Our I.S. Systems Support team at BIDMC is moving its HP/UX infrastructure to Red Hat Enterprise Linux. The success from our first implementation achieved high marks in all aspects: improved availability and performance, an inexpensive capital entry point with lowered operating costs, and users did not notice a disruptive change with the new platform. When asked if we could change anything in the new architecture, we could only come back with one response: a proliferation of rack-and-stack implementations will only greatly increase our systems management complexity, because it does not provide enough hardware management and monitoring required for enterprise-class applications — all of which was integrated as part of HP/UX server technology. Thus, we are engaged with IBM to determine if their blade systems can meet those needs:

  • Superior hardware management and monitoring from its Advanced Management Module over ILO and other IPMI cards
  • Eliminate KVM cabling: concurrent KVM with role-based access controls, with improved remote management access using a Java-based applet instead of an IE-only ActiveX control
  • Eliminate server I/O cabling: embedded LAN/SAN switches map to the variety of I/O cards
  • Eliminate internal storage: which greatly adds to TCO when running disks within enterprise servers — using boot-from-SAN management instead reduces power, cooling, and complexity
  • Improve longevity: as improvement in processors and chipsets are introduced, it can be easily introduced by swapping blades out — leaving the chassis and its infrastructure intact. No more duplicate infrastructure requirements as newer hardware is introduced on the datacenter floor.

But even with all of these incremental improvements in place, the “burning” question that remains to be gauged is can bladecenter be powered and thermally controlled within the datacenter. Escalation in electrical and cooling costs has made for a more acute provisioning process that is steered by reducing hardware per application whenever feasible. Our early numbers from bladecenter suggest there is already some improvement over an equivalent rack server, with room for much more when we leverage any of the software throttling and virtualization techniques available to us.

With this significant reduction in cabling, it leads to easy replacement of any of bladecenter’s internal parts. Everything that inserts into the chassis is hot removable. Our Windows Server Team is quick to point out that a hard disk on a blade is not hot removable, but that only adds to our goal of a disk-less blade configuration.

Our Network Services Group has endeared themselves with CISCO-branded technologies throughout their infrastructure. But while CISCO provides their product to the major blade technology vendors, NSG refuses to allow that switch into the topology. So we have instead settled on trying IBM’s Server Connectivity Module, which extends a 6-port uplink to any or all of a blade’s NIC — we have four NICs per blade, so there are four SCMs installed in the chassis. In theory, we have diminished bandwidth capacity, but at the very least, it reduces the copper cable count from 56 (14×4) to 24 (6×4). And given the density of the blade form factor, this reduction is most welcomed.

Our Storage Team has trunked the bladecenter’s CISCO SAN switches into their topology with no issues, with the exception that the Advanced Management Module is not reporting the blade’s HBA WWN properly — an aesthetic we can certainly live without, and we get to test resolution of bug reporting. Only two pair of fiber cables (with four empty pair left for expansion) are required to allow for the 28 HBA cards to connect up into the various EMC storage arrays — with an additional 28 HBA cards left for expansion. That is a lot of I/O capacity at our demand within a dense form factor — and with minimal cabling!!

In the coming weeks, we will be installing its included IBM Director software suite for detailed statistics gathering for monitoring and trending. I look forward to any other such enhancements to this “simply elegant” architecture.