Highest possible bandwidth, lowest possible latency—this is the ideal scenario for large-scale enterprise and scientific computing and the subject of some discussions this week with numerous individuals we spoke with at ISC in Hamburg this year. InfiniBand continues to make this possible in the face of mounting data demands and so too does cloud, at least for some applications. Institutions can finally ditch the expensive proposition of building and maintaining their own cluster since cloud providers with ultra-infrastructure can cater to these needs in many application contexts. If your application is one that is well suited to cloud computing, this change in delivery of compute resources and management is nothing short of revolutionary.
To discuss the role of InfiniBand in cloud and HPC, we met with Brian Sparks from Mellanox Technologies here in Hamburg at ISC to get at the heart of why innovations in InfiniBand matter in the cloud, especially in scientific application areas. Sparks talks in some detail about what the role of InfiniBand is currently in the HPC and cloud space and what is coming in subsequent years as it matures further.
The InfiniBand Trade Association (IBTA) announced its new roadmap for the next year of growth that now includes 4x EDR at 100Gbps data rate in 2011. While we’re on the topic, we should mention that InfiniBand in general was a topic of discussion (albeit in scattered senses) among several of the exhibitors we spoke with. According to the IBTA, “InfiniBand is now deployed in 208 systems of the TOP500 supercomputer sites—an increase of 37 percent from one year ago.”