Why would the performance improvement be visible only a specific hardware?
Ellison said users can expect real-time analytics queries 100 times faster and online transaction processing that is two times faster as long as they are using hardware that supports the Oracle 12c database.
I’ll assume that this could only mean that these results will be seen when data fits in memory. And not that one will need custom hardware to enable this feature. As a side note, I’m not sure I’m reading the announcement correctly, but it looks like a paying Oracle database customer will have to pay extra for the in-memory option.
Can anyone explain how data can be stored both in columnar and row format?
Additionally, the software will allow people to store data in both columns (used for analytics) and rows (used for transactions) as opposed to only one method; Ellison described this function as being “the magic of Oracle.”
Magic has very little to do with databases and performance.
Original title and link: Two questions about the Oracle in-memory database ( ©myNoSQL)
If you have a quiet Sunday and want to listen to something extremely awesome, you should try this episode of ThinkDistributed podcast covering causality in distributed systems with guest hosts: Peter Bailis, Carlos Baquero, and Marek Zawirski.
The links in the show notes will fill your reading list for a good while.
Original title and link: Causality: A discussion of causality, vector clocks, version vectors, and CRDTs ( ©myNoSQL)
Authored by a team from Karlsruhe Institute of Technology, the paper “Parallel graph partitioning for complex networks” presents a parallelized and adapting label propagation technique for partitioning graphs:
The graph partitioning problem is NP-complete ,  and there is no approximation algorithm with a constant ratio factor for general graphs . Hence, heuristic algorithms are used in practice.
A successful heuristic for partitioning large graphs is the multilevel graph partitioning (MGP) approach depicted in Figure 1, where the graph is recursively contracted to achieve smaller graphs which should reflect the same basic structure as the input graph.