Theoretical breakthrough could boost data storage

A trio of researchers that includes William Kuszmaul—a machine subject Ph.D. pupil astatine MIT—has made a find that could pb to much businesslike information retention and retrieval successful computers.

The team's findings subordinate to alleged "linear-probing hash tables," which were introduced successful 1954 and are among the oldest, simplest, and fastest data structures disposable today. Data structures supply ways of organizing and storing information successful computers, with hash tables being 1 of the astir commonly utilized approaches. In a linear-probing hash table, the positions successful which accusation tin beryllium stored prevarication on a linear array.

Suppose, for instance, that a database is designed to store the Social Security numbers of 10,000 people, Kuszmaul suggests. "We instrumentality your Social Security number, x, and we'll past compute the hash relation of x, h(x), which gives you a random fig betwixt 1 and 10,000." The adjacent measurement is to instrumentality that random number, h(x), spell to that presumption successful the array, and enactment x, the Social Security number, into that spot.

If there's already thing occupying that spot, Kuszmaul says, "you conscionable determination guardant to the adjacent escaped presumption and enactment it there. This is wherever the word 'linear probing' comes from, arsenic you support moving guardant linearly until you find an unfastened spot." In bid to aboriginal retrieve that Social Security number, x, you conscionable spell to the designated spot, h(x), and if it's not there, you determination guardant until you either find x oregon travel to a escaped presumption and reason that x is not successful your database.

There's a somewhat antithetic protocol for deleting an item, specified arsenic a Social Security number. If you conscionable near an bare spot successful the hash array aft deleting the information, that could origin disorder erstwhile you aboriginal tried to find thing else, arsenic the vacant spot mightiness erroneously suggest that the point you're looking for is obscurity to beryllium recovered successful the database. To debar that problem, Kuszmaul explains, "you tin spell to the spot wherever the constituent was removed and enactment a small marker determination called a 'tombstone,' which indicates determination utilized to beryllium an constituent here, but it's gone now."

This wide process has been followed for much than fractional a century. But successful each that time, astir everyone utilizing linear-probing hash tables has assumed that if you let them to get excessively full, agelong stretches of occupied spots would tally unneurotic to signifier "clusters." As a result, the clip it takes to find a escaped spot would spell up dramatically—quadratically, successful fact—taking truthful agelong arsenic to beryllium impractical. Consequently, radical person been trained to run hash tables astatine debased capacity—a signifier that tin nonstop an economical toll by affecting the magnitude of hardware a institution has to acquisition and maintain.

But this time-honored principle, which has agelong militated against precocious load factors, has been wholly upended by the enactment of Kuszmaul and his colleagues, Michael Bender of Stony Brook University and Bradley Kuszmaul of Google. They recovered that for applications wherever the fig of insertions and deletions stays astir the same—and the magnitude of information added is astir adjacent to that removed—linear-probing hash tables tin run astatine precocious storage capacities without sacrificing speed.

In addition, the squad has devised a caller strategy, called "graveyard hashing," which involves artificially expanding the fig of tombstones placed successful an array until they inhabit astir fractional the escaped spots. These tombstones past reserve spaces that tin beryllium utilized for aboriginal insertions.

This approach, which runs contrary to what radical person customarily been instructed to do, Kuszmaul says, "can pb to optimal show successful linear-probing hash tables." Or, arsenic helium and his coauthors support successful their paper, the "well-designed usage of tombstones tin wholly alteration the … scenery of however linear probing behaves."

Kuszmaul wrote up these findings with Bender and Kuszmaul successful a insubstantial posted earlier this twelvemonth that volition beryllium presented successful February astatine the Foundations of Computer Science (FOCS) Symposium successful Boulder, Colorado.

Kuszmaul's Ph.D. thesis advisor, MIT machine subject prof Charles E. Leiserson (who did not enactment successful this research), agrees with that assessment. "These caller and astonishing results overturn 1 of the oldest accepted wisdoms astir hash array behavior," Leiserson says. "The lessons volition reverberate for years among theoreticians and practitioners alike."

As for translating their results into practice, Kuszmaul notes, "there are galore considerations that spell into gathering a hash table. Although we've precocious the communicative considerably from a theoretical standpoint, we're conscionable starting to research the experimental broadside of things."

---

---

More information: Linear Probing Revisited: Tombstones Mark the Death of Primary Clustering, arXiv:2107.01250 [cs.DS] arxiv.org/abs/2107.01250