2024年5月11日发(作者：尾初柔)
Technical bulletin: Remote visualisation with VirtualGL
Dell/Cambridge HPC Solution Centre
Dr Stuart Rankin, Dr Paul Calleja, Dr James Coomer
© Dell
Introduction
This technical bulletin is a reduced form of the whitepaper entitled “Remote visualisation using open
source software & commodity hardware”. It is aimed at High Performance Computing (HPC) users
and administrators who wish to develop remote visualisation solutions using open-source software
and commodity hardware. The whitepaper provides background reading as well as a command-line
configuration guide for those wishing to recreate the solution.
It is commonplace today that HPC users produce large scale multi-gigabyte data sets on a daily basis
and that these data sets require interactive post processing with some form of real time 3D or 2D
visualisation. The traditional HPC workflow process requires that these data sets be transferred back
to the user’s workstation, remote from the HPC data centre over the network. This process has several
disadvantages, firstly it requires large I/O transfers out of the HPC data centre which is time consuming,
and also it requires that the user has significant local disk storage and a workstation setup with the
appropriate visualisation software and hardware.
The remote visualisation procedures described here remove the need to transfer data out of the
HPC data centre. The procedure allows the user to logon interactively to the Dell | NVIDIA remote
visualisation server within the HPC data centre and access their data sets directly from the HPC file
system and then run the visualisation software on the remote visualisation server in the machine room
sending the visual output over the network to the users remote PC. The visualisation server consists of
a T5500 Dell Precision Workstation equipped with a NVIDIA Quadro FX 5800 configured with an open
source software stack facilitating sending of the visual output to the remote user.
By deploying this methodology users can gain direct visual access to their HPC data wherever and
whenever they require it. Also it reduces the need to transfer large scale HPC data over the network and
removes the need for users to configure complex and expensive visualisation software and hardware
on their local machine. In this technical bulletin we summarise work performed with an OS-neutral
extension of familiar remote desktop techniques using open source software for remote visualisation.
Whilst traditional X11/GLX over SSH methods may require very high quality end-to-end network
connection and additional hardware resources at the local workstation or PC, remote-desktop software
limits the required network bandwidth by offloading pixel rendering to the server but is not designed
to support 3D applications with hardware acceleration. VirtualGL (VGL) is open source software
which addresses these problems by separating off the 3D rendering task and invoking the hardware-
accelerated graphics system on the remote server to produce the required images. The resulting (2D)
images are then fed to the target (local) X display with the rest of the 2D drawing instructions. The key
advantages of this are (i) any 3D image which can be rendered by the server's graphics hardware and
software can in principle be rendered by VGL, and (ii) since only 2D commands are ultimately sent to the
X display, the destination display need have no native 3D capability at all.
In the following sections we summarise various implementations of VirtualGL in combination with
Dell and NVIDIA hardware and other widely available software to demonstrate the effectiveness of
commodity systems in supporting remote visualisation solutions.
1 Dell | Remote visualisation technical bulletin
2024年5月11日发(作者：尾初柔)
Technical bulletin: Remote visualisation with VirtualGL
Dell/Cambridge HPC Solution Centre
Dr Stuart Rankin, Dr Paul Calleja, Dr James Coomer
© Dell
Introduction
This technical bulletin is a reduced form of the whitepaper entitled “Remote visualisation using open
source software & commodity hardware”. It is aimed at High Performance Computing (HPC) users
and administrators who wish to develop remote visualisation solutions using open-source software
and commodity hardware. The whitepaper provides background reading as well as a command-line
configuration guide for those wishing to recreate the solution.
It is commonplace today that HPC users produce large scale multi-gigabyte data sets on a daily basis
and that these data sets require interactive post processing with some form of real time 3D or 2D
visualisation. The traditional HPC workflow process requires that these data sets be transferred back
to the user’s workstation, remote from the HPC data centre over the network. This process has several
disadvantages, firstly it requires large I/O transfers out of the HPC data centre which is time consuming,
and also it requires that the user has significant local disk storage and a workstation setup with the
appropriate visualisation software and hardware.
The remote visualisation procedures described here remove the need to transfer data out of the
HPC data centre. The procedure allows the user to logon interactively to the Dell | NVIDIA remote
visualisation server within the HPC data centre and access their data sets directly from the HPC file
system and then run the visualisation software on the remote visualisation server in the machine room
sending the visual output over the network to the users remote PC. The visualisation server consists of
a T5500 Dell Precision Workstation equipped with a NVIDIA Quadro FX 5800 configured with an open
source software stack facilitating sending of the visual output to the remote user.
By deploying this methodology users can gain direct visual access to their HPC data wherever and
whenever they require it. Also it reduces the need to transfer large scale HPC data over the network and
removes the need for users to configure complex and expensive visualisation software and hardware
on their local machine. In this technical bulletin we summarise work performed with an OS-neutral
extension of familiar remote desktop techniques using open source software for remote visualisation.
Whilst traditional X11/GLX over SSH methods may require very high quality end-to-end network
connection and additional hardware resources at the local workstation or PC, remote-desktop software
limits the required network bandwidth by offloading pixel rendering to the server but is not designed
to support 3D applications with hardware acceleration. VirtualGL (VGL) is open source software
which addresses these problems by separating off the 3D rendering task and invoking the hardware-
accelerated graphics system on the remote server to produce the required images. The resulting (2D)
images are then fed to the target (local) X display with the rest of the 2D drawing instructions. The key
advantages of this are (i) any 3D image which can be rendered by the server's graphics hardware and
software can in principle be rendered by VGL, and (ii) since only 2D commands are ultimately sent to the
X display, the destination display need have no native 3D capability at all.
In the following sections we summarise various implementations of VirtualGL in combination with
Dell and NVIDIA hardware and other widely available software to demonstrate the effectiveness of
commodity systems in supporting remote visualisation solutions.
1 Dell | Remote visualisation technical bulletin