Bright Cluster

Home

Products

Sales

Support

Data Sheet

White Paper

Cluster I/O Challenge

Bright Cluster Pulse

any-to-many
high throughput data replication

Rapid data replication is one of the most critical components of data-intensive distributed computing. Replicating data or making data available to multiple systems simultaneously has however been complex, slow and unreliable.

NFS mounting shares on each node results in numerous connections being made simultaneously to the file server for the same piece of data. The thrashing this causes on the file server and network will bring data transfer to a grinding halt. To solve this bottleneck the next step is usually to purchase ever more expensive high bandwidth file servers and networks. The cost of these solutions usually become the primary constraint on computational ROI and scalability. To lessen this burden extensive data staging is often implemented which is tedious, time consuming, and prone to error that continues to limit computational production and ROI.

Parallel file systems once appeared to be the magic solution but reality has proven these to be extremely complex, temperamental and unreliable.

Pulse provides a simple, easy solution for any-to-many data replication.

No matter how many nodes need data, Pulse makes a single unicast connection to the data source that allows for any file server to move data at maximum speed. Data arriving at the Pulse Server is then sent non-stop to the entire group of target systems via multicast. Each target system receives the data simultaneously.

Pulse replaces NFS by transferring data directly to the local node hard drive. Applications can then access the data as they need it with certainty and speed from the local disc.

Pulse replaces expensive high bandwidth storage and networks by using a single connection to the data source which means no thrashing and a streamlined data flow to the destination. Pulse let's you use existing storage or low cost file servers.

Pulse replaces tedious data staging by providing true end-to-end data transfer. Pulse accesses any network reachable data source and replicates data directly to the end destination.

Bright Cluster™ Pulse™ delivers data as fast a as a cluster can absorb. The Data Absorption Rate of a cluster is simply the speed that a node can accept data. Small files can be absorbed at full network speed because they fit in a node's memory buffer. Files or directories of files larger than the memory buffer though must begin writing to the local hard drive before all the data can be absorbed. As the write speed to a hard drive is much slower than the network transfer speed the result is a data jam that causes thrashing and packet loss in the data stream. Pulse regulates the data flow below the absorption threshold resulting in a smooth data stream that delivers the maximum possible aggregate throughput.

Pulse shell commands can be easily scripted to integrate seamlessly with your scheduling environment for automated workflow operation.

Web interface provides status, administration including creation of target node groups, transfer history, and data transfer operation.

Same look and feel simplicity as our web site.

Pulse Server is a 1U appliance system that connects directly to your Ethernet network. Pulse Server is the physical engine that moves your data. Small agent applications are installed on all target nodes to receive data. User interface with the Pulse Server is via http whether by Web GUI or command line interface.

Pulse Server can access any network reachable data source that is accessible via NFS, SMB, HTTP, or FTP protocols.

When commanded to transfer data Pulse Server makes a single connection to the data source and initiates transfer. As the data packets arrive at the Pulse Server they are sent non-stop to the target group via UDP/IP multicast.

Guaranteed Data Delivery – Pulse is fault tolerant & provides for MD5 sum verification of data transfer.

Compatible – Pulse operates over standard network protocols and only in user space. Supported operating systems include Linux distributions: Red Hat, SuSE, Debian, and Fedora

Ethernet – Pulse works over commodity Ethernet and standard switches.

Pulse is simple, easy performance.