Optimizing power consumption, CPU performance, and form factor is a never-ending battle in server design and IBM’s Bladecentre HS22 succeeds on all counts.
The HS22 that we tested came with two eight-core Intel 5570 Nehalem processors, two 10Gigabit Ethernet network interface cards, 50GB SSD drives, 150GB 15k hard drives and 48GB of RAM.
Intel claims improved performance and power consumption with their new Nehalem processors, so we tested the HS22 against an IBM HS20 blade with Intel Core-based processors. The Nehalem-based blade delivered a 10x increase in performance, while the amount of energy per transaction was only 12 per cent of the Core-based Xeon processor.
In terms of total overall power consumption, the Nehalem processors consumed 17 per cent more power than the Xeon-based system, but that’s a pretty good bargain when you’re gaining a 10x boost in performance.
IBM claims that solid state drives (SSD) are desirable because of their increased power efficiency as compared with conventional rotating spindle hard drives. This makes sense because of the SSD’s lack of power-consuming moving parts.
We compared the power consumption of the two drive types in the HS22 and found that the SSDs use 5 watts less power at idle than a conventional spindle drive and 5 watts less power under load than a conventional spindle drive.
However, spindle drives are able to sustain 37 per cent more bandwidth with writes than SSDs. This translates to 30 per cent more energy consumed per megabyte written for the SSDs than for the conventional spindles. In other words, spindles are more efficient than SSDs for high I/O rates; SSD drives are a better choice if the drives aren’t under heavy load.
Our 10G Ethernet tests showed good network and I/O performance with a peak of 8Gbps and a sustained rate of 7Gbps of HTTP traffic out of the server. The 10Gbps daughter card in our HS22 has two 10G Ethernet ports. The onboard 1G Ethernet ports remain active with the 10G Ethernet card in place.
The 10G Ethernet ports are accessible from the rear of the BladeCentre chassis via a 10G Ethernet switch or a 10G Ethernet pass-through module. The pass-through module allows direct connection to a blade’s 10G Ethernet port from outside the chassis. We conducted all our tests with the 10G Ethernet switch in place. The chassis has 10G option slots for a choice of connection methods to the two internal blade 10G ports if installed.
The HS22 makes a great virtualization platform because of the Nehalem processor, which implements Intel’s virtualization technology, and for the large RAM capacity. This hardware combination opens the possibility of running many virtual machines on the blade. The number of virtual machines is only limited by the processor and RAM requirements of the VMs.
Management
There are four categories of management for the HS22 – hardware/firmware, resource management, platform management, and enterprise management. Each category has a set of management tools and a loose definition.
The hardware/firmware category refers to the hardware and firmware configuration and revision management. There is a Web interface for the network attached power distribution unit, the advanced management module for the blade chassis and the I/O modules such as the 10G Ethernet switch in the chassis.
These Web interfaces provide a means to configure and manage these elements — including a remote video console for viewing the OS video output through the chassis management module. No external software tools are needed at this level. It is worth noting that the management Web interfaces were all we needed during all our testing.
The HS22 uses unified extensible firmware interface (UEFI) instead of BIOS. This gives more flexibility for configuration and management of the blade because access to components isn’t restricted by the limits of BIOS. If an operating system doesn’t support UEFI, the system will boot into a ‘legacy mode,’ which operates like BIOS to support the non-UEFI OS. With legacy mode, the boot time is effectively doubled. Since most all hypervisors are not UEFI aware, hypervisor boot times are very long.
The resource management category refers to the ability to configure systems level items, such as install operating systems on the blade, and blade firmware updates. IBM provides a Web site called toolscentre which provides access to a suite of tools that operate at the resource management level.
Platform management eases the management of multiple blades. Deciding when to use resource management instead of platform management is up to the systems engineer responsible for these tasks. Some environments can be served by either method, but the general idea is that the more devices you have to manage, the more likely you will need to perform platform management.
Management at this level is implemented by IBM Systems Director. Basic tools such as system discovery, firmware and software management, and system availability monitoring are included for free. Other plug-ins to Systems Director such as virtualization management and energy configuration can be purchased.
The highest management category, called enterprise management, is implemented by IBM’s Tivoli, which is available for a fee. Tivoli was not used in this review.
We found the management platforms to be comprehensive, easy to use and efficient. The only problem we discovered, if you consider this a problem, is deciding which tool to use. Fortunately, IBM has defined and documented the best practices for these tools, so that the question of which tool to use can be answered with some research.
It is worth noting that IBM does a good job with backward compatibility. The HS22 will work in a 6-year-old chassis with no problems – without 10G Ethernet support of course. The blade is easy to install and remove from the chassis. Once removed, the blade cover can be removed without tools to expose the internal components.
The RAM is a special low-height DIMM stick that is inserted at a 90-degree angle to the motherboard. This allows the RAM to be more densely packed in the blade. The HS22 has two drive bays on the front panel of the chassis so that the drives can be removed without removing the blade from the chassis.
We found the included documentation and Web site support to be useful and effective when troubleshooting configuration, capabilities and software revisions.
The HS22 is a well-designed blade that’s easy to use with lots of management tools. The Nehalem processor performance and energy efficiency is astounding. When considering power efficiency, the issue of whether to choose conventional hard drives or SSDs depends on your environment. The 10G Ethernet networking performed as expected. Finally, the combination of Nehalem processors, 10G Ethernet, and lots of RAM make the HS22 a great choice for virtualization.
Bass is the technical director of NEXT Services, an IT extension service of North Carolina State University. He can be reached at john_bass@ncsu.edu. Michael Brown, NEXT Services Engineer, assisted with the testing. He can be reached at msbrown3@ncsu.edu.