While scale-out IT architectures are all the rage these days, there are still quite a few new and old applications that benefit more from scale-up architectures like, for example, SAP S/4 enterprise resource planning (ERP) applications running on the HANA in-memory computing platform. There are instances where running HANA applications on a scale-out architecture is fine. But when it comes to combining transaction processing and real-time analytics, scale-up architectures are going to be required.
Cisco has become the latest systems vendor to decide to resell a platform from SGI that is specifically designed to scale up applications in memory. Earlier this year SGI struck a similar alliance with Dell, and Hewlett Packard Enterprise (HPE) has an OEM agreement in place surrounding an eight-socket appliance that HPE intends to build based on SGI server technology.https://o1.qnsr.com/log/p.gif?;n=203;c=204663295;s=11915;x=7936;f=201904081034270;u=j;z=TIMESTAMP;a=20410779;e=iBill Dunmire, head of product management and marketing for SGI’s High Performance Data Analytics group, says the issue with scale-out architectures in many mission-critical use cases is that all those separate compute and storage engines have to be networked. That’s fine for two, four and even eight socket systems, but all that additional latency isn’t present in a SGI UV 300H system that is purpose-built to scale up using Intel Xeon-class processors.
Of course, Cisco, Dell and others will continue to sell scale-out systems because there are many scenarios where that architecture makes sense. But when it comes to in-memory computing, many of the precepts that companies such as SGI originally developed for the high-performance computing (HPC) market now apply to the high-performance sector of traditional enterprise data centers.
Whether an IT organization opts to acquire an SGI platform directly from SGI or through one of its OEM partners will depend on the relationship it has with a particular vendor and the level of support required. But one thing is certain: The one-size-fits-all notion regarding application workloads in the data center can be filed away with every other piece of IT gospel that has turned out to be a myth.