On 11/18/2015 04:42 PM, John R Pierce wrote:
the /really/ hard one when rolling your own highly redundant systems with high data integrity needed for things like transactional database servers, is implementing redundant storage controllers with shared writeback cache... you pretty much have to get into EMC class hardware for this level of reliability with data integrity and performance. and thats /really/ expensive stuff.
Yes it is, because it really is that hard to do shared writeback cache. EMC, Nimble, NetApp, and the like cost what they do because of those HA features. EMC storage processors have specialized shared backplanes and replicated write caches just in case an SP goes down while the data to be written is in cache and has yet to be committed (so that the trespassing SP can write the correct data to disk). They also have dedicated battery backup units and the whole concept of the 'vault' drives to specifically save the write cache in a powerfail emergency.
But I would love to see something in the free software space that did that kind of thing, with appropriate hardware.