Recent gains in processing power have enabled scientists and researchers to build world-class supercomputers from off the shelf components. Xsan makes it even easier to build clusters using this strategy since it lets you manage your storage devices collectively the same way you add workstations together to make the CPU cluster.
Network vs. direct
Computational clusters feature multiple computers running an application against a single large data set, which is typically hosted on a file server. Individual cluster nodes access the data using a network file system protocol, such as NFS. This limits scalability and performance, however. As in traditional data center environments, Xsan can remove these limitations, providing faster, more scalable data sharing across small and large clusters.
128 processors crunch one data pool
Xsan can support hundreds of Xserve nodes connected directly to the SAN via Fibre Channel. In a cluster of any size, all nodes can mount the same SAN volume and read directly from the same files with zero data replication. For small clusters working on very large data sets, this should provide a significant performance improvement over the traditional NFS file server. Data is never transferred over the network, which prevents LAN bottlenecks and avoids idle computing cycles while nodes wait for data to process.
Or Many, Many More
In a larger cluster, you would use your SAN as network attached storage. Its very easy to design this topology if youve set up head nodes to manage portions of the cluster. For example, a large cluster may use one head node for every 25 to 50 cluster nodes. Xsan would be ideal for data management at the head node. Each head node can connect to the storage pool via Fibre Channel HBA and Xsan. These nodes then distribute the data to their nodes using network file system protocols (NFS). This method for accessing and distributing data is faster and uses less network bandwidth across the cluster than typical network-attached storage solutions. At the end of the run, nodes return the processed data to the head nodes, which in turn write directly to the shared SAN volume.
Breaks the Mold, Not the Bank
Until Xsan, enterprise-class SAN solutions were too expensive to deploy with most computational clusters. Apples storage networking solutions Xsan, Xserve RAID, Xserve and Apple HBA let computer scientists design even more powerful supercomputers at reasonable costs.
Xserve RAID Storage Pool. Manage data centrally.
Metadata Controller. Directs servers to files in pool.
Head Nodes. Retrieve data over Fibre Channel.
Client Nodes. Process data received from head node.
Fibre Channel Switch. Gives high-speed access to storage.
Standby Controller. Provides high-availability insurance.
Xsan for HPC
Xsan is ideal for a supercomputer:
- Data Consolidation. Increases efficiency and scalability of your storage resources by aggregating storage pools and eliminates the need for storage provisioning.
- File-level Locking. Lets multiple head nodes connect to shared volume for improved performance and scalability.
- Direct Fibre Channel Connectivity. Reduces load on local area network or cluster interconnect technology, providing fast access to storage without bottlenecks.
- Metadata Controller Failover. Maximizes availability to ensure SAN operation in the event of a metadata controller failure.
- Flexible Volume Management. Makes it easy to scale SAN volumes by adding storage pools as capacity needs grow.