HP reckons its blade server systems use almost a third less power than those of arch-rival IBM.
The company has announced publicly available benchmark results which suggest that, under certain conditions, the HP BladeSystem c-Class uses up to 27 per cent less power than the IBM BladeCenter-H in similar configurations. It said that the benchmarks reflect real-world customer usage of blade server environments - although it is notoriously difficult to devise scenarios that meaningfully recreate a wide range of end-user experiences.
According to HP, the results derive from a week-long study conducted by Sine Nomine Associates (SNA), which examined the overall power consumption and external airflow requirements of a variety of blade server and 1U rack server configurations in what HP described as a typical data centre environment, experiencing light to heavy use.
The aim of the tests according to SNA was to show what happens when the servers perform tasks that generate the most heat. Processors and memory are the components that generate the most heat, so SNA used PRIME95, which uses Fourier transforms as a filtering mechanism to identify very large prime numbers, and is both compute- and RAM-intensive.
The labs reported that the tests "were run in an environmentally-controlled chamber that represents a typical rack area in a hypothetical large data centre. The chamber's available HVAC (heating, ventilating, and air conditioning) resources were such that there was no possibility of the servers under test requiring more airflow than could be delivered by the HVAC, and the temperature of the air at the inlet of the enclosure under test was monitored independently using a digital thermometer."
The report argues that the HP BladeSystem c-Class and its zoned cooling properties lower power usage through efficient power distribution and optimise airflow which between them are significant components in helping to keep data centres cool. The study argues that the HP BladeSystem c-Class requires 60 per cent less airflow than the IBM BladeCenter-H.
However, the results showing HP far ahead of IBM were only obtained under certain circumstances, according to SNA's report - specifically, when eight memory DIMMs were used and hotplug drives were installed.
SNA found that: "When only 4-DIMMs are needed and there is no requirement of hot-pluggable disk drives, the IBM and HP blade servers are closely competitive with one another for power consumption per server, differing by less than one per cent in these tests (non-interleaved memory)."
However, it went on to say that: "As soon as more memory or hot-pluggable drives are needed, IBM's use of expansion boards [for additional memory] costs dearly, and the HP BladeSystem with ProLiant BL460c significantly outperforms the IBM BladeCenter-H with HS21. In these tests, the BL460c used only about 340 VA/server versus 407 VA/server for the BladeCenter-H in its minimal 8-DIMM configuration, meaning the HP blade consumed about 16.5 per cent less power than the IBM blade, per server."
The report concluded that: "If hot-pluggable drives and larger memory are both needed, then the triple-wide IBM configuration is required and the power usage rises to 465 VA/server, which means the HP BladeSystem uses 26.8 per cent less in this instance."
SNA also found that the HP systems' fan speeds were more responsive to required cooling load than IBM's systems, and so consumed 20 per cent less energy.
IBM refused to comment.
“Power and cooling are serious issues for customers,” said HP's Mark Potter. “Sine Nomine took an objective look at this real-world customer problem and the test results show what we already knew: HP BladeSystem c-Class featuring Thermal Logic is the clear choice for customers addressing their power and cooling challenges.”