2.1.0

Benchmark

Scripts to benchmark the system and gather the basic hardware information.

Basic Benchmark

Basic Information

In this benchmark, it will

  • gather hardware information

    • get cpu information by lscpu
    • get memory information by free -h and cat /proc/meminfo
    • get gpu information by nvidia-smi
    • get the disk partitions by lsblk

  • benchmark

    • cpu

Prerequisite

  1. NVIDIA device driver: driver can be downloaded here. Once install, please check nividia-smi is installed, we use this command to get GPU device information.

Command

Run the benchmark script

cd modules/benchmark
sudo ./benchmark.sh

Report

The output goes to report.txt in the working directory. Here is a sample result in GCP GCE. The profile is

  • instance type: n1-standard-1 (1 vCPU, 3.75 GB ram)
  • gpu: Tesla K80

the result: report.txt

Storage Benchmark

Raw Disk

We use fio to benchmark the disk performance. There are 5 tests to run

  • RANDOM_WRITE: It simulates the traditional write-intensive DB work load. We observe the write IOPS and bandwidth for this test.
  • RANDOM_READ: It simulates the traditional read-intensive DB work load. We observe the read IOPS and bandwidth for this test.
  • SEQ_WRITE: It simulates to write big file to a disk.
  • SEQ_READ: It simulates to read big file from a disk.
  • LATENCY: In this test, we can understand the latency between client and the disk. Typically, local disks can outperform network disks on this metric.

The parameters of each test are listed as follow

NameFile SizeBlock SizeConcurrent Thread
SEQ_WRITE10G1M64
SEQ_READ10G1M64
RANDOM_WRITE2G4k64
RANDOM_WRITE2G4k64
LATENCY (30sec)10G4k1

Prerequisite

  1. fio: we use fio to benchmark the disk performance

Command

modules/benchmark/storage/run-fio.sh <device|file>

Caution: for raw device, the data in the disk will be destroyed. Please don't sepcify raw device for a formatted disk.

Ceph

In this benchmark, it will

  • Use rados bench to benchmark against ceph rados cluster
  • Use osd bench to benchmark against OSD daemon
  • Use rbd bench to benchmark against ceph rbd

Prerequisite

  1. An primehub-provisioned ceph cluster. From modules/rook.
  2. kubectl-cephtools: a kubectl plugin to run command against the rook ceph tools pod. Please put ./bin/kubectl-cephtools in to PATH environment variable

Command

  1. run rados benchmark

    modules/benchmark/storage/run-ceph-rados.sh

  2. run osd benchmark

    modules/benchmark/storage/run-ceph-osd.sh

  3. run rbd benchmark

```
modules/benchmark/storage/run-ceph-rbd.sh
```

PVC

In the benchmark, it will create a pvc for a given storageclass, create a pod mount on created pvc, and do fio against the mount point.

Prerequisite

  1. kubectl-mountpvc: a kubectl plugin to create a pod to mount a specified pvc. Please put ./bin/kubectl-mountpvc in to PATH environment variable

Command

  1. Create a pvc for this storage class. And create a pod to mount the pvc.

    cd modules/benchmark/storage
./mount-pvc-rwo.sh <storage-class>

  2. Run the fio test

    ./run-pvc-rwo.sh

  3. Clean up

    ./clean-pvc.sh

If read performance is ridiculously fast, please try to clean cache for each benchmark test

echo 3 > /proc/sys/vm/drop_cache

PVC NFS

In the benchmark, it will create a NFS pvc and the backing storage is on top of the given storageclass. At the same time, it creates a pod mount on the created pvc, and do fio against the NFS mount point.

Prerequisite

  1. Group volume NFS required. modules/primehub-groupvolume-nfs/
  2. kubectl-mountpvc: a kubectl plugin to create a pod to mount a specified pvc. Please put bin/kubectl-mountpvc in to PATH environment variable

Command

  1. Create a pvc for this storage class. And create a pod to mount the pvc.

    cd modules/benchmark/storage
./mount-pvc-nfs.sh <storage-class>

  2. Run the fio test

    ./run-pvc-nfs.sh

  3. Clean up

    ./clean-pvc.sh
← PreviousNext →