Hi Rafa On Fri, Oct 24, 2014 at 12:47 PM, Rafa Haro <[email protected]> wrote: > I completely understand your point but I partially disagree. I find the > memory consumption requirement quite tough. It may prevent a lot of people to > give it try or experiment with it. Actually, Chalitha had serious problems to > find a machine for testing it. I agree that probably it is difficult to find > a valid architecture for any disambiguation approach, but focusing only in > Aida-Light I think that it is worth to provide also an other data management > solution. It will affect the performance for sure, but at least it will let > people use it without a super machine. I’m actually thinking in alternatives > that can “simulate” memory access using the disk like LevelDB or similar. > > Makes sense? >
IMO no and here is why: * If you need to solve this problem for a developer - find a way to trim down the size of the dataset - yago - so that he can test in setups with 5Gbyte of RAM * For a real installation you do not want to solve the problem as for dedicated machines memory is really cheap. Implementing an alternative that can use some hard drive technology would just be so much slower that you would always opt for the more memory option. So if your concern is about ease of deployment for developers or demoability I would recommend to work on a way to extract meaningful sub-sets of Yago that can be managed by Aida-Light setups requiring < 10GByte of RAM. best Rupert -- | Rupert Westenthaler [email protected] | Bodenlehenstraße 11 ++43-699-11108907 | A-5500 Bischofshofen | REDLINK.CO .......................................................................... | http://redlink.co/
