Does the Random algorithm have any practical benefit?

Short answer

Yes. Random in TR7 is not a completely blind random selection: for each request, it draws two candidates in proportion to Weight and selects the one that is less loaded at that moment (the "power of two random choices" method). This reduces the tendency of algorithms that look for the least loaded server to pile up on a newly added or newly freed server. It is beneficial in large or frequently changing server pools and with many short, independent requests.

How Does It Work?

1. For each request, two candidates are drawn among the servers with a probability proportional to their Weight values.
2. Of the two candidates, the one with less load at that moment is selected; this way, the probability of the most loaded server in the pool being selected drops markedly.
3. When a server is added or removed, the selection adapts gradually; not all new requests are directed to the same server.

When Is It Beneficial?

Situation	Contribution of Random
The server pool is large or scales frequently	No sudden pile-up on a newly added server
More than one load balancer sends traffic to the same servers	The balancers do not select the same "emptiest" server at the same time
Many short, independent requests	Balanced distribution at low cost
Server capacities differ	Weight is reflected in the probability of the candidates being drawn

For loads that require persistence or where connections last very long, Least Connection or Persistence is more suitable.

In TR7

• Selection: vService → Details → LB Algorithm → Algorithm → Random.
• vService type: HTTP and L7 TCP; Random does not exist on a Network vService.
• Relation to Fastest / Fastest+: When the Fastest algorithms cannot determine a server according to the selected metric, distribution is done with Random.
• Monitoring: The distribution is monitored with the request and session counts of the Backend Services on the Live Stats screen of the vService.

Common Misconceptions

• "Random leaves the distribution entirely to chance." Of the two drawn candidates, the less loaded one is selected.
• "Random ignores Weight." The probability of the candidates being drawn is proportional to Weight.
• "Random looks at the response time." The criterion is the current load of the server; Least Response Time is used for response time.
• "Random keeps the user on the same server." It does not provide persistence; if needed, Persistence is selected.

How it is done in the documentation

• HTTP (L7) vService · Algorithms
• L7 TCP vService · Algorithms
• Live Stats · Backend Services

Related questions

• Which algorithm should I choose; is there a decision guide?
• When is Least Connection better than Round Robin?
• Which load balancing algorithms are supported?