diff --git a/sql/mysql-compatibility.md b/sql/mysql-compatibility.md
index a685cf706eaf2..5709e097c2e6f 100644
--- a/sql/mysql-compatibility.md
+++ b/sql/mysql-compatibility.md
@@ -41,14 +41,14 @@ The auto-increment ID feature in TiDB is only guaranteed to be automatically inc
 > create table t(id int unique key auto_increment, c int);
 > ```
 > 
-> The principle of the auto-increment ID in TiDB is that each tidb-server instance caches a section of ID values for allocation and fetches the next section after this section is used up.
+> The principle of the auto-increment ID in TiDB is that each tidb-server instance caches a section of ID values (currently 30000 IDs are cached) for allocation and fetches the next section after this section is used up.
 >
-> Assume that the cluster contains two tidb-server instances, namely Instance A and Instance B. Instance A caches the auto-increment ID of [1, 5000], while Instance B caches the auto-increment ID of [5001, 10000].
+> Assume that the cluster contains two tidb-server instances, namely Instance A and Instance B. Instance A caches the auto-increment ID of [1, 30000], while Instance B caches the auto-increment ID of [30001, 60000].
 > 
 > The operations are executed as follows:
 >
 > 1. The client issues the `insert into t values (1, 1)` statement to Instance B which sets the `id` to 1 and the statement is executed successfully.
-> 2. The client issues the `insert into t (c) (1)` statement to Instance A. This statement does not specify the value of `id`, so Instance A allocates the value. Currently, Instances A caches the auto-increment ID of [1, 5000], so it allocates the `id` value to 1 and adds 1 to the local counter. However, at this time the data with the `id` of 1 already exists in the cluster, therefore it reports `Duplicated Error`.
+> 2. The client issues the `insert into t (c) (1)` statement to Instance A. This statement does not specify the value of `id`, so Instance A allocates the value. Currently, Instances A caches the auto-increment ID of [1, 30000], so it allocates the `id` value to 1 and adds 1 to the local counter. However, at this time the data with the `id` of 1 already exists in the cluster, therefore it reports `Duplicated Error`.
 
 ### Built-in functions