Testing EVS Disk Performance

For enterprise applications, storage performance has a direct impact on service running efficiency and user experience. However, many users overlook storage performance test during deployment, which often leads to performance bottlenecks under heavy workloads. Ensuring that storage devices can deliver stable, high performance is therefore critical. By testing block storage performance, you can better understand the capabilities of your storage devices and make necessary optimizations to achieve optimal performance.

The testing procedures differ between Windows and Linux environments.

Windows
Linux

Prerequisites

The partition's first sector is aligned to 4 KiB. Otherwise, disk performance will be significantly affected.
To test the performance of a shared disk, the following conditions must be met:
- The shared disk must be attached to multiple cloud servers.
- If the shared disk is attached to multiple ECSs, all the ECSs must belong to the same anti-affinity ECS group.
  If they do not meet this requirement, the shared disk cannot reach the optimal performance.
For an Extreme Single-copy SSD disk, a protection group has been configured to ensure high reliability of service data. There is no SLA guarantee for disk IOPS, throughput, or reliability metrics. Fluctuations in the metrics are normal.
For a General Purpose SSD V2 disk, you have confirmed the IOPS and throughput requirements to ensure that they can be configured correctly.

When using fio to perform a raw disk performance test, ensure that no partitions and file systems have been created on the disk and that no data is stored on it. Otherwise, the raw disk test will damage the file system, and the data on the disk will become read-only. In this case, the only option is to delete the disk and purchase a new one to continue the test. Therefore, avoid testing disks that store service data. If testing on such disks is a must, perform the test as follows:
- Back up the disk data before the test.
- In the test command, set the filename parameter to a file, for example, -filename=/opt/fiotest/fiotest.txt.
Performance tests are conducted in a specific test environment and are for reference only. In a production environment, the actual performance of an EVS disk may vary due to factors such as network conditions and concurrent access. If the fio test results do not meet your expectations, see Why Does My Disk Performance Test Using Fio Have Incorrect Results? for troubleshooting.

The way you test disk performance depends on the server OS. This section uses Windows Server 2019 Standard 64-bit as an example. For other Windows OSs, see the corresponding OS documents.

Install the performance-measurement tool Iometer before the test.

Log in to the server.

For how to log in to an ECS, see How Do I Log In to My ECS?

For how to log in to a BMS, see Windows BMS Login Methods.
Press win+R to open the Run window. Enter msinfo32 and click OK.

The system information window is displayed.
Choose components > storage > disks. In the right pane, view the partition offset.
- If 4096 can be divided by the parameter value, the partition is 4-KiB aligned. Go to 4.
- If 4096 cannot be divided by the parameter value, the partition is not 4-KiB aligned. Ensure 4-KiB alignment for the partition before continuing the test.
  
  If you delete the partition and select another first sector number for 4-KiB alignment, you will lose all the data on that partition.
Use Iometer to test the disk performance. For details, see the Iometer product documentation.

When the disk IOPS and throughput are tested, the parameter settings for Iometer and fio are the same. For details, see Table 1.

The following example uses Iometer to test the disk performance.
1. Set the workflow.
2. Set the test run time.
  In this example, the test run time is set to 10 minutes, with a 60-second ramp-up time. Disk performance is tested after the writes become stable.
3. Set the data block size and read/write policy. In this example, the disk size is set to 64 KiB, the policy is 100% sequential write.
4. View the test results.

If you use an old version Linux OS, for example CentOS 6.5, and run fdisk to create partitions, the default first sector number will not be 4-KiB aligned, which will greatly affect the test performance. For this reason, if such an OS is used, you are advised to select a new first sector number, one that is 4-KiB aligned, when creating partitions.

The way you test disk performance depends on the server OS. This section uses CentOS 7.2 64-bit as an example. For other Linux OSs, see the corresponding OS documents.

Log in to the server as user root.

For how to log in to an ECS, see How Do I Log In to My ECS?

For how to log in to a BMS, see Linux BMS Login Methods.
Install the performance test tool fio.

yum install fio

Before you start the test, run the following command to check whether the first sector number is 4-KiB aligned:

fdisk -lu

Information similar to the following is displayed:

[root@ecs-centos sdc]# fdisk -lu

Disk /dev/xvda: 10.7 GiB, 10737418240 bytes, 20971520 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk label type: dos
Disk identifier: 0x7db77aa5

   Device Boot      Start         End      Blocks   Id  System
/dev/xvda1   *        2048    20968919    10483436   83  Linux

Disk /dev/xvdb: 10.7 GiB, 10737418240 bytes, 20971520 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/xvdc: 53.7 GiB, 53687091200 bytes, 104857600 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk label type: dos
Disk identifier: 0x3cf3265c

   Device Boot      Start         End      Blocks   Id  System
/dev/xvdc1            2048    41943039    20970496   83  Linux

If the first sector number is divisible by 8, the partition is 4-KiB aligned. Go to 4.
If the first sector number is not divisible by 8, the partition is not 4-KiB aligned. Delete the partition and select a first sector number that is 4-KiB aligned for the new partition before continuing the test.

If you delete the partition and select another first sector number for 4-KiB alignment, you will lose all the data on that partition.

Run the following commands and use fio to test the disk performance:

To test random write IOPS, run the following command: fio -direct=1 -iodepth=128 -rw=randwrite -ioengine=libaio -bs=4k -size=10G -numjobs=1 -runtime=600 -group_reporting -filename=/opt/fiotest/fiotest.txt -name=Rand_Write_IOPS_Test
To test random read IOPS, run the following command: fio -direct=1 -iodepth=128 -rw=randread -ioengine=libaio -bs=4k -size=10G -numjobs=1 -runtime=600 -group_reporting -filename=/opt/fiotest/fiotest.txt -name=Rand_Read_IOPS_Test
To test write throughput, run the following command: fio -direct=1 -iodepth=32 -rw=write -ioengine=libaio -bs=1024k -size=10G -numjobs=1 -runtime=600 -group_reporting -filename=/opt/fiotest/fiotest.txt -name=Write_BandWidth_Test
To test read throughput, run the following command: fio -direct=1 -iodepth=32 -rw=read -ioengine=libaio -bs=1024k -size=10G -numjobs=1 -runtime=600 -group_reporting -filename=/opt/fiotest/fiotest.txt -name=Read_BandWidth_Test

To test single-queue, random read latency, run the following command: fio -direct=1 -iodepth=1 -rw=randread -ioengine=libaio -bs=4k -size=10G -numjobs=1 -runtime=60 -group_reporting -filename=/opt/fiotest/fiotest.txt -name=Rand_Read_LATE_Test

Table 1 lists the fio test parameters.

**Table 1** Parameter description
Parameter	Description
direct	Defines whether direct I/O is used. Set to 0: buffered I/O is used. Set to 1: direct I/O is used.
iodepth	Defines the I/O queue depth. This queue depth refers to the queue depth of each thread regardless of whether a single or multiple threads are used in the test. Total concurrent I/Os of fio = iodepth × numjobs Examples: If there is a single thread and -iodepth=32, the I/O queue depth of this thread is 32 and the total concurrent I/Os of fio is 32 (32 × 1). If there are three threads and -iodepth=32, the I/O queue depth of each thread is 32 and the total concurrent I/Os of fio is 96 (32 × 3).
rw	Defines the test read/write policy. randread: random read randwrite: random write read: sequential read write: sequential write randrw: mixed random read/write
ioengine	Defines how fio delivers the I/O request (synchronously or asynchronously). Synchronous I/O: Only one I/O request is delivered at a time, and the response is returned after the kernel has processed the request. That said, the single-thread I/O queue depth is always less than 1, and multi-thread concurrent processing can be used to handle such issues. Normally, 16 to 32 concurrent working threads fully occupy the I/O queue depth. Asynchronous I/O: Multiple I/O requests are delivered using libaio at a time. Wait for the process to complete and reduce the interaction times to improve efficiency.
bs	Defines the I/O block size. The unit can be KiB, KB, MiB, and MB, and the default value is 4 KiB.
size	Defines the amount of data processed by the test I/Os. If parameters, such as runtime, are not specified, the test ends when fio has processed all the specified data amount. The value can be a number with a unit or percentage. A number with a unit indicates the read/write data amount, for example size=10G, indicating a 10-GiB read/write data amount. A percentage indicates the ratio of read/write data amount to the total size of files, for example size=20%, indicating the read/write data amount takes 20% of the total file space.
numjobs	Defines the number of concurrent threads.
runtime	Defines the test time. If this parameter is not specified, the test ends until the specified amount of data is processed by the block size defined using parameter size.
group_reporting	Defines the test result display mode. The parameter value displays the statistics on a single thread instead of that on all jobs.
filename	Defines the name of the test file or device. If a file is specified, the performance of the file system is tested. Example: -filename=/opt/fiotest/fiotest.txt If a device name is specified, the performance of the raw disk is tested. Example: -filename=/dev/vdb NOTICE: If the test is performed on a disk that already has partitions and file systems created and data on it, use parameter filename to specify a file so that the original file system is not damaged and the data is not overwritten.
name	Defines the test task name.

Follow-Up Operations

After the test is complete, handle the disk as follows:
- If you want to retain the disk, do not use it directly. Instead, initialize it first.
- If you do not need to retain the disk, detach and delete it by referring to Detaching and Deleting an EVS Disk.