On Tue, Jan 8, 2019 at 4:28 PM Mark Nelson <mark.a.nelson@xxxxxxxxx> wrote: > >> take this model to the extreme? > > > > could you name some of them so we can be more specific? > > Say in 5 years if optane or nvdimm style technologies become > significantly cheaper and yet we still can't scale cores to be any > faster (but we get more of them). What happens if we start needing > dozens of OSDs on one device to achieve high performance? Do the > underlying layers hold up? Does the model in general hold up? Mark, you're rising here an *extremely* important problem of RADOS scalability that IMHO is being mixed with another one in the whole discussion. Let me decompose. The two issues we're talking are: 1. Efficient consumption of a very fast device -- that's the primary goal for crimson-osd, I assume. 2. Scalability of the RADOS infrastructure. The junction point I see between them is the difference in supply and demand for the RADOS name resolution entities. A new osd design could increase (by partitioning) or decrease demand (through aggregation). **If** we're scarce on the resource, considering *solely* OSD might be not the best way to address #2 as it entirely ignores the supply side. The topic would require much broader view. However, the essential thing is to judge how much RADOS capacity we have. Driven by the 10k OSD testing I'm living in world where plenty of it seems to be available. Regards, Radek
