BeeGFS

From ArchWiki
Jump to navigation Jump to search

Tango-edit-clear.pngThis article or section needs language, wiki syntax or style improvements. See Help:Style for reference.Tango-edit-clear.png

Reason: please use the first argument of the template to provide a brief explanation. (Discuss in Talk:BeeGFS#)

BeeGFS is a scalable network-storage platform with a focus on being distributed, resilient, highly configurable and having good performance and high reliability. BeeGFS is extremely configurable, with administrators being able to control virtually all aspects of the system. A command line interface is used to monitor and control the cluster.

From Wikipedia: BeeGFS (software):

BeeGFS (formerly FhGFS) is a parallel file system, developed and optimized for high-performance computing. BeeGFS includes a distributed metadata architecture for scalability and flexibility reasons. Its most important aspect is data throughput. BeeGFS was originally developed at the Fraunhofer Center for High Performance Computing in Germany by a team around Sven Breuner, who later became the CEO of ThinkParQ, the spin-off company that was founded in 2014 to maintain BeeGFS and offer professional services.

From BeeGFS.io:

BeeGFS is the leading parallel cluster file system, developed with a strong focus on performance and designed for very easy installation and management. If I/O intensive workloads are your problem, BeeGFS is the solution.

Terminology

Tip: A full glossary is available in the official documentation.
Node Type and Description Packages
Management Server (one node)
  • Manages configuration and group membership
  • Hostname or IP address must be known by other nodes at service start time
beegfs-mgmtdAUR
Metadata Server (at least one node)
  • Stores directory information and allocates file space on storage servers
beegfs-metaAUR
Storage Server (at least one node)
  • Stores raw file contents
beegfs-storageAUR
InfluxDB / Grafana based Monitoring Server (optional)
  • Continuous monitoring of servers
  • Live statistics
  • beegfs-admon (Java based administration and monitoring GUI), must not be installed on the same server
beegfs-monAUR
BeeGFS utilities for administrators
  • beegfs-ctl tool for command-line administration
  • beegfs-fsck tool for file system checking
  • Several small helper scripts such logging and DNS lookup functionality
beegfs-utilsAUR
BeeGFS Common beegfs-commonAUR
Client
  • Kernel module to mount the file system
  • Requires userspace helper daemon for logging and hostname resolution
beegfs-clientAUR


In addition to the free and open-source packages described here, BeeGFS also offers a number of Enterprise Features and Professional Support, which include:

  • High Availability
  • Quota Enforcement
  • Access Control Lists (ACLs)
  • Storage Pools
  • Burst buffer function with BeeOND
Warning: Whilst the BeeGFS server components are userspace daemons, the client is a native kernel module. The latest version of BeeGFS v7.1.3 has support for Kernels up to 4.19.x. Hence there are a number of adhoc patches to the client source build files included in the beegfs-clientAUR PKGBUILD. This in turn may lead to instability with the client kernel module.

Installation

Example cluster deployment

The following hardware configuration will be used in this example:

Hostname IP Address Description
node01 192.168.0.1 Management Server and Monitoring (optional) Server
node02 192.168.0.2 Metadata Server
node03 192.168.0.3 Storage Server
node04 192.168.0.4 Client
Tip: One if free to choose the option of using dedicated hosts for all BeeGFS services. BeeGFS allows running any combination of services (including client and storage/metadata service) on the same machine. Especially the management and mon daemons are not performance-critical and thus are typically not running on dedicated machines.

NTP client

Install and run a time synchronization client on all the nodes. See Time synchronization for details.

Note: It is strongly recommended to synchronize the clocks on all cluster nodes to prevent clock drift (see System time#Time skew for details), which can degrade the performance of your cluster or stop it from functioning altogether. The official documentation recommends that nodes run some form of clock synchronization.

Management server

Install it with the package beegfs-mgmtdAUR on the management node 192.168.0.1.

The management service needs to know where it can store its data. It will only store some node information like connectivity data, so it will not require much storage space and its data access is not performance critical. Thus, this service is typically not running on a dedicated machine.

/etc/beegfs/beegfs-mgmtd
storeMgmtdDirectory = /mnt/beegfs/beegfs-mgmtd

Start/enable the beegfs-mgmtd.service on the management node:

 # systemctl start beegfs-mgmtd@node01.service
 # systemctl enable beegfs-mgmtd@node01.service

Monitoring server

Install the package beegfs-monAUR on the management/monitoring node 192.168.0.1, which collects statistics from the system and provides them to the user using a time series database InfluxDB. For visualization of the data beegfs-mon provides predefined Grafana panels that can be used out of the box.

Before running beegfs-mon, you need to edit the configuration file /etc/beegfs/beegfs-mon.conf. If you have everything installed on the same host, you only need to specify the management host:

/etc/beegfs/beegfs-mon.conf
sysMgmtHost = localhost
Tip: If your InfluxDB is installed on another host, say the client for example or you need to use a different database port or name, you also need to modify the corresponding entries:
/etc/beegfs/beegfs-mon.conf
dbHostName = node04
dbHostPort = 9096
dbHostName = beegfs_mon_client

Start/enable the beegfs-mon.service on the management/monitoring node:

 # systemctl start beegfs-mon@node01.service
 # systemctl enable beegfs-mon@node01.service

Configuration of default Grafana panels

You can use the provided installation script for default InfluxDB and Grafana deployments on the same host.

 # cd /etc/beegfs/grafana
 # ./import-dashboards default

Accessing Grafana panels

Access the application on localhost, e.g.: http://127.0.0.1:3000 . Refer to Custom Grafana Panel Configuration for non-default installations and for the Reference to All Metrics monitored.

Note: Different services on the same machine cannot share the same storage directory, so different directories have to be used, i.e. /mnt/beegfs/beegfs-mgmtd for management servers and /mnt/beegfs/beegfs-mon for monitoring servers.

Metadata server

Install the package beegfs-metaAUR on the metadata server(s), i.e. 192.168.0.2.

The metadata service needs to know where it can store its data and where the management service is running. Typically, one will have multiple metadata services running on different machines.

/etc/beegfs/beegfs-meta.conf
sysMgmtdHost = node01
storeMetaDirectory = /mnt/beegfs/beegfs-meta

Start/enable the beegfs-meta.service on the metadata node.

 # systemctl start beegfs-meta@node02.service
 # systemctl enable beegfs-meta@node02.service

Storage server

Install the package beegfs-storageAUR on the storage server(s), i.e. 192.168.0.3.

The storage service needs to know where it can store its data and how to reach the management server. Typically, one will have multiple storage services running on different machines and/or multiple storage targets (e.g. multiple RAID volumes) per storage service.

/etc/beegfs/beegfs-storage.conf
sysMgmtdHost = node01
storeStorageDirectory = /mnt/beegfs/beegfs-storage

Start/enable the beegfs-storage.service on the storage node.

 # systemctl start beegfs-storage@node03.service
 # systemctl enable beegfs-storage@node03.service

Client

Install the package beegfs-clientAUR on the client node, which will build the client Kernel module.

The client service needs to know where it can reach the management server.

/etc/beegfs/beegfs-client.conf
sysMgmtdHost = node01

The client service needs to know where it can mount the cluster storage, as well as the location of teh client configuration file.

/etc/beegfs/beegfs-mount.conf
/mnt/beegfs/beegfs-mount /etc/beegfs/beegfs-client.conf

Load the Kernel module and its dependencies.

 # modprobe beegfs

Start/enable the beegfs-helperd.service on the client node:

 # systemctl start beegfs-helperd@node04.service
 # systemctl enable beegfs-helperd@node04.service

Start/enable the beegfs-client.service on the client node:

 # systemctl start beegfs-client.service
 # systemctl enable beegfs-client.service

Utilities

Install the package beegfs-utilsAUR.

Tip: Best to install on either the management server or client node, or both. For the purposes of this example the client node is used.

Check connectivity

Check the detected network interfaces and transport protocols from a client node with the following commands:

 # beegfs-ctl --listnodes --nodetype=mgmt --nicdetails 
   node01 [ID: 1]
     Ports: UDP: 8008; TCP: 8008
     Interfaces: 
     + enp0s31f6[ip addr: 192.168.0.1; type: TCP]
 # beegfs-ctl --listnodes --nodetype=meta --nicdetails 
   node02 [ID: 2]
     Ports: UDP: 8005; TCP: 8005
     Interfaces: 
     + eno1[ip addr: 192.168.0.2; type: TCP]
 # beegfs-ctl --listnodes --nodetype=storage --nicdetails 
   node03 [ID: 3]
     Ports: UDP: 8003; TCP: 8003
     Interfaces: 
     + eno1[ip addr: 192.168.0.3; type: TCP]
 # beegfs-ctl --listnodes --nodetype=client --nicdetails 
   4E451-5DAEDCBF-node04 [ID: 4]
     Ports: UDP: 8004; TCP: 0
     Interfaces: 
     + wlo1[ip addr: 192.168.0.4; type: TCP]

Server tuning and advanced features

InfiniBand Support

ACLs

Storage Pools

Quota Enforcement

High Availability

See also