On 07/09/14 11:01 PM, Keith Keller wrote: > On 2014-09-08, Valeri Galtsev <galtsev at kicp.uchicago.edu> wrote: >> >> Even more: system failure or power loss is more likely to destroy all data >> on software RAID than on a single drive when there is a lot of IO present >> (to the best of my understanding, loss of cache software RAID is using is >> more catastrophic compared to journaled filesystem under same >> circumstances - somebody may correct me). So, there may be worth thinking >> about hardware RAID. > > I think an essential feature of any md RAID that's not a RAID1 is a UPS > and a mechanism for a clean shutdown in case of extended power failure. > (An md RAID1 might be okay in this instance, but I wouldn't want to risk > it.) But this is true for any RAID, which is why many controllers come > with a BBU (and if you don't have a BBU on your hardware RAID controller, > then you absolutely need the UPS setup I described). > > OTOH, the OP wasn't clear on what he was doing; perhaps he is just > playing around, and doesn't care about data preservation at this time. > If you're just testing performance then data integrity in the face of a > power failure is less of a concern. > > --keith A UPS is certainly better than nothing, but I would not consider it safe enough. A BBU/FBU will protect you if the node loses power, right up to the failure of the PSU(s). I've seen shorted cable harnesses taking out servers with redundant power supplies, popped breakers in PDUs/UPSes, knocked out power cords, etc. So a UPS is not a silver-bullet to safe write-back caching in software arrays. Good, yes, but not perfect. -- Digimer Papers and Projects: https://alteeve.ca/w/ What if the cure for cancer is trapped in the mind of a person without access to education?