What Is the Hi/Lo Algorithm?

1. Introduction

In this tutorial, we'll explain the Hi/Lo Algorithm. It is used mostly as a database identifier generation strategy.

We'll start with the algorithm overview. Then, we'll show a practical example based on the Hibernate framework. Finally, we'll discuss the algorithm's use cases, its benefits, and its drawbacks.

2. Hi/Lo Algorithm Overview

2.1 Definition

The main purpose of the Hi/Lo algorithm is to create a range of numbers that can be safely used as database identifiers. In order to do that, it uses three number variables commonly called high, low, and incrementSize.

The incrementSize variable holds the maximum number of identifiers that can be generated in one batch. It should be treated as a constant value defined at the beginning of the algorithm. Any runtime modification might cause serious problems in environments where multiple clients use the same Hi/Lo configuration to persist entries.

The high variable is usually assigned from a database sequence. In that case, we're sure that no one will get the same number twice.

The low variable holds the currently assigned number in the range [0, incrementSize).

Given these points, the Hi/Lo algorithm generates values in range [(hi – 1) * incrementSize + 1, (hi * incrementSize)).

2.2 Pseudocode

Let’s take a look at the steps for generating a new value using the Hi/Lo algorithm:

• if low is greater than or equal to incrementSize, assign a new value to high and reset low to 0
• generate a new value with the formula: (high – 1) * incrementSize + low
• increment low by 1
• return the generated value

3. Practical Example

Let's see the Hi/Lo algorithm in action. To do that, we'll use the Hibernate framework and its Hi/Lo implementation.

First, let's define a database entity to work with:

@Entity
public class RestaurantOrder {
    @Id
    @GeneratedValue(strategy = GenerationType.SEQUENCE, generator = "hilo_sequence_generator")
    @GenericGenerator(
      name = "hilo_sequence_generator",
      strategy = "sequence",
      parameters = {
        @Parameter(name = "sequence_name", value = "hilo_seqeunce"),
        @Parameter(name = "initial_value", value = "1"),
        @Parameter(name = "increment_size", value = "3"),
        @Parameter(name = "optimizer", value = "hilo")
      }
    )
    private Long id;
}

It's a simple restaurant order with one id field. To correctly define the Hi/Lo algorithm in Hibernate, in the definition of the id field, we must choose a sequence strategy – hilo optimizer – and specify the increment_size parameter.

To show the Hi/Lo algorithm in action, we'll persist nine restaurant orders in a loop:

public void persist() {
    Transaction transaction = session.beginTransaction();

    for (int i = 0; i < 9; i++) {
        session.persist(new RestaurantOrder());
        session.flush();
    }

    transaction.commit();
}

According to the specified increment size in the entity, we should have only three calls to the database for the next high value. Assuming the database sequence starts from 1, the first batch of generated identifiers will be in the range [1,3].

When the Hi/Lo algorithm returns 3 and Hibernate asks for the next identifier's value, the value of the low variable is equal to the incrementSize constant. In that case, the next call to the database for the new high value must be made. Having 2 as the new high value, the algorithm generates values in the range [4,6].

Finally, the last call to the database for the next high value is made, and values in the range [7, 9] are assigned to the entities.

Hibernate logs captured during the execution of the persist() method confirm those values:

Hibernate: call next value for hilo_seqeunce
org.hibernate.id.enhanced.SequenceStructure  - Sequence value obtained: 1
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 1, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 2, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 3, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
Hibernate: call next value for hilo_seqeunce
org.hibernate.id.enhanced.SequenceStructure  - Sequence value obtained: 2
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 4, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 5, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 6, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
Hibernate: call next value for hilo_seqeunce
org.hibernate.id.enhanced.SequenceStructure  - Sequence value obtained: 3
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 7, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 8, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator
org.hibernate.event.internal.AbstractSaveEventListener  - Generated identifier: 9, using strategy: org.hibernate.id.enhanced.SequenceStyleGenerator

4. Algorithm Benefits and Drawbacks

The main advantage of the Hi/Lo algorithm is the reduced number of database calls for the next sequence values. Increasing the value of incrementSize decreases the number of round-trips to the database. Obviously, that means a performance gain in our application. In addition to that, the Hi/Lo algorithm is a preferred choice in environments with a weak Internet connection.

On the other hand, the Hi/Lo algorithm isn't the best choice in environments where multiple different clients persist data to the same table in a database. Third-party applications might be unaware of the Hi/Lo strategy we're using to generate identifiers. As a result, they might use entity ids from the generated range of numbers used currently in our application. In that case, when persisting data, we may encounter errors that are difficult to fix.

5. Conclusion

In this tutorial, we discussed the Hi/Lo algorithm.

First, we explained how it works and discussed its pseudocode implementation. Then, we showed a practical example using Hibernate's algorithm implementation. Finally, we listed Hi/Lo benefits and drawbacks.

As always, the code shown in this article is available over on GitHub.