An information technology research company, OrionX, recently published a report detailing how the World Bank evaluated SPARC/Solaris systems for their hardware refresh. We were highly interested in this report, since it compares our own Fujitsu M10 to Oracle T5 systems.
The World Bank manages an investment portfolio of over $207 billion which consists of financial commitments to developing countries and regions. The bank issues bonds in the capital market and has to manage its lending and borrowing portfolios with sophisticated tools.
In order to manage these portfolios, the World Bank needs systems with both strong integer and floating point performance. Before the procurement, the bank was using 15 individual Fujitsu M4000 and M5000 servers. With the new systems, the bank was looking to consolidate these older systems down to a common set of six physical systems.
Given that the bank was already using SPARC/Solaris systems, they confined their search to servers from Oracle (Oracle T5) and Fujitsu (Fujitsu M10). They were happy with the SPARC/Solaris platform and didn’t see the need to port their applications over to another architecture.
World Bank received loaner systems from each vendor so they could test them with their own applications and workflows. Both of the test systems had 128 threads and 128GB of memory so that it was an apples-to-apples comparison.
The bank performed several tests on the machines, all mimicking real world workloads that the bank uses in day to day business. On three separate application benchmarks run in-house, the Fujitsu M10 system outperformed the Oracle T5 server and demonstrated remarkably predictable and stable performance throughout the tests.
The reason for these great performance disparities is due to processor design. Oracle designed their SPARC processor to handle lightweight transactional workloads and didn’t apportion much of the chip to handle floating point operations.
Fujitsu, on the other hand, designed their version of the SPARC CPU to perform well on both integer and floating point workloads. This is why it performed so well on the World Bank test workloads.
I encourage you to read the details about the World Bank evaluation here. It’s a fascinating real-world evaluation based on actual workloads and could help inform decisions you may have about your environment.