Skip to content

Scale_Up_and_Scale_Out.md

Latest commit

 

History

History
74 lines (44 loc) · 2.65 KB

Scale_Up_and_Scale_Out.md

File metadata and controls

74 lines (44 loc) · 2.65 KB

Scale Up and Scale Out

Reference: library.gabia.com



Two Methods for Server Expansion - Scale Up, Scale Out


When more server capacity and performance are needed, you can expand the system using two methods: Scale Up and Scale Out


1. Scale Up

  • Scale-up means increasing the server's own performance by adding CPU or RAM to the server, or replacing it with higher-performance components or a server
    • In other words, it means upgrading from the existing server to a higher-performance server
      • Also called "Vertical Scaling"
  • Since all data is processed on a single server, it is a suitable method for DB Servers where data updates occur frequently

![Screen Shot 2020-09-22 at 3.12.23 PM](../../kor/images/Screen Shot 2020-09-22 at 3.12.23 PM.png)


Problems with Scale-up

  • Since all the load is concentrated on a single server, there is a risk of being significantly affected in the event of a failure

  • The problem of scalability limitations of the storage controller

  • Migration costs that arise when performance & capacity expansion limits are reached and a new system needs to be added



2. Scale Out

  • A method of improving processing capability by increasing the number of servers with the same or similar specifications as the existing server
    • Also called "Horizontal Scaling"
  • Expansion is somewhat more flexible compared to Scale-up
  • Since there are multiple servers, load balancing must be implemented to evenly distribute the load across each server

![Screen Shot 2020-09-22 at 3.20.20 PM](../../kor/images/Screen Shot 2020-09-22 at 3.20.20 PM.png)


Advantages of Scale Out

  • Even if one server goes down due to a failure, service can continue to be provided through other servers

Problems with Scale Out

  • Since all servers must have the same data, it is a suitable method for web servers where data changes are infrequent

  • Difficulty in designing and implementing Parallel Computing

    • There are fundamentally parts that need to be serialized (single processing),
    • There are bandwidth and synchronization issues, and
    • Performance does not simply increase as cores increase; bandwidth increases with the number of cores, potentially causing delays


Wrap-up

  • The appropriate method should be adopted depending on the situation
    • ex)
      • When heavy traffic occurs due to increased visitors, Scale-out is more effective and also cheaper
      • For OLTP (Online Transaction) that requires frequent database updates, Scale-up is more effective