Daniel, > I'd really, really, rather have smaller nodes: single sockets, less > drives per node. Our unit of failure is a node, so I'd rather do > recovery in smaller, rather than larger pieces. "We just lost a > node; there goes 100TB; wait, what?" We need 65PB of data for DR11. We're scheduled to have around 250 nodes. That means 65PB/250 = 260TB per node (with replicas). Each chunk is going to be at least a TB. There's a lot more overhead for few-core, few-spindle blade-like servers than many-core, many-spindle boxes; I think there's going to be an ever-increasing minimum size to get good cost-effectiveness. -- Kian-Tat Lim, LSST Data Management, [log in to unmask] ######################################################################## Use REPLY-ALL to reply to list To unsubscribe from the QSERV-L list, click the following link: https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=QSERV-L&A=1