On 04/14/2013 08:38 AM, Stas Oskin wrote:
Hi.I mean the sl4500 can have 2 nodes with 50 drives instead of 1 node with 60 drives. See this: http://h18000.www1.hp.com/__products/quickspecs/14406_div/__14406_div.HTML <http://h18000.www1.hp.com/products/quickspecs/14406_div/14406_div.HTML> That lets you use cheaper CPUs and also gives you better performance, but is slightly less dense and probably more expensive than the 1x60 model. The more important thing though is that when a node fails you only lose 25 drives instead of 60 so recovery will be smoother. Aha, but supermicro has a new king: http://www.supermicro.com/__products/system/4u/6047/ssg-__6047r-e1r72l.cfm <http://www.supermicro.com/products/system/4u/6047/ssg-6047r-e1r72l.cfm> 72 3.5" drives in 4U along with a motherboard and triple redundant power supplies! Crazy! I have no idea how well ceph would run on such a beast, but you'd probably need the fastest CPUs you could get your hands on and a lot of RAM. Back to this topic, what would your general advice be regarding the below: 1) If data availability and redundancy is most important, you would go with multiple 2U boxes to minimize cluster impacts in case of any downtime?
My general feeling here is that it depends on the size of the cluster. For small clusters, 2U or even 1U boxes may be ideal. For very large clusters, it is probably fine to use a denser chassis. It's all about the ratio of how many nodes are left to absorb the loss when a single node fails. I don't have any hard numbers, but my instinct is that in production I wouldn't want to lose more than about 10% of my cluster if a node dies.
2) Barring service and SLA, is it really worth taking HP over SuperMicro, or it's simply overpaying for a brand?
I've had great experiences with both Supermicro hardware and HP hardware. Products from both companies can work great for Ceph, but they are different companies and have different benefits and downsides. Service, support, and price are all things that may make one or the other a better fit depending on your needs.
Thanks, Stas.
-- Mark Nelson Performance Engineer Inktank _______________________________________________ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
