When quoting this document, please refer to the following
DOI: 10.4230/LIPIcs.ESA.2018.41
URN: urn:nbn:de:0030-drops-95042
Go to the corresponding LIPIcs Volume Portal

Goswami, Mayank ; Medjedovic, Dzejla ; Mekic, Emina ; Pandey, Prashant

Buffered Count-Min Sketch on SSD: Theory and Experiments

LIPIcs-ESA-2018-41.pdf (1 MB)


Frequency estimation data structures such as the count-min sketch (CMS) have found numerous applications in databases, networking, computational biology and other domains. Many applications that use the count-min sketch process massive and rapidly evolving data sets. For data-intensive applications that aim to keep the overestimate error low, the count-min sketch becomes too large to store in available RAM and may have to migrate to external storage (e.g., SSD.) Due to the random-read/write nature of hash operations of the count-min sketch, simply placing it on SSD stifles the performance of time-critical applications, requiring about 4-6 random reads/writes to SSD per estimate (lookup) and update (insert) operation. In this paper, we expand on the preliminary idea of the buffered count-min sketch (BCMS) {[Eydi et al., 2017]}, an SSD variant of the count-min sketch, that uses hash localization to scale efficiently out of RAM while keeping the total error bounded. We describe the design and implementation of the buffered count-min sketch, and empirically show that our implementation achieves 3.7 x-4.7 x speedup on update and 4.3 x speedup on estimate operations compared to the traditional count-min sketch on SSD. Our design also offers an asymptotic improvement in the external-memory model over the original data structure: r random I/Os are reduced to 1 I/O for the estimate operation. For a data structure that uses k blocks on SSD, w as the word/counter size, r as the number of rows, M as the number of bits in the main memory, our data structure uses kwr/M amortized I/Os for updates, or, if kwr/M > 1, 1 I/O in the worst case. In typical scenarios, kwr/M is much smaller than 1. This is in contrast to O(r) I/Os incurred for each update in the original data structure. Lastly, we mathematically show that for the buffered count-min sketch, the error rate does not substantially degrade over the traditional count-min sketch. Specifically, we prove that for any query q, our data structure provides the guarantee: Pr[Error(q) >= n epsilon (1+o(1))] <= delta + o(1), which, up to o(1) terms, is the same guarantee as that of a traditional count-min sketch.

BibTeX - Entry

  author =	{Mayank Goswami and Dzejla Medjedovic and Emina Mekic and Prashant Pandey},
  title =	{{Buffered Count-Min Sketch on SSD: Theory and Experiments}},
  booktitle =	{26th Annual European Symposium on Algorithms (ESA 2018)},
  pages =	{41:1--41:15},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-081-1},
  ISSN =	{1868-8969},
  year =	{2018},
  volume =	{112},
  editor =	{Yossi Azar and Hannah Bast and Grzegorz Herman},
  publisher =	{Schloss Dagstuhl--Leibniz-Zentrum fuer Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{},
  URN =		{urn:nbn:de:0030-drops-95042},
  doi =		{10.4230/LIPIcs.ESA.2018.41},
  annote =	{Keywords: Streaming model, Count-min sketch, Counting, Frequency, External memory, I/O efficiency, Bloom filter, Counting filter, Quotient filter}

Keywords: Streaming model, Count-min sketch, Counting, Frequency, External memory, I/O efficiency, Bloom filter, Counting filter, Quotient filter
Seminar: 26th Annual European Symposium on Algorithms (ESA 2018)
Issue Date: 2018
Date of publication: 08.08.2018

DROPS-Home | Fulltext Search | Imprint | Privacy Published by LZI