6
Cray DVS
6.1
Introduction to DVS
The Cray Data Virtualization Service (Cray DVS) is a distributed network service that provides transparent access
to file systems residing on the service I/O nodes and remote servers in the data center. Cray DVS provides a
service analogous to NFS™. It projects local file systems resident on I/O nodes or remote file servers to compute
and service nodes within the Cray system. Projecting is simply the process of making a file system available on
nodes where it does not physically reside. DVS-specific options to the
mount
command enable clients (compute
nodes) to access a file system projected by DVS servers. Thus, Cray DVS, while not a file system, represents a
software layer that provides scalable transport for file system services. See the
mount(8)
and
dvs(5)
man
pages for more information.
Cray DVS uses the Linux-supplied VFS interface to process file system access operations. This allows DVS to
project any POSIX-compliant file system.
Figure 30. Cray DVS Use Case
/home
(NFS)
Input Files
Small Data Files
Applications
Lustre
/scratch
User
Application
DVS Network
Service
Shared Data File
Input Files
Large Data Files
Applications
Compute Nodes
/home
,
/gpfs
,
/scratch
/gpfs
Cray DVS provides I/O performance and scalability to a large number of nodes, far beyond the typical number of
clients supported by a single NFS server. Operating system noise and impact on compute node memory
resources are both minimized in the Cray DVS configuration.
IMPORTANT: DVS servers use unlimited amounts of CPU and memory resources based directly on the
I/O requests sent from DVS clients. For this reason, DVS servers should be dedicated and not share
nodes with other services (Lustre nodes, login nodes, etc.).
Cray DVS
S3016
140
