HyperDex The Next-Generation Key Value Store

NEW RELEASE! HyperDex now has an improved fault-tolerant coordinator, a more efficient storage backend, and an expanded API that includes floats, range searches on strings, sorted search, group delete, and count. As usual, packages are available for Debian, Ubuntu and Fedora with source available for others.

Slides from the HyperDex talk at VMware Academic Research Symposium are now online.

Slides from the HyperDex talk at SIGCOMM are now online.

New feature: HyperDex now supports sorted-searches, counts and group-deletes. Sorted search returns the top N objects in sorted order. Count will efficiently count the number of objects that match a search predicate. Group-delete removes objects which match a search predicate. With these additions, HyperDex continues to be the fastest distributed key-value store with the richest API.

NEW RELEASE! HyperDex now supports full cluster restart, uses a faster messaging layer and provides better semantics for math operations. If you're using prebuilt packages for Debian, Ubuntu, or Fedora, the update is immediately available in your package manager.

Talks at MSR, LinkedIn and Twitter on HyperDex. Here is the latest slide deck.

NEW RELEASE! HyperDex now supports lists, sets, and maps natively, with atomic operations on each of these structures. This enables HyperDex to be used in ever-more demanding applications that make use of these rich datastructures. As usual, packages are available for Debian, Ubuntu and Fedora with source for others.

Slides from the University of Buffalo presentation are now online.

In response to user feedback, we've put together an official API guide and documentation. Check out the HyperDex documentation (and HyperDex documentation in PDF). The documentation is for the current development version.

NEW RELEASE! HyperDex now supports conditional puts, atomic increments, a command-line interface to the coordinator, and natively supported types. Additionally, we've fixed a few bugs we found along the way. Check out the tutorial to see the new types in action. If you're using HyperDex, we encourage you to upgrade. If you're using the pre-built packages, you can upgrade to the new release with your package manager.

We have created two mailing lists, hyperdex-announce and hyperdex-discuss. Come and join the discussion!

There's new performance numbers comparing HyperDex to Redis. In every workload, HyperDex is faster than Redis. This is especially true for the SEARCH workload where HyperDex is a full 14 times faster than Redis.

There is now a brand new list of Frequently Asked Questions, with answers!

We are excited to announce version 0.2b6 of HyperDex, which now comes with full Python bindings. This makes experimenting with HyperDex significantly easier, as you can see in our new HyperDex tutorial.

This new release also includes a number of minor bug fixes. As always, please feel free to contact us at if you have any questions about HyperDex or the new release.

It is now easier than ever to check out HyperDex! We now provide precompiled Ubuntu 11.10 and Fedora 16 packages. Just follow the instructions on our download page to find out how to install HyperDex on your favorite platform.

As promised, we have updated our performance page with new numbers from the epoll patch. We now outperform MongoDB and Cassandra by a factor of 2-13!

Some of our internal performance tests found that, although HyperDex scales very well with increasing request load on the servers from a fixed number of clients, it did not scale nearly as well when we increased the number of clients dramatically. This is primarily due to our use of poll to handle client connections in order to provide platform independence. In response to these performance results, we have decided to use platform specific code for multiplexing I/O, and have replaced poll with epoll on Linux. The performance difference is pretty dramatic; we will update our performance page soon with the performance of HyperDex using epoll.

We're working on getting forums/mailing lists available. In the mean time, if you're interested in the system, contact us at:

We welcome patches that add features, improve performance, or fix bugs. Our contributors page should provide all the information that you need to help us improve HyperDex.

We are excited to announce that HyperDex is now available as a precompiled Debian package! Moving forward, we plan to add precompiled package for Ubuntu and Fedora in the near future.

We have posted a technical report that describes in depth the technical details of HyperDex. The report details how hyperspace hashing works as well as other tricks required to implement a high performance key-value store.

As part of our performance analysis of HyperDex, we have deployed an early version of our system on 256 EC2 nodes. This deployment enables us to evaluate the current scalability of our system, find problem areas that we need to address, and get a sense of the performance of the system in what we expect to be a typical deployment size.

Our performance results are exceeding our initial expectations and we are seeing amazing scaling results given the preliminary nature of our implementation.

HyperDex now has a home at http://hyperdex.org/! This page will keep you up to date on all the exciting developments relating to HyperDex.