Most server vendors -- Dell, HP, IBM, etc. -- have a blade server strategy. Many view blade servers as processing engines to bring online, do a job, and then be re-purposed for another job, with each change of application involving a bare-metal restore.
Fujitsu Siemens Computer (FSC) approaches things a bit differently. It resells Egenera for its BladeFrame EX technology in Europe, the Middle East and Africa as FSC Primergy BladeFrame. Management software automates this blade re-provisioning process and make it responsive to changes in the processing load of the collection of server blade servers in the cabinet.
The idea is that server blades are CPU plus RAM and nothing else. The BladeFrame has extra hardware and software to assign disk, a network connection, operating system, middleware and application software to the raw server blades.
In effect a virtual server exists as a configuration file on a BladeFrame disk. When a new server is needed then this virtual server is instantiated on a real server blade which then becomes, for example, a Linux server running a procurement application. When that application is no longer needed the physical server blade is shutdown and its disk and network connection returned to the pool of unused resources. The blade itself is returned to the processing resource pool.
Egenera states that 'the BladeFrame combines the utility of stateless servers with powerful software that virtualises processing, storage and networking resources into a computing fabric'.
The BladeFrame is constructed as a rack cabinet with blade slots connecting to a backplane called, inevitably, the BladePlane. There are several kinds of blade:
- Processing Blade (pBlade) - FSC says this can have four single or dual-core x86 architecture CPUs and there can be 24 of them for a total of 96 CPUs. They can run Red Hat or SuSE Linux, Solaris 10 or Windows Server 2003.
- Switch Blade (sBlade) - Two switched-fabric interconnects, providing switching for internal blade-to-blade, TCP/IP or external traffic and do automatic load-balancing and failover.
- Control Blade (cBlade) - Two blades hosting the BladeFrame control software, PAN Manager. PAN stands for processor area network. These blades have connections to external devices via four-port GigE and four-port Fibre Channel adapters.
The BladeFrame uses external SAN or NAS as a disk resource and blades are booted from this external networked disk.
Four four-socket pBlades can be combined to form a 16-socket core.
The pBlades can be grouped together into logical processor area networks or LPANs. Each LPAN can be dedicated to a group of users or specific applications.
FSC pushes the idea of a dynamic data centre responding and adapting to a business's needs. At the end of the month, blades are scheduled to come into play and run payroll applications for example. Ones previously running a lower-priority procurement application are returned to the resource pool and re-provisioned in ten minutes or so to become a payroll server.
A BladeFrame administrator has access to the PAN Manager GUI and can set low and high water marks on a graphed display. If application response times goes above the high water mark then another pBlade can be provisioned to ensure that enough processing resource is bought to bear.
The pBlades are virtual servers but not in the VMware sense. They are real servers which are decoupled from NICs, host bus adapters (HBAs) and external storage so that they can be re-provisioned in minutes. There are virtual connections to operating system and application software, network connections and disk resource and these can be instantiated on real resource and the pBlade booted to turn a virtual server into a real server.
It takes longer to re-purpose blade servers with their own local disk, according to FSC and Egenera.
A potential disadvantage is that if an application runs out of memory and needs to page or swap files to disk, then that paging has a longer latency because it goes via the cBlade and through it to the external disk. Having several pBlades take up the BladeFrames's shared backplane bandwidth with paging activity could slow things down.
Blade server application
According to a Gartner Group report, BladeFrame is a good fit for server consolidation applications where there need to be a relatively large number of servers sharing resources and passing lots of data or messages between themselves.
Bernhard Brandwitte, FSC's server product marketing director, says that total cost of server ownership is lowered with BladeFrame because a customer doesn't need so many dedicated servers. Effective server utilisation goes up because when particular applications, like payroll, are idle between month-end times the servers running their work can be quickly and easily re-provisioned to do something else. Instead of computational silos you now have a shared computational resource pool.
High-availability is covered by configuring pBlades for automatic fail-over, and also by redundancy features in the BladeFrame itself.