From Troika Networks:
....Storage networking has emerged to allow any-to-any connectivity, or a network, between storage systems and servers. This framework increases the availability of storage resources and enables storage resource sharing among multiple servers. Typically, many paths exist between a server and a given storage device. A path consists of all the components between the server (host) and storage (target), including controller, cable, switch ports and internals, and controllers or other components on the storage device.
A "failed path" indicates that the connection is broken, which could be caused by failure of any component along the way. The availability of multiple paths indicates redundancy for many or all components, which means there are alternative ways to re-establish the connection. This introduces the capability for SANs to find alternative paths to a given target and automatically fail over when a
path is interrupted....
......Dynamic multipathing (or DMP) is a concept that addresses two important and related, yet distinct features. They are automatic failover and dynamic load balancing. Multiple independent paths from a server to a storage device make it possible to automatically recover from a path failure. Automatic failover refers to any technique used to implement such a recovery. Outstanding input/output (I/O) requests can be retried and new requests can be issued on the surviving paths. Duplicating paths and tolerating the failure of one path increases the overall reliability of the SAN beyond the reliability of any one component.
Performance degradation or partial inaccessibility can result when a given path has not failed but is overloaded. Even if the overloading of one path relative to another path does not result in "virtual inaccessibility," it usually causes less than optimal response time on I/O requests sent over the highly loaded path. Load balancing is a concept that in storage networking addresses distributing the load of I/O traffic among all available paths from servers to storage systems, under non-failure conditions. The objective is to reduce the average response time, and use the enormous bandwidth potential available
in a SAN......
.....A solution that encompasses both host and target-side failover is clearly superior. With such an implementation, failures can be tolerated at any point along the path between application and physical storage device. This includes fabric equipment, RAID controllers, and other components whose failure or manual removal can result in loss of use in more traditional host-only failover implementations. In such a system, the total number of available paths and the potential to overcome a wide range of failures increases exponentially, which in turn boosts data availability to the same degree.....
troikanetworks.com
From EMC:
....The roots of dynamic load-balanced multipathing and channel failover technology — commonly known as dynamic path selection (DPS) and dynamic path reconnect (DPR) — originated in the 1980s mainframe environment. When UNIX servers began sharing many tasks performed by mainframe systems, the need for a similar high-availability technology became apparent. UNIX server vendors attempted to meet this requirement with primitive channel failover features built into an operating system. These early offerings eliminated single points of failure for I/O paths, but did not provide scalability, application tuning, ease-of-use and maximum uptime.
EMC recognized the opportunity to bring mainframe-like capabilities to the UNIX and Windows NT server environments with PowerPath. With SCSI protocols as the mainstay technology in open systems, the PowerPath architecture could manage these protocols in both SCSI and Fibre Channel
physical interconnects. The Small Computer System Interface (SCSI) protocol connects multiple peripheral devices to a single controller via a common cable. All I/O in this environment follows a single path. A protocol layer, cabling, host bus adapters (HBA) and drivers make up the SCSI interface.........
Advanced Applications
PowerPath brings advanced applications to the Fibre Channel and clustered environments, and to segmenting paths for environments with more than four paths. PowerPath in a Fibre Fibre Channel technology provides flexibility to configure a topology to meet any requirement for Channel Environment extended distance, storage consolidation and expanded storage capacity. Additionally, Fibre Channel meets high-availability requirements. Fibre Channel configurations consist of a physical topology and a logical topology.
a) Physical topology includes physical interconnects for a host computer, a Symmetrix storage system and other Fibre Channel components, such as hubs and switches.
b) Logical topology describes the logical paths established between the operating system device names and their associated Symmetrix ports and volumes.
PowerPath works within all three physical topologies — switched fabric, arbitrated loop, and point-to-point connections — to enhance high availability and I/O congestion avoidance in a Fibre Channel configuration.
• High availability: Achieved within the Fibre Channel storage area network (SAN) by configuring dual paths between connections, configuring alternate paths to SAN components, and deploying redundant SAN components.
Some SAN switch components (such as the EMC Connectrix Enterprise Storage Network System) feature redundant subsystems to ensure high availability and a reliable fabric.
• Congestion avoidance: SAN congestion can develop when multiple computer servers send I/O into a single SAN component, especially when the SAN component is part of a fan-out or fan-in strategy for storage consolidation or expanded storage capacity.
By monitoring each path’s throughput and responsiveness within the host computer, PowerPath can redirect new I/O away from congestion points, as detected by an unresponsive path (assuming an alternate path exists on the host and in the SAN to redirect the I/O). The host computer may not cause the SAN congestion that redirects I/O away from the congestion with the use of PowerPath. Still, PowerPath improves the I/O performance on the host computer and enhances SAN traffic patterns by avoiding bottlenecks.....
.... Typical switch or hub configurations make management of computer server and application performance increasingly difficult because I/O performance tuning must accommodate the varying loads of all the computer severs within the storage area network. Only a time-intensive analysis of
the environment could statically allocate loads to avoid SAN congestion.
PowerPath intelligently calculates the pending I/O queue depth for each channel, and then balances with PowerPath the new I/O load across the channels on each server (Figure 13). This load balancing process continues throughout the fabric or loop, with PowerPath adjusting path selection for each server based on the performance of each channel. No matter how many computer servers exist in the fabric or loop, PowerPath can continually adjust the I/O loading without interrupting computer server applications.....
emc.com
As you know, Compaq is using Troika. PowerPath is already EMC's bestselling software product with over 25,000 licenses sold and the possible early release of the SAN version of PowerPath sets up everything that EMC wants to do in virtualization, IMO. EMC started working with Conley Software in 1997 and worked with them on the PowerPath product before deciding to acquire it in August 1998. |
