are processed in parallel.
• Reliablity: data is still preserved even if a server goes down since you have 4 copies of the same data
• High availability: your site is up as long as one database instance is up, and if master goes down, one of the slaves will replace it.

Caching

A cache tier is a fast in-memory storage that stores the result of expensive results or frequently accessed data.

Caches make reads faster, since the web server can query the cache first, and if there’s a cache hit, it can serve that data instead.

you would cache normally like so (using memcached):

const Memcached = require(;
const cache = new Memcached();

cache.set("myKey", "hi there", 3600); // 1 hour
cache.get(; // should return 'hi there'

Caching improves performance, but is complicated when synchronizing the webserver with the cache.

Content Delivery Network

You can use a CDN to cache data, which serves static assets as long as the TTL has not expired.

Stateful vs Stateless Architecture

Stateful architectures allow the web server to keep state in between requests – but that requires that sessions are always routed to the same web server, which can be challenging and makes adding or removing servers problematic, since if a server goes down, some state will disappear for our users.

Stateless architectures do not keep the user’s state, instead, they store state in a data layer that has its own mechanisms for reliability. This both has higher consistency and availability.

Data Centers

• If you a lot of international users, you’ll want to divide the web servers and databases out into data centers.

If one data center is down, the load balancer will load balance to another region.

This adds some complication:

• Traffic redirection based on IP
• Data Synchronization: make all regions synchronized.

Message Queues

To decouple the server from other servers/workers, you can use a message queue. This message queue keeps a buffer of tasks to do, completes the task, then notifies consumers who are subscribed.

Loggin, metrics, automation

When working with a large system, you must have logging and metrics.

• Logging can be centralized or per-server.
• Metrics help you understand the performance of the system and the business.

Database Scaling