- HPC Buyers Guide
Detailed Buyers Reference- Applications
- Budget
- Processors
- Interconnects
- Memory
- Distributions
- Cluster Management
- Warranty and Support
- Cluster Racking
- Cluster Logical Layout
- Master Node
- Storage
- Nodes
- Power
- Cooling
- Commercial Software
- Schedulers
- Account Management
- Security
- Windows Integration
- Shipping and Delivery
- On-Site and Training
- Next Steps
HPC Applications- Processor
- System
- Network
- Storage
- Power
- Cooling
- Facilities Estimator
- Remote Management
- Green Computing
- Visualization
- Compilers
- Operating Systems
- Schedulers and Tools
- Virtualization
Backup
What are your backup options?
Data backup is a needed part of any HPC cluster deployment, however the size of today's storage solutions can make that problematic. Aspen Systems makes disaster recovery images of your cluster when it is shipped. These images do not contain your user data in most cases, such as /home or /data, but instead are meant to rebuild the cluster should a catastrophic hardware or software failure occur. You may also contract with Aspen to store additional disaster recovery images if the cluster environment is changed in any significant way. These images are normally smaller than 20 Gigabytes, and often less than a gigabyte for a compute node. So, where does that leave your data in event of a total RAID failure or other catastrophic failure? You need a backup solution for at least some of your most critical data, and there are several options you might consider.
Backup Hardware
Often, larger organizations will already have an enterprise-wide backup solution, and your new cluster can be licensed or configured as a client to the existing resource. If that resource does not exist, Aspen can provide your cluster with Digital Linear Tape (DLT), Advanced Intelligent Tape (AIT), and Linear Tape-Open (LTO) tape drives and tape drive libraries, depending on your needs. Tape capacities range from 160 to 800 GB per tape, with write speeds from 12 MB/s to 120 MB/s. Optical worm drives are available as well.
Almost all of the single tape drive solutions can be mounted as internal devices in your master or storage node(s) if they are 3U or 4U chassis, while tape libraries are normally configured as rack mount units or on rack shelves. Internal drives are normally Small Computer Systems Interconnect (SCSI) interfaces, while external drives can be SCSI or Fiber Channel.
You may need to decide what data on your cluster is critical and back up only that data. Often, saving only results or application output data can significantly reduce your backup requirements, allowing even a single tape drive of the right type to be used for your backup needs.
Aspen can provide your system with larger tape drive libraries, which can back up your entire system, including your data storage. Some of these systems can archive petabytes of data, speak to your Aspen sales engineer about your specific backup needs.
Backup Software
You have many choices for your backup software solution as well. If you wish an open source solution, Aspen recommends Amanda or Bacula, and Aspen offers Storix, IBM Tivoli, and Veritas Backup as commercial products. Amanda is command line based, while Bacula has X windows clients and a command line text console user interface. The commercial products all offer Graphical User Interfaces (GUIs) as well as command line based options. Veritas, now owned by Symantec, is arguably the most widely deployed commercial backup solution, while IBM Tivoli Storage Manager is ideal for an IBM hardware based cluster. Storix is a relatively full featured and price competitive commercial offering, and has a web user interface.
As data sets become larger and larger, full backups of all data spaces becomes problematic, forcing some organizations to rely only on their RAID systems, with no additional backup capabilities. In these cases, it is critical that hot spares be configured on all RAID sets, notification to your administrators and to Aspen be configured and routinely tested, and that the RAID system(s) be located on conditioned power sources.
