Contact UsGet started
© 2024 Iron Hive LLC Belgrade

Deploying the GlusterFS parallel file system in high performance mode

Written by
Updated at October 17, 2024

GlusterFS is a parallel distributed file system with linear scalability. With horizontal scaling, the system provides the cloud with an aggregate bandwidth of tens of GB/s and hundreds of thousands of IOPS.

Use this tutorial to create an infrastructure made up of 30 segments sharing a common GlusterFS file system. Placing storage disks in a single availability zone will ensure high performance of your file system. In this use case, performance is limited by the speed of accessing physical disks, while network latency is less important.

To configure a high-performance file system:

  1. Prepare your cloud.
  2. Configure the CLI profile.
  3. Prepare an environment for deploying the resources.
  4. Deploy your resources.
  5. Install and configure GlusterFS.
  6. Test the solution availability.
  7. Test the solution performance.

If you no longer need the resources you created, delete them.

Prepare your cloud

Sign up for Yandex Cloud and create a billing account:

  1. Go to the management console and log in to Yandex Cloud or create an account if you do not have one yet.
  2. On the Yandex Cloud Billing page, make sure you have a billing account linked and it has the ACTIVE or TRIAL_ACTIVE status. If you do not have a billing account, create one.

If you have an active billing account, you can go to the cloud page to create or select a folder for your infrastructure to operate in.

Learn more about clouds and folders.

The infrastructure support costs include:

Configure the CLI profile

  1. If you do not have the Yandex Cloud command line interface yet, install it and sign in as a user.

  2. Create a service account:

    1. In the management console, select the folder where you want to create a service account.
    2. In the Service accounts tab, click Create service account.
    3. Enter a name for the service account, e.g., sa-glusterfs.
    4. Click Create.

    The folder specified in the CLI profile is used by default. You can specify a different folder using the --folder-name or --folder-id parameter.

    Run the command below to create a service account, specifying the sa-glusterfs name:

    yc iam service-account create --name sa-glusterfs

    Where name is the service account name.

    Result:

    id: ajehr0to1g8b********
folder_id: b1gv87ssvu49********
created_at: "2023-06-20T09:03:11.665153755Z"
name: sa-glusterfs

    To create a service account, use the ServiceAccountService/Create gRPC API call or the create REST API method for the ServiceAccount resource.

  3. Assign the service account the administrator role for the folder:

    1. On the management console home page, select the folder.
    2. Go to the Access bindings tab.
    3. Find the sa-glusterfs service account in the list and click .
    4. Click Edit roles.
    5. Click Add role in the dialog box that opens and select the admin role.

    Run this command:

    yc resource-manager folder add-access-binding <folder_ID> \
   --role admin \
   --subject serviceAccount:<service_account_ID>

    To assign the service account a role for the folder, use the setAccessBindings REST API method for the ServiceAccount resource or the ServiceAccountService/SetAccessBindings gRPC API call.

  4. Set up the CLI profile to run operations on behalf of the service account:

    1. Create an authorized key for the service account and save it to the file:

      yc iam key create \
--service-account-id <service_account_ID> \
--folder-id <id_of_folder_with_service_account> \
--output key.json

      Where:

      • service-account-id: Service account ID.
      • folder-id: ID of the folder in which the service account was created.
      • output: Name of the file with the authorized key.

      Result:

      id: aje8nn871qo4********
service_account_id: ajehr0to1g8b********
created_at: "2023-06-20T09:16:43.479156798Z"
key_algorithm: RSA_2048

    2. Create a CLI profile to run operations on behalf of the service account:

      yc config profile create sa-glusterfs

      Result:

      Profile 'sa-glusterfs' created and activated

    3. Set the profile configuration:

      yc config set service-account-key key.json
yc config set cloud-id <cloud_ID>
yc config set folder-id <folder_ID>

      Where:

      • service-account-key: File with the authorized key of the service account.
      • cloud-id: Cloud ID.
      • folder-id: Folder ID.

    4. Add the credentials to the environment variables:

      export YC_TOKEN=$(yc iam create-token)
export YC_CLOUD_ID=$(yc config get cloud-id)
export YC_FOLDER_ID=$(yc config get folder-id)

Prepare an environment for deploying the resources

  1. Create an SSH key pair:

    ssh-keygen -t ed25519

    We recommend leaving the key file name unchanged.

  2. Install Terraform.

  3. Clone the yandex-cloud-examples/yc-distributed-ha-storage-with-glusterfs GitHub repository and go to the yc-distributed-ha-storage-with-glusterfs folder:

    git clone https://github.com/yandex-cloud-examples/yc-distributed-ha-storage-with-glusterfs.git
cd ./yc-distributed-ha-storage-with-glusterfs

  4. Edit the variables.tf file, specifying the parameters of the deployed resources:

    Warning

    The values set in the file result in deploying a resource-intensive infrastructure.
    To deploy the resources within your available quotas, use the values below or change the values according to your specific needs.

    1. In the is_ha section, change default to false.

    2. In client_node_per_zone, change default to 30.

    3. In storage_node_per_zone, change default to 30.

      Note

      In this use case, we will deploy 30 VMs. You can change this number depending on the requirements for the final storage size or total bandwidth.
      The maximum aggregate bandwidth of the entire system is calculated as each segment's bandwidth (450 MB/s for network SSDs) multiplied by the number of segments (30), which is around 13.5 GB/s.
      The system capacity is calculated as the number of segments (30) multiplied by the size of each storage (1 TB), which amounts to 30 TB.

    4. If you specified a non-default name when creating the SSH key pair, under local_pubkey_path, change default to <path_to_the_public_SSH_key>.

    5. If you need enhanced performance without guaranteed data durability, you can use non-replicated SSDs. For this, in disk_type, change default to network-ssd-nonreplicated. In addition, make sure the default value under disk_size is a multiple of 93.

Deploy your resources

  1. Run Terraform initialization: 
    terraform init
  2. Check the Terraform file configuration: 
    terraform validate
  3. Check the list of cloud resources you are about to create: 
    terraform plan
  4. Create resources: 
    terraform apply -auto-approve
  5. Wait until a process completion message appears: 
    Outputs:

connect_line = "ssh storage@158.160.108.137"
public_ip = "158.160.108.137"

This will create 30 VMs for hosting client code (client01, client02, etc.) in the folder and 30 VMs for distributed data storage (gluster01, gluster02, etc.) bound to the client VMs and placed in the same availability zone.

Install and configure GlusterFS

  1. Connect to the client01 VM using the command from the process completion output:

    ssh storage@158.160.108.137

  2. Switch to the root superuser mode:

    sudo -i

  3. Install ClusterShell:

    dnf install epel-release -y
dnf install clustershell -y
echo 'ssh_options: -oStrictHostKeyChecking=no' >> /etc/clustershell/clush.conf

  4. Create the configuration files:

    cat > /etc/clustershell/groups.conf <<EOF
[Main]
default: cluster
confdir: /etc/clustershell/groups.conf.d $CFGDIR/groups.conf.d
autodir: /etc/clustershell/groups.d $CFGDIR/groups.d
EOF      

cat > /etc/clustershell/groups.d/cluster.yaml <<EOF
cluster:
   all: '@clients,@gluster'
   clients: 'client[01-30]'
   gluster: 'gluster[01-30]'
EOF

  5. Install GlusterFS:

    clush -w @all hostname # check and auto add fingerprints
clush -w @all dnf install centos-release-gluster -y
clush -w @all dnf --enablerepo=powertools install glusterfs-server -y
clush -w @gluster mkfs.xfs -f -i size=512 /dev/vdb
clush -w @gluster mkdir -p /bricks/brick1
clush -w @gluster "echo '/dev/vdb /bricks/brick1 xfs defaults 1 2' >> /etc/fstab"
clush -w @gluster "mount -a && mount"

  6. Restart GlusterFS:

    clush -w @gluster systemctl enable glusterd
clush -w @gluster systemctl restart glusterd

  7. Check the availability of the gluster02 through gluster30 VMs:

    clush -w gluster01 'for i in {2..9}; do gluster peer probe gluster0$i; done'
clush -w gluster01 'for i in {10..30}; do gluster peer probe gluster$i; done'

  8. Create a vol0 folder in each storage VM and configure availability and fault tolerance by connecting to the stripe-volume shared folder:

    clush -w @gluster mkdir -p /bricks/brick1/vol0
export STRIPE_NODES=$(nodeset -S':/bricks/brick1/vol0 ' -e @gluster)
clush -w gluster01 gluster volume create stripe-volume ${STRIPE_NODES}:/bricks/brick1/vol0

  9. Configure additional performance settings:

    clush -w gluster01 gluster volume set stripe-volume client.event-threads 8
clush -w gluster01 gluster volume set stripe-volume server.event-threads 8
clush -w gluster01 gluster volume set stripe-volume cluster.shd-max-threads 8
clush -w gluster01 gluster volume set stripe-volume performance.read-ahead-page-count 16
clush -w gluster01 gluster volume set stripe-volume performance.client-io-threads on
clush -w gluster01 gluster volume set stripe-volume performance.quick-read off
clush -w gluster01 gluster volume set stripe-volume performance.parallel-readdir on
clush -w gluster01 gluster volume set stripe-volume performance.io-thread-count 32
clush -w gluster01 gluster volume set stripe-volume performance.cache-size 1GB
clush -w gluster01 gluster volume set stripe-volume performance.cache-invalidation on
clush -w gluster01 gluster volume set stripe-volume performance.md-cache-timeout 600
clush -w gluster01 gluster volume set stripe-volume performance.stat-prefetch on
clush -w gluster01 gluster volume set stripe-volume server.allow-insecure on
clush -w gluster01 gluster volume set stripe-volume network.inode-lru-limit 200000
clush -w gluster01 gluster volume set stripe-volume features.shard-block-size 128MB
clush -w gluster01 gluster volume set stripe-volume features.shard on
clush -w gluster01 gluster volume set stripe-volume features.cache-invalidation-timeout 600
clush -w gluster01 gluster volume set stripe-volume storage.fips-mode-rchecksum on

  10. Mount the stripe-volume shared folder on the client VMs:

    clush -w gluster01  gluster volume start stripe-volume
clush -w @clients mount -t glusterfs gluster01:/stripe-volume /mnt/

Test the solution availability

  1. Check the status of the stripe-volume shared folder:

    clush -w gluster01  gluster volume status

  2. Create a text file:

    cat > /mnt/test.txt <<EOF
Hello, GlusterFS!
EOF

  3. Make sure the file is available on all client VMs:

    clush -w @clients sha256sum /mnt/test.txt

    Result:

    client01: 5fd9c031531c39f2568a8af5512803fad053baf3fe9eef2a03ed2a6f0a884c85  /mnt/test.txt
client02: 5fd9c031531c39f2568a8af5512803fad053baf3fe9eef2a03ed2a6f0a884c85  /mnt/test.txt
client03: 5fd9c031531c39f2568a8af5512803fad053baf3fe9eef2a03ed2a6f0a884c85  /mnt/test.txt
...
client30: 5fd9c031531c39f2568a8af5512803fad053baf3fe9eef2a03ed2a6f0a884c85  /mnt/test.txt

Test the solution performance

IOR is a benchmark for concurrent I/O operations that can be used to test the performance of parallel data storage systems using various interfaces and access patterns.

  1. Install the dependencies:

    clush -w @clients dnf install -y autoconf automake pkg-config m4 libtool git mpich mpich-devel make fio
cd /mnt/
git clone https://github.com/hpc/ior.git
cd ior
mkdir prefix

  2. Close the shell and open it again:

    ^C
sudo -i
module load mpi/mpich-x86_64
cd /mnt/ior

  3. Install IOR:

    ./bootstrap
./configure --disable-dependency-tracking  --prefix /mnt/ior/prefix
make 
make install
mkdir -p /mnt/benchmark/ior

  4. Run IOR:

    export NODES=$(nodeset  -S',' -e @clients)
mpirun -hosts $NODES -ppn 16 /mnt/ior/prefix/bin/ior  -o /mnt/benchmark/ior/ior_file -t 1m -b 16m -s 16 -F
mpirun -hosts $NODES -ppn 16 /mnt/ior/prefix/bin/ior  -o /mnt/benchmark/ior/ior_file -t 1m -b 16m -s 16 -F -C

    Result:

    IOR-4.1.0+dev: MPI Coordinated Test of Parallel I/O
Options:
api                 : POSIX
apiVersion          :
test filename       : /mnt/benchmark/ior/ior_file
access              : file-per-process
type                : independent
segments            : 16
ordering in a file  : sequential
ordering inter file : no tasks offsets
nodes               : 30
tasks               : 480
clients per node    : 16
memoryBuffer        : CPU
dataAccess          : CPU
GPUDirect           : 0
repetitions         : 1
xfersize            : 1 MiB
blocksize           : 16 MiB
aggregate filesize  : 120 GiB

Results:

access    bw(MiB/s)  IOPS       Latency(s)  block(KiB) xfer(KiB)  open(s)    wr/rd(s)   close(s)   total(s)   iter
------    ---------  ----       ----------  ---------- ---------  --------   --------   --------   --------   ----
write     1223.48    1223.99    4.65        16384      1024.00    2.44       100.39     88.37      100.44     0
read      1175.45    1175.65    4.83        16384      1024.00    0.643641   104.52     37.97      104.54     0

How to delete the resources you created

To stop paying for the resources created, delete them:

terraform destroy -auto-approve
Previous
Deploying the GlusterFS parallel file system in high availability mode
Next
Backup to Object Storage via Bacula