Storage Latency Times: Google vs. Yahoo vs. Dropbox

By Henry Newman

The phrase sounded catchy to me, but what is non-cloudable data?  I would suggest that data that is cloudable is data that you can afford to have high latency access to, or have an application or system figure out how to cache it.  

I decided to ping three sites from my office over the period of the day and took the averages.  The three sites were and times were:

Average Latency in milliseconds from my computer:

Google: 51

Yahoo: 63

Dropbox: 83

These number are just from my computer to the site and do not including the latency to access storage you need to add in the storage path latency, which will at least add another 15 milliseconds getting to another host and then getting to a disk drive. And even if it is not a disk but an SSD, say another 3 milliseconds would be my best guess.

I think both numbers are very conservative.  So lets look at: the low end/high end latency range:

Low end latency

Google: 54/66

Yahoo: 66/78

Dropbox: 86/98

Now think of a local disk drive or SSD. Your range from your server to the device is likely sub-millisecond for SSD, say maybe a .5  of a millisecond to likely say 10 milliseconds for an enterprise 6 TB drive. 

So let’s look at: Low end latency differences cloud vs. local server/ High end latency differences cloud vs. local server

Google: 6.6x/108x

Yahoo: 7.8x/132x

Dropbox: 9.8x/172x

Clearly data that you need to access quickly and potentially randomly is not going to work well in the cloud as compared to locally, unless the whole file can be cached locally, which if you are writing and others are writing has is own set of problems.  I/O performance does matter for some applications and some clearly must be redesigned to work in the cloud.  

Photo courtesy of Shutterstock.

This article was originally published on August 18, 2014