By now, the news is out about IBM’s mainframe-blade server hybrid play, the latest part of its zEnterprise initiative. Big Blue announced it in January with all the ruffles and flourishes one might expect from a “unified computing strategy.” Based on the coverage I read, the trade press heard “a solution for consolidating the Wild West of the Microsoft Windows server-based computing environment into the stolid and dependable mainframe data center,” which I suppose it might be.
Truth be told, IBM had been on point about consolidating distributed computing workload into the mainframe for some time; they’ve touted the lower cost and higher resiliency of using mainframe LPARs to host distributed computing workload as compared to x86-based consolidation using Tinkertoy server virtualization platforms such as VMware or Microsoft Hyper-V. We covered it here, noting the enormous potential benefits inherent in the strategy—but also a few potential limitations, including the recalcitrance of Microsoft wares to behave as LPAR guests and the need to manage distributed peripherals (Fibre Channel [FC] fabric and Network-Attached Storage [NAS], in particular) separately from mainframe peripherals and DASD. A hybrid architecture, networking the mainframe with IBM’s Windows-friendly blade servers, both of which have existed for quite awhile, provided a fix.
Integration of the platforms is accomplished using a Unified Resource Manager tool and Gigabit Ethernet network interconnecting the different rigs. Via the Unified Resource Manager, Simple Network Management Protocol (SNMP) information from the Power and Xeon blades is collected so the gear can be managed as another mainframe resource. Additional data is collected from the PowerVM and KVM hypervisors running on the blade servers so, in theory, the x86 hypervisors can be managed like virtual partitions in the mainframe itself.
Sounds good on paper, but it still leaves me concerned. I won’t geek out on the merits and demerits of using GigE interconnects vs. Infiniband, or my even larger concerns about a lack of management support for peripherals hanging off the backs of the blades. On the latter point, IBM says it’s working on its storage virtualization appliance to extend common management to FC fabrics as Scale Out Network Attached Storage (SONAS) is supposed to eventually do with NAS.
Actually, I have a completely different issue with this cobble. Why is management of the resource pool done via SNMP? I know SNMP has been around for a long time, and over the years I’ve seen real benefits derived from its use. Still, SNMP is a rather archaic protocol that requires software agents to be installed on monitored equipment that typically collect pretty rudimentary information from the monitored device and format it for collection on request by a Network Management System (NMS) or SNMP NMS. Rarely is SNMP used to actually do anything to the monitored device (other than to change the info being collected). It tends to be implemented in a very “passive” way.
This flies in the face of the work that IBM began crowing about in 2007 with respect to Web services and the REST protocol. Their Project Zero initiative pretty much denounced passive management techniques and presented REST (which stands for REpresentational State Transfer) as the future of management of both their SOA architecture and the next evolution of WebSphere. Not long ago, I heard a presentation by an IBM guy that championed REST-based management as the centerpiece of cloud computing because of the simple, yet capable, mechanism it provided for managing everything from software processes to hardware resources.
To see what you can do, management-wise, with REST that you would really be challenged to do with SNMP, go to CORTEXDEVELOPER.com. Storage-maker Xiotech put together this site to demonstrate the application of REST-based management to its storage rigs. It’s an eye-opener.
If applications, hardware, and clouds are all embracing REST and Web services standards for management, why isn’t IBM making this the centerpiece of its zEnterprise strategy, too? It’s a no-brainer to connect freestanding servers into a mainframe-centric network and to extend some centralized mainframe management to blades, but it would look a lot more “Mainframe 2.0” to also get out in front of the Web services/REST-ful management trend and show the world that mainframes are in fact the fundamental building blocks of clouds.
That’s my two cents.