Advanced Server Clusters

Hewlett-Packard has coined a new phrase in network technology. This white paper details its concept of Advanced Server Clusters

Hewlett-Packard has coined a new phrase in network technology. This white paper details its concept of Advanced Server Clusters

Introduction

Advanced Server Clusters, or ASCs, are simply networks of servers connected via an advanced low-overhead, high-performance interconnect. Servers linked in an Advanced Server Cluster configuration can communicate with each other much faster and with far less server overhead than traditional LANs, and thus can provide excellent benefits for demanding applications in areas such as:

( High availability clusters, specifically with network-based failover solutions (for example, Vinca and NSI)

( Scalability clusters utilising high-capacity parallel database solutions (for example, Oracle(r) Parallel Server)

( Application clusters for three-tier client-server applications

( Large-scale data transfer, including local network-backup solutions and loading data marts

This white paper provides IT managers with a high-level overview of Advanced Server Clusters, with an emphasis on the benefits ASC provides for Microsoft Windows NT applications. If you have demanding requirements in the above types of application areas, then you may find it useful to review this white paper.

A closer look

Hewlett-Packard's interest in Advanced Server Clusters is based upon an objective assessment of how customer needs are evolving in two key areas:

Increasing demand for high availability: Businesses increasingly require applications and data to be available 24 hours a day, seven days a week. To ensure minimal interruption to application and data availability upon hardware failure, several application vendors provide "failover" capabilities via file mirroring or disk mirroring solutions. These solutions also referred to as "vaulting", "shadowing" or "electronic journalling", copy data over a network to maintain a "mirror" on another server. For large installations that use traditional LAN technologies, mirroring results in substantial network traffic and in server processing overhead. This overhead negatively impacts both application performance and high availability functionality.

Thus, a low-overhead interconnect will increasingly be required by the most demanding network-based failover installations.

Increasing need for capacity: Applications such as data warehousing illustrate the continually increasing needs for additional performance and capacity. Computer vendors traditionally have addressed such needs by simply providing larger servers with more processors and more storage. However, increasing the number of processors in a server presents scalability issues for both the operating system and the server hardware.

Based on extensive experience with multiprocessor systems, HP believes that over the near term, traditional symmetric multiprocessor servers utilising Windows NT will effectively scale to a maximum of perhaps 16 processors. A solution for demanding applications that require even more capacity is a topology of multiple servers that are tightly linked via a low-overhead communications interconnect.

High performance, low-overhead interconnects have been available on mainframes and UNIX systems to meet customer demands for expanded capacity and high availability. However, such products and technologies are not applicable for Windows NT environments, either because they are not supported under Windows NT (for example, NUMA), or they utilise expensive proprietary technologies that are inconsistent with the Windows NT paradigm of low-cost, open, multi-platform computing.

What's the difference between an ASC, a Server Area Network, and a Storage Area Network?

Advanced Server Cluster is a term used by HP to refer to a tightly linked network of servers connected via an advanced, high-performance, low-overhead interconnect. Elsewhere in the industry, such configurations may be referred to as Server Area Network, or System Area Network, or SAN. However, the "SAN" nomenclature generates confusion as that term is increasingly used to instead refer to "Storage Area Network". A Storage Area Network represents an entirely different technology, which provides a specialised interconnect between servers and storage systems. HP has adopted the term ASC in an attempt to avoid confusion with "Storage Area Network", and to more accurately reflect the actual usage of this new server-to-server interconnect technology.

What is an Advanced Server Cluster?

An Advanced Server Cluster is a tightly linked network of servers connected via an advanced, high-performance interconnect. Compared with LAN technology, an ASC provides:

( Dramatically reduced communications overhead, thus enabling both ultra-fast message passing and improved server performance (as "more CPU" is available for other work)

( A high-bandwidth interconnect, which has the capacity to transfer large amounts of data much faster than traditional LANs

An ASC configuration can contain as few as two servers, and in the near term there will be many deployments of ASCs consisting only of a few clustered servers. However, ASC architecture can support much larger clusters distributed throughout a building or across a campus. The underlying technology that enables the low-overhead communications capability of ASCs is the Virtual Interface Architecture (VI). VI is a new standard for advanced server-to-server communications, which is rapidly being adopted for Windows NT products. VI enables applications to bypass the high-overhead TCP/IP stack when passing messages over an ASC.

With Windows 2000 server, most applications will utilise the basic capabilities of ASCs and VI without modification through the same Application Programming Interface (API) set, known as WinSock, that they can use to access the LAN. Presently, application providers (ISVs) can choose to certify their applications to prove functionality when used in an ASC/VI environment. Application vendors whose products realise the most dramatic benefits from an ASC will often modify their applications to utilise the VI Application Programming Interface (API) set, with the result being approximately a 100X reduction in CPU overhead for message transmission. The combination of standards-based Windows NT servers, and the new VI architecture standard for server-to-server interconnects, makes it possible for practical, low-cost ASCs to be offered for Windows NT servers.

Advanced Server Clusters: applications and benefits

Various types of applications will derive substantial benefit from ASCs. Examples of four such applications follow.

High availability clusters utilising network-based failover solutions

Several popular high availability solutions (for example, Vinca and NSI) rely on mirroring files or databases over a network. Such "network-based failover solutions" require constant passing of messages and data between servers, such that the resulting overhead can impact both server performance and failover / resynchronisation time. Thus, a low-overhead interconnect can help maximise server performance, and optimise overall solution effectiveness.

Scalability clusters utilising parallel database applications

Parallel database applications enable multiple servers to be linked together to provide a higher level of database capacity than could be supported on a single server. Such products are increasingly popular for large-scale decision support and OLTP applications. Leading parallel database technologies for Windows NT include Oracle Parallel Server (OPS), IBM DB2 Universal Database Enterprise Extended Edition (DB2 UDB EEE), and Informix Extended Parallel Server (XPS). In order to scale effectively, parallel databases require constant, time-sensitive message passing between servers, allowing multiple servers access to the database while maintaining database integrity. A low-overhead, high-performance interconnect is an ideal technology to make such solutions practical and scalable. ASC is expected to be very popular with customers implementing parallel databases.

Applications Clusters utilising three-tier client-server applications

Typical three-tier client-server applications require fast, reliable communications for transmitting data between application servers and database servers.

These applications are often performance-sensitive. The performance of both application servers and database servers can be enhanced via a low-overhead interconnect, which helps ensure that processing power is utilised for the application, not for communications overhead.

Enterprise Resource Planning (ERP) solutions (and many others) often utilise a three-tier architecture, and such applications can realise better response time and/or higher throughput via ASC solutions. Figure 4 shows an example of an ASC utilised for application server to database server communications.

Network-based file and database backups

There is an ever-increasing need to reduce the time window for regular backups of files and databases. For customers who utilise network backup solutions, a high-bandwidth, low-overhead interface to backup servers can enable a substantial reduction in backup time, and can also improve performance of production machines by reducing CPU overhead.

Note that ASCs can specifically provide an improved solution for network backup of servers that are in relatively close proximity (for example, a computer room)

In addition to the four examples we have reviewed here, there are various other applications that will also benefit from ASC. For example, ASC interconnects provide a low-overhead, high-bandwidth "pipe" for bulk data transfer, such as loading a data mart. ASCs are also applicable for technical computing by utilising the Message Passing Interface to support clustering of workstations and/or servers for scientific and engineering applications (for example, ME/EE design simulation and analysis).

Conclusion

ASC technology provides real customer benefits in areas including scalability clustering, high availability clustering, applications clustering, and data backups. If you have demanding requirements in such application areas, then you may want to further consider ASC technology.

The full text of this white paper can be found on the Hewlett-Packard website. Compiled by Richard Pitt

Read more on Server hardware

SearchCIO
SearchSecurity
SearchNetworking
SearchDataCenter
SearchDataManagement
Close