Blade Server Standards
Principal Consultant Technical Marketing, ZNA Communications
copyright (c) 2003, John Harker
Vendors in the blade server market today need to face and embrace a hard reality. Open Systems-based standardization helps sell product. Currently standards are still evolving for blade servers and that means data center managers face a wide range of incompatible choices. Customers are reluctant to install proprietary architectures into their corporate systems because they fear limited choices for add-ons and technological dead-ends. They also want compatibility so they don't have to learn entirely new methods or buy new equipment or software. The hanging question is "Is the market going to break out of this proprietary standards mode soon and how long will blade platforms remain OEM specific". Proprietary designs are a higher price/margin play but for a variety of reasons will continue to lose volume to cheaper rack mount standard PC systems. This article seeks to explore areas of standards appropriate for Blade Servers whose adoption would move blade architectures to the actuality and perception of being open standards based.
What Standards Apply to Blade Servers?
Open Standards design possibilities exist at several levels in blade systems. One is hardware form factor and bus designs. Another is remote blade management, both hardware and software. Another is a common distributed applications environment. Using open standards for blades is not an all-or-nothing affair. Standards can be used in some areas, and proprietary designs in others.
Here are some standards areas that apply:
Bus Standards and I/O standards - These are covered elsewhere in this newsletter in the article on "Interconnect Technologies in Servers". Currently, customer perception of a lack of open standards and component interchangability in this area continues to dampen the blade server market. The argument for standardized bus and I/O designs is clear to all. The issues are largely centered on which standards – to which, because different design points do exist, the answer is and may remain more than one. But to enable what customers want - a variety of vendors selling compatible CPUs, I/O devices, and high-speed network switch interconnects - there is still benefit attached to using standards based designs such as CompactPCI, PCI-X, PCI Express or AdvancedTCA (PICMG 3.x).
High Speed Interconnect Standards – Likewise covered in the "Interconnect Technologies in Servers" article. Areas of agreement on standards are emerging however – TCP/IP sockets are supported now over both high-speed Ethernet and Infiniband connections. Also RDMA (remote DMA) looks like a candidate for adoption.
Management Standards - The argument for standardized management is from the point-of-view of the IT customer, who wants to manage all his/her systems – standalone, rack mount bricks, blades, all using the same tools. In the data center or hosting center manageability has long been a key issue for customers. The management system is likely to have been chosen and already be in place when the blade system is selected and installed. If the blade server management and application environments are standards-based then common management becomes practical. In this customer view, blade servers are just a compatible variant of a standard modular computing environment, as espoused by Intel and other vendors. Then, on a level playing field of applications and management compatibility, blade servers can compete on the basis of the real advantages they have – lower costs, simplicity, smaller footprint, smaller power requirements, and so on.
Hardware Management Standards – Blades don’t have individual consoles or installation devices, so they require remote installation, provisioning and management capabilities. And because blades are usually hot-swappable, administrators must be able to quickly discover new blades, identify the proper configuration and allocate the necessary images to the blades. Blade servers are acting as a catalyst for making these remote capabilities a requirement for IT customers. But this is not a call for different standards and products. Managing blades shouldn't vary much from managing typical servers where users need to monitor availability and performance, as well as spot potential hardware problems before they cause downtime.
One Standard that is useful in this area is Intel’s Intelligent Platform Management Interface (IPMI). IPMI defines a common interface to how vendors monitor their system hardware and sensors (temperature, voltage, fan, etc.), control system components (power, blades, etc.), log important system events (chassis intrusion, CPU performance, etc.), and to allow administrators to remotely manage and recover failed systems. Promoted by Intel, Dell, HP and NEC, IPMI standardizes the way embedded management subsystems communicate with systems such as, the CPU, other embedded management subsystems and remote management applications. IPMI also complements and incorporates other management standards like CIM, DMI, WMI, ASF, and SMBIOS.
SNMP and MIBs are well known IETF standards which I won’t describe here. CIM, DMI, and SMBIOS are Desktop Management Task Force (DMTF) standards. CIM is a standard for a common information model that features object friendly data content format that can be expressed in XML. DMI is a component instrumentation interface for workstations and servers. SMBIOS (System Management BIOS) is the standard for management extensions to the BIOS interface on Intel-based architectures). Another related standard developed by the DMTF is ASF (Alert Standard Format) which is an evolution of IBM and Intels Alert on LAN (AoL) standard for status and control of devices before an O/S is loaded. And the DMTF is working on developing a DEN (Directory Enabled Network) standard so management services can use standard distributed directories (via LDAP).
Other related standards incorporating the DMTFs work include the open source movement driven WBEM (Web Based Enterprise Management) initiative and Microsoft's WMI (Windows Management Instrumentation) standard.
Software Managagment Standards - Software Management products for blades falls into a variety of categories such as: change and configuration management; image cloning and management; provisioning; application monitoring and control, and policy-based management efforts. Policy-based management is where systems can automatically deploy predefined actions (such as for configuration, security, provisioning and performance) across servers and other networks elements in response to events, triggers, or administrative commands. At a low level a variety of standards exist and are used for these applications, such as the DMTF standards, SNMP and syslog, but at higher levels there is little standardization. Instead what may be most appropriate and is happening is for the major cross-platform management tools to add blade management capabilities.
Distributed Environment Standards - Applications are what sell systems, and commercial applications follow volume platforms. Just as with hardware there may be more than one such environment, but the systems software for the blades should provide for common standards-based distributed environment support. For example it appears that the web services standards community (WW3, DMTF, Oasis) and the grid computing standards community (Globus) are standardizing on a common set of Internet based standards for distributed services (IPC, directory, security, data formats, events). These will be important for both remote management and for commercial application support.
What is still needed
One thought is that there is a need for a common architecture and lexicon standard. Just as
an example, before SNIA moved into the standards arena, they first cemented a taxonomy and framework for storage interoperability. A similar first step for the blade industry would provide a common ground that does not impinge on the proprietary approaches of various vendors, but is useful in moving forward with solving the customers problems with proprietary designs.
After that the SBTA could work on adopting or creating sets of standards in the areas mentioned above. We should agree there are different bus, I/O and interconnect standards appropriate for different design points, and adopt a catalog of them. We should adopt the DMTF standard set and the Intelligent Platform Management Interface and work to modify them if needed. We should adopt and support distributed environment standards. We could also emphasize recommended practices in hardware areas outside of the scope of bus, I/O, and hardware management standards. Examples would be areas such as mechanical systems, system power management, rack setup and design, EMI shielding, mechanical design issues, heat dissipation and air conditioning, signal integrity, and power failure issues.
In sum, in the long run, blade vendors will best be served if there is a convergence on interoperable standards for platform hardware and system software along the lines of a common extension of the standard PC platform. This will remove a major barrier to sales for many customers and allow users to easily integrate blade servers with existing systems setups - for new and different functionality and for lower ownership costs.