Why is Google so fast?

• Jan. 13, 2004

Perform any search on Google today and you should see results in about 1/4 second. If you see anything over 1/2 second than there's a problem. In fact, except for when the Florida update occurred, I rarely see results take longer than 4/10 of a second.

I guess in this age we live in - the wired age - we expect computers to run quickly. In fact, recent studies show that many people leave a website within about 7 seconds if it hasn't loaded in their browser by then. Think about that - 7 seconds. The average human takes about 2 breaths in that time. Sporting events have been won and lost in that span of time. In Google terms however, you could return results for over 14 Google queries in those 7 seconds (based on a less than a 1/2 second average).

So how does Google do this? If you break it down, Google is responding to about 4 searches per second per server (based on 250 million searches served up over 10,000 servers).

Well, they could do multiple reads and writes to a hard drive, but the average end-to-end hard drive access is 3/10 of a second. That only leaves 1/10 of a second or less to process and display the average search result. And what of those results which display in less than 3/10 of a second? How could Google possibly recall the data from a drive faster than the drive itself can read and write? (An interesting site note - my query for "average hard drive write" returned 1,120,000 results in 0.25 seconds)

The simple answer is memory. Google servers must have incredible amounts of RAM in order to serve up pages so fast. They do not write to drives. The whole index (or at least the important part) is cached in memory. This allows for the incredible return times on results. This is why they can not only compare and categorize results, but also return them to your browser at lightning fast speeds. If you consider that there are upwards of 100 different elements affecting the final search results, you can see why multiple reads and writes to the hard drives are inefficient. The calculations for results MUST be performed in memory to be able to return them so fast.

To put this in perspective, consider my query for "average hard drive write." Google would have had to review a certain number (probably between 1000 and 10,000) of those 1.1 million sites to determine their relevance to the query. Then, once they determined the relevancy, they then applied the various ranking factors (remember that's about 100, that we know of) to determine the final rank of those 1.1 million pages. That all happened in a quarter of a second.

Of course they have to write certain parts to the drive - such as cached versions of the pages, and so on, but the majority of Google is, in essence, accessed in RAM. Talk about virtual reality.

So before you complain the next time a web page is slow to load, consider what needs to be done to serve up that page. Now, consider what Google has to do to serve up its search results pages a quarter of a billion times per day.


Rob Sullivan
Production Manager
Searchengineposition.com
Search Engine Positioning
specialists