Cloud storage benchmarking highlights effect of distance

Distance to data centres remains the most important factor affecting performance for cloud storage services, a team of researchers from Dutch and Italian universities have found.

The researchers tested popular personal cloud storage services Dropbox, Microsoft SkyDrive, La Cie Wuala, Google Drive and Amazon Cloud Drive, running a set of benchmarks to ascertain the different factors that affect performance.

Google Drive and Wuala were the two fastest services at 26.49 and 33.34 megabit per second respectively - both had data centres nearby the researchers' test bed. Wuala has data centres located in Europe, whereas Google Drive terminates connections at its closest edge node, from where traffic is routed via the web services giant's private network.

Dropbox and SkyDrive data centres were located the furthest away, and therefore offered up the slowest results. A one megabyte file took four seconds to upload on SkyDrive, which had a 160 millisecond packet round trip time (RTT) to the researchers' test bed.

In comparison, Google Drive with a 15ms RTT only required 300ms to upload the one megabyte size file.

Advanced client features also play an important role.

Only Dropbox has a complete set of client capabilities that boost performance for networked storage especially when it comes to uploading multiple files and synchronising them with servers.

This includes chunking or splitting up large files into smaller pieces, and bundling multiple files instead of opening up several transmission control protocol (TCP) connections.SkyDrive, Wuala and Google Drive also do chunking, with different approaches as to how they split up the files.

Dropbox also supports deduplication of files to avoid unnecessary uploads, and delta encoding that transfers just the modified portion of data. Wuala also supports deduplication and client-side encryption.

File compression is always applied with Dropbox, while Google Drive uses it on a case by case basis. 

Testing with multiple files, Dropbox emerged as the winner by a factor of two, thanks to its advanced client features, uploading at 800kbps.

Google Drive's distributed network topology advantage was cancelled out by having to use separate TCP and secure sockets layer (SSL) connections for each file, which slowed down the service to just 189kbps.

Amazon's Cloud Drive is the only storage service that does not offer any of the above client capabilities the researchers noted, taking around sixty seconds to complete some tests at a slow 132kbps and having protocol overhead that caused it to use 5MB of data for each 1MB of content committed.

"Cloud Drive bandwidth wastage is an order of magnitude higher than other offerings, and its lack of client capabilities results in performance bottlenecks," the researchers wrote.

Even though the client capabilities help boost performance in general, they add to protocol overhead, the researchers note. Dropbox had the highest overhead, at 47 per cent for 100 kilobyte files and 22 per cent for 1 megabyte ones, due to the increased signalling between client and server thanks to the advanced features.

Ultimately the researchers found there was no clear winner, with all services tested suffering from some limitations.

"In some scenarios, the upload of the same file set can take seven times more, wasting twice as much capacity," the researchers concluded.

The benchmark scripts and the test methodology have been made available for public use by the team, which intends to collect data from volunteers to create new realistic tests.