Recently, plenty of researchers are looking at designing efficient data-oblivious algorithms. Roughly speaking, an algorithm is said to be data-oblivious if its data access patterns are independednt of the input i.e. the data access pattern of the algorithm does not leak any information about the input of the algorithm. Few of the algorithms already looked at include sorting, BFS, minimum spanning tree and convex-hull. These oblivious algorithms have various applications for the cloud and efficient secure computation. My queires regarding the same are as follows:

  • Is ORAM a generic method to transform any non-oblivious algorithm to its oblivious counterpart?
  • Is it always possible to construct a oblivious algorithm with the same asymptotic complexity as that of its non-oblivious counterpart? Please provide me with pointers to works which discuss on the same.
  1. Yes. There is an $\Omega(\log n)$ lower bound on ORAM. Therefore directly using ORAM to transform a non-oblivious algorithm to oblivious algorithm would incur a logN overhead. It is an open problem to design an ORAM matching the lower bound.

  2. No. There exists algorithms that do not have more efficient solution. As an apparent example, accessing a memory cell out of N cells requires $O(1)$ time in non-oblivious case but $\Omega(\log n)$ time in oblivious case, implied by the lower bound. I cannot think of more "useful" algorithms since there are not many lower bound proven.

I'm posting two relevant citations:

  • Nicholas Pippenger and Michael J. Fischer. Relations Among Complexity Measures. Journal of the ACM, Vol. 26, No 2, April 1979, pp 361–381. ACM PDF
  • Oded Goldreich and Rafail Ostrovsky. Software protection and simulation on oblivious RAMs. Journal of the ACM, Vol. 43, No 3, May 1996, pp 431–473. doi: ACM PDF

[follow up]

Adding more relevant (but incomplete list of) papers.

  • Goodrich, Michael T. "Randomized shellsort: A simple oblivious sorting algorithm." In Proceedings of the twenty-first annual ACM-SIAM symposium on Discrete Algorithms, pp. 1262-1277. Society for Industrial and Applied Mathematics, 2010.
  • Mitchell, John C., and Joe Zimmerman. "Data-oblivious data structures." In LIPIcs-Leibniz International Proceedings in Informatics, vol. 25. Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik, 2014.
  • Zahur, Samee, and David Evans. "Circuit structures for improving efficiency of security and privacy tools." In Security and Privacy (SP), 2013 IEEE Symposium on, pp. 493-507. IEEE, 2013.
  • Goodrich, Michael T., and Michael Mitzenmacher. "Privacy-preserving access of outsourced data via oblivious RAM simulation." In International Colloquium on Automata, Languages, and Programming, pp. 576-587. Springer Berlin Heidelberg, 2011.
  • Blanton, Marina, Aaron Steele, and Mehrdad Alisagari. "Data-oblivious graph algorithms for secure computation and outsourcing." In Proceedings of the 8th ACM SIGSAC symposium on Information, computer and communications security, pp. 207-218. ACM, 2013.
  • Nikolaenko, Valeria, Stratis Ioannidis, Udi Weinsberg, Marc Joye, Nina Taft, and Dan Boneh. "Privacy-preserving matrix factorization." In Proceedings of the 2013 ACM SIGSAC conference on Computer & communications security, pp. 801-812. ACM, 2013.
  • Keller, Marcel, and Peter Scholl. "Efficient, oblivious data structures for MPC." In International Conference on the Theory and Application of Cryptology and Information Security, pp. 506-525. Springer Berlin Heidelberg, 2014.
  • Wang, Xiao Shaun, Kartik Nayak, Chang Liu, T. H. Chan, Elaine Shi, Emil Stefanov, and Yan Huang. "Oblivious data structures." In Proceedings of the 2014 ACM SIGSAC Conference on Computer and Communications Security, pp. 215-226. ACM, 2014.

A lot has happened in this area in the past year regarding your second question.

Lower Bounds:

  • This paper proves that any oblivious RAM has an overhead of $\Omega(\log n)$. The lower bound from the original ORAM paper of Goldreich and Ostrovsky only applied to a certain type of ORAM constructions as discussed here.

  • This paper shows that the same lower bound applies to differentially private ORAM, which is a significantly weaker primitive than regular ORAM.

  • This paper shows that basically the same lower bound applies to oblivious stacks, queues, deques, priority queues and search trees.

Upper Bounds:

  • This paper constructs oblivious ram with an amortized overhead of $\mathcal{O}(\log n)$. Amortized here means that on average the operations have $\mathcal{O}(\log n)$ overhead, but from time to time some operations have an overhead that is linear in $n$. The construction in this paper is asymptotically optimal, but the hidden constants in the big-O notation are so big that it's not useful for any practical scenario.

  • This paper constructs the first oblivious priority queue with asymptotically optimal amortized overhead of $\mathcal{O}(\log n)$. The hidden constant is small, i.e. somewhere around 10, but since the overhead is amortized the practicality is again unclear.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.