2. New product: XtreemStore
Storage Management Software (HSM; Archive; Backup) for high
performance applications such as
Lustre
Web Services
Others
GRAU ArchiveManager (GAM) with built in Parallel HSM Filesystem
Parallel access through a Meta File System with POSIX interface to
unlimited amount of storage
Grid structure build on standard PC hardware
3. Need for high performance POSIX Archive
POSIX interface is very common to all applications
But Standard File Systems have limitations
ArchiveManager inherits those limitations when working with Standard
Filesystems
Major limits:
The amount of files within one File System
The throughput of one single File System
Throughput for huge files (10TB and more)
The integration of a parallel File System into GAM breaks these limits
XtreemStore
4. NFS / CIFS NFS / CIFSNFS / CIFS
Driving an unlimited amount of standard hardware
Software Architecture
Parallel File System
GAM Client
GAM Server
XtreemStore
6. Parallel POSIX Data Mover
Master
Data
Mover
Data
Mover
Data
Mover
Data
Mover
Data
Mover
Data
Mover
The parallel POSIX Data Mover is just software
It may run on
the source maschine
the target machine
or on dedicated computer nodes
The amount of streams running in parallel is not limited
7. HSM for Lustre with API – Overview
API Lustre Server
Primary Storage
Lustre
Clients
Parallel Data
Mover
XtreemStore
Archive Storage Disk / Tape
Master
Data
Mover
Data
Mover
Data
Mover
P / HSM FS
GAM Clients
GAM Server
Data
Mover
8. Status of integration with Lustre
Cooperation with Intel (earlier Whamcloud), CEA and GRAU DATA started
in May 2012
A Beta installation of XtreemStore plus Lustre and CEA code is planned
for April 2013 at the High Performance Computing Center (HLRS) at the
University of Stuttgart
Official availability of XtreemStore in combination with Lustre 2.4 inkl.
HSM – API is planned for June 2013
9. HSM for Fraunhofer Global File System (FhGFS)
FhGFS Server
Primary Storage
FhGFS
Clients
XtreemStore
Archive Storage Disk / Tape
P / HSM FS
GAM Clients
GAM Server
FhGFS Server
APIAPI
10. Throughput
A node based on standard hardware is expected to run a sustained data
rate above 200 MB/second.
XtreemStore scales almost linear with the amount of nodes
So the performance target for a system with 20 storage nodes is
expected with:
HSM - throughput per hour = 12TB
HSM - throughput per day = 250 TB
A system with 80 nodes should be able to run as much as
1 PetaByte per day