Calibrate the disk performance|V4.3.1| docs|Distributed Database

Calibrate the disk performance

Last Updated：2024-06-28 05:30:31 Updated

Before you perform input/output operations per second (IOPS) resource isolation, you must calibrate the disk performance. If IOPS resource isolation is not required, skip this step.

Calculate the baseline IOPS value

Disk performance calibration is mainly about calibrating the IOPS value of the disk. The formula for calculating the baseline IOPS value is as follows:

Baseline IOPS value = Min(Maximum IOPS value supported by the disk, (Target bandwidth/Amount of data involved in I/O operations))

You must consider the following aspects when you determine the target bandwidth:

Specifications of the server, specifically, the bandwidth limit of the disk
Sensitivity requirements on the response time (RT)
Bandwidth used by the I/O manager, which must be excluded
10 to 20 MB/s of bandwidth reserved for clogs

For example, when calculating the baseline IOPS value for 16 KB read data for a disk, the method is as follows: Assume that the bandwidth limit of the disk is 600 MB/s, with an IOPS of 30000, and the business is highly sensitive to the RT. If using the full 600 MB would significantly increase the probability of jitter, and after considering the reserved bandwidth for clogs, the target bandwidth is confirmed as 400 MB. By using 16 KB read data, the maximum achievable IOPS for 400 MB bandwidth is calculated as (400 * 1024 KB) / 16 KB = 25600, hence the IOPS value can be calibrated to 25600. If the calculated IOPS value already reaches or exceeds the maximum IOPS limit of the disk, for example, if a disk has a maximum IOPS of 10000, then the IOPS value can be calibrated to 10000.

Calibration methods

OceanBase Database provides the disk performance calibration feature for you to calibrate the read/write performance of the disk where an OBServer node resides. At present, OceanBase Database supports two disk performance calibration modes:

Automatic calibration: A background automatic calibration job is triggered by the ALTER SYSTEM RUN JOB statement.
Manual calibration: You can refresh the disk performance information for manual calibration.

Automatic calibration

If the current OceanBase cluster runs with zero load and the data disk has much idle space, you can execute the ALTER SYSTEM RUN JOB statement to trigger a background job to automatically calibrate the disk performance. In this mode, the system automatically calibrates the data disk.

Log on to the sys tenant of the cluster as the root user.

Note that you must specify the corresponding parameters in the following sample code based on your actual database configurations.
```
obclient -h10.xx.xx.xx -P2883 -uroot@sys#obdemo -p***** -A
```
For more information about how to connect to a database, see Overview.
Trigger a disk calibration job by using an appropriate statement based on business scenarios.
- Trigger a disk calibration job for all OBServer nodes in a cluster
  The syntax is as follows:
```
ALTER SYSTEM RUN JOB "job_name";
```
  In the statement, job_name specifies the name of the background job. To trigger a background disk calibration job, specify io_calibration.
  
  Here is an example:
```
ALTER SYSTEM RUN JOB "io_calibration";
```
- Trigger a disk calibration job for all OBServer nodes in a specified zone
  
  Here is a sample statement:
```
ALTER SYSTEM RUN JOB "job_name" ZONE [=] zone_name;
```
  The parameters are described as follows:
  - job_name: the name of the background job. To trigger a background disk calibration job, specify io_calibration.
  - zone_name: the name of the zone for which the disk calibration job is to be triggered. At present, you can specify only one zone.
  Here is an example:
```
ALTER SYSTEM RUN JOB "io_calibration" ZONE = zone1;
```
- Trigger a disk calibration job for a specified OBServer node
  
  Here is a sample statement:
```
ALTER SYSTEM RUN JOB "job_name" SERVER [=] 'svr_ip:svr_port';
```
  The parameters are described as follows:
  - job_name: the name of the background job. To trigger a background disk calibration job, specify io_calibration.
  - svr_ip: the IP address of the OBServer node for which the disk calibration job is to be triggered. At present, you can specify only one OBServer node.
  - svr_port: the remote procedure call (RPC) port of the OBServer node for which the disk calibration job is to be triggered.
  Here is an example:
```
ALTER SYSTEM RUN JOB "io_calibration" SERVER = 'xx.xx.xx.1:2882';
```

Query the disk I/O calibration status.

After you trigger a disk calibration job, you can query the GV$OB_IO_CALIBRATION_STATUS or V$OB_IO_CALIBRATION_STATUS view for the I/O calibration status.

SELECT * FROM oceanbase.V$OB_IO_CALIBRATION_STATUS;

The query result is as follows:

+----------------+----------+--------------+-------------+----------------------------+-------------+
| SVR_IP         | SVR_PORT | STORAGE_NAME | STATUS      | START_TIME                 | FINISH_TIME |
+----------------+----------+--------------+-------------+----------------------------+-------------+
| xx.xx.xx.197   |     2882 | DATA         | IN PROGRESS | 2023-06-27 14:30:38.393482 | NULL        |
+----------------+----------+--------------+-------------+----------------------------+-------------+
1 row in set

Possible I/O calibration states include the following ones:

NOT AVAILABLE: I/O calibration is not started.
IN PROGRESS: I/O calibration is in progress.
READY: I/O calibration is completed.
FAILED: I/O calibration failed.

In the sample query result, the value of the STATUS field is IN PROGRESS, indicating that disk I/O calibration is in progress. After the disk I/O calibration is completed, the value of the STATUS field changes to READY, and the completion time is displayed in the FINISH_TIME field.

+----------------+----------+--------------+--------+----------------------------+----------------------------+
| SVR_IP         | SVR_PORT | STORAGE_NAME | STATUS | START_TIME                 | FINISH_TIME                |
+----------------+----------+--------------+--------+----------------------------+----------------------------+
| xx.xx.xx.197   |     2882 | DATA         | READY  | 2023-06-27 14:25:20.202022 | 2023-06-27 14:27:00.398748 |
+----------------+----------+--------------+--------+----------------------------+----------------------------+
1 row in set

Verify whether the disk I/O calibration takes effect.

After the disk I/O calibration is completed, you can query the GV$OB_IO_BENCHMARK or V$OB_IO_BENCHMARK view to check whether the calibration takes effect.

SELECT * FROM oceanbase.GV$OB_IO_BENCHMARK;

A sample query result is as follows:

+----------------+----------+--------------+-------+---------+--------+------+---------+
| SVR_IP         | SVR_PORT | STORAGE_NAME | MODE  | SIZE    | IOPS   | MBPS | LATENCY |
+----------------+----------+--------------+-------+---------+--------+------+---------+
| xx.xx.xx.197   |     2882 | DATA         | READ  |    4096 | 124648 |  486 |     128 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |    8192 | 118546 |  926 |     134 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |   16384 |  98870 | 1544 |     161 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |   32768 |  73857 | 2308 |     216 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |   65536 |  48015 | 3000 |     332 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |  131072 |  33780 | 4222 |     473 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |  262144 |  20650 | 5162 |     774 |
| xx.xx.xx.197   |     2882 | DATA         | READ  |  524288 |  12111 | 6055 |    1321 |
| xx.xx.xx.197   |     2882 | DATA         | READ  | 1048576 |   6237 | 6237 |    2565 |
| xx.xx.xx.197   |     2882 | DATA         | READ  | 2097152 |   2762 | 5524 |    5795 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |    4096 |  49771 |  194 |     321 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |    8192 |  48566 |  379 |     329 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |   16384 |  42784 |  668 |     373 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |   32768 |  35187 | 1099 |     454 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |   65536 |  24892 | 1555 |     642 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |  131072 |  12720 | 1590 |    1257 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |  262144 |   6889 | 1722 |    2322 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE |  524288 |   3452 | 1726 |    4636 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE | 1048576 |   1689 | 1689 |    9481 |
| xx.xx.xx.197   |     2882 | DATA         | WRITE | 2097152 |    876 | 1752 |   18296 |
+----------------+----------+--------------+-------+---------+--------+------+---------+
20 rows in set

where:

STORAGE_NAME indicates the name of the storage disk, and DATA indicates a data disk.
MODE indicates the I/O mode, which can be read or write.
SIZE indicates the size of each I/O request, in bytes.
IOPS indicates the number of I/O requests completed per second.
MBPS indicates the disk bandwidth, in MB/s.
LATENCY indicates the response time of the disk, in us.

Manual calibration

If the current OceanBase cluster runs at a load, you can actively refresh the disk performance information to manually calibrate the disk performance.

Log on to the sys tenant of the cluster as the root user.

Note that you must specify the corresponding parameters in the following sample code based on your actual database configurations.
```
obclient -h10.xx.xx.xx -P2883 -uroot@sys#obdemo -p***** -A
```
For more information about how to connect to a database, see Overview.
Actively refresh the disk performance information by using an appropriate statement based on business scenarios.
- Refresh the disk performance information for all OBServer nodes in the cluster
```
ALTER SYSTEM REFRESH IO CALIBRATION [STORAGE [=] 'storage_name'] [CALIBRATION_INFO [=] ("mode : size : latency : iops" [, "mode : size : latency : iops"])];
```
- Clear the disk performance information of all OBServer nodes in the cluster
```
ALTER SYSTEM REFRESH IO CALIBRATION [STORAGE [=] 'storage_name'] CALIBRATION_INFO = ("");
```
- Refresh the disk performance information for all OBServer nodes in a specified zone
```
ALTER SYSTEM REFRESH IO CALIBRATION [STORAGE [=] 'storage_name'] [CALIBRATION_INFO [=] ("mode : size : latency : iops "[, "mode : size : latency : iops"])] ZONE [=] zone_name;
```
- Refresh the disk performance information for a specified OBServer node
```
ALTER SYSTEM REFRESH IO CALIBRATION [STORAGE [=] 'storage_name'] [CALIBRATION_INFO [=] ("mode : size : latency : iops" [, "mode : size : latency : iops"])] SERVER [=] 'svr_ip:svr_port';
```
where:
- STORAGE indicates the name of the storage disk of OceanBase Database. At present, the value can be only DATA, which indicates a data disk.
- CALIBRATION_INFO indicates the disk performance information to be refreshed. If you do not specify this parameter, the disk performance information is refreshed from the internal table by default. When you set CALIBRATION_INFO, you must specify at least one record for both the read and write modes. Otherwise, the system returns an error.
  
  Note
  
  When you specify disk performance information, if no stress testing data of the current disk is available, you can use the flexible I/O (FIO) tester to perform a performance stress test to obtain performance data of the current disk. For more information about the FIO tester, visit its official website.
  - mode: the I/O mode. Valid values are r, w, read, and write.
  - size: the size of a single I/O request. The value must be followed by a unit, such as 4K. Supported units are K, KB, M, MB, G, and GB.
  - latency: the response time of the disk. If a numeric value is specified, the default unit is seconds. We recommend that you specify a unit together with a value and do not use the default unit. Supported units are us, ms, s, min, and h.
  - iops: the number of I/O requests processed per second.
- zone_name indicates the zone for which the disk performance information is to be refreshed. At present, you can specify only one zone.
- svr_ip indicates the IP address of the OBServer node for which the disk performance information is to be refreshed.
- svr_port indicates the RPC port of the OBServer node for which the disk performance information is to be refreshed.
Here is an example:
- Refresh the disk performance information for the data disks of all OBServer nodes in the cluster from the internal table
```
ALTER SYSTEM REFRESH IO CALIBRATION;
```
- Clear the disk performance information of data disks of all OBServer nodes in the cluster
```
ALTER SYSTEM REFRESH IO CALIBRATION CALIBRATION_INFO = ("");
```
- Refresh the disk performance information for the data disks of all OBServer nodes in zone1
```
ALTER SYSTEM REFRESH IO CALIBRATION STORAGE = 'DATA' CALIBRATION_INFO = ("read:4K:100us:200000","write:2M:5ms:1500") ZONE = zone1;
```
  In this example, the refreshed disk performance information is as follows: The response time of random reads of 4 KB data is 100 us with an IOPS value of 200000. The response time of random writes of 2 MB data is 5 ms with an IOPS value of 1500.
- Refresh the disk performance information for the data disk of a specified OBServer node
```
ALTER SYSTEM REFRESH IO CALIBRATION STORAGE = 'DATA' CALIBRATION_INFO = ("read:4K:100us:200000","write:2M:5ms:1500") SERVER = 'xx.xx.xx.1:2882';
```
Verify whether the disk I/O calibration takes effect.

After the disk performance information is refreshed, you can query the GV$OB_IO_BENCHMARK or V$OB_IO_BENCHMARK view to confirm whether the disk I/O calibration takes effect.
```
SELECT * FROM oceanbase.GV$OB_IO_BENCHMARK;
```
A sample query result is as follows:
```
+----------------+----------+--------------+-------+---------+--------+------+---------+
| SVR_IP         | SVR_PORT | STORAGE_NAME | MODE  | SIZE    | IOPS   | MBPS | LATENCY |
+----------------+----------+--------------+-------+---------+--------+------+---------+
| xx.xx.xx.197   |     2882 | DATA         | READ  |    4096 | 200000 |  781 |     100 |
+----------------+----------+--------------+-------+---------+--------+------+---------+
1 rows in set
```
where:
- STORAGE_NAME indicates the name of the storage disk, and DATA indicates a data disk.
- MODE indicates the I/O mode, which can be read or write.
- SIZE indicates the size of each I/O request, in bytes.
- IOPS indicates the number of I/O requests completed per second.
- MBPS indicates the disk bandwidth, in MB/s.
- LATENCY indicates the response time of the disk, in microseconds.

References

Configure resource isolation within a tenant