As we talked this morning wth IBM representative, Our Bank, for its baremetal BigData clusters will probably split Spark compute nodes and non-Spark compute nodes (using hadoop label) for financial and optimization reasons.
it might be interesting for some usecases (ie Spark) to use Power S822L system instead of intel servers expecially for their performance and their ability to scale up
An hadoop Biginsights version allowing hybrid BigInsights clusters integrated in Ambari (Intel & Power) will really be interested.
For sample , it's currently possible to mixed redhat, suse etc ... nodes in the same cluster ? why not mix Redhat Power with Redhat linux ?
Thanks for your answer.