Google's Instant search generates additional traffic that without QoS could impact network performance.
Google’s Instant search feature promises faster searches, smarter predictions and instant results, but could the behind-the-scenes work to speed searches cause additional network traffic — enough to impact network performance?
One software engineer tested Google ‘Classic’ against the newer search feature and found that some end users might generate additional traffic by performing traditional searches in the Instant interface. Jesse Najera, software engineer for CA Technologies, says he started wondering how the “additional request/response traffic” of Google Instant could be impacting end users’ network performance for other applications. Because if an end user quickly types, say, five words into the search field, the page gets reloaded at least five times before the end user looks at the result set, perhaps more if taking typos into account.
“Since paradigm shifts take some time to get used to, I’d say this type of behavior will be found for the near future – in other words, adjusting behaviors to view the results set after just one or half a word typed in, which should then reduce the overall load on the network from the Instant search feature,” Najera explains.
After thinking on the topic, Najera decided to test his theory. Using a search string of “seafood restaurants gavleston,” which includes a typo that Najera corrected in three trials, he looked for information outside of his location to test if Google would first search in the Austin, Texas, area based on the ISP/IP address.
“There were some interesting results, such as the amount of total traffic decreased as I repeated my Instant test with the typo. Perhaps this is due to the speed difference in typing the query, as this varied between each trial,” Najera says.
A quick Q&A with Najera reveals his inspiration for the Instant tests and what he thinks IT managers could do to avoid potential performance problems.
What made you think that the Google Instant search setting might cause additional network traffic?
Seeing the result set change as quickly as it does made me suspect the search application (using the term ‘application’ generically here) is sending additional requests and getting additional responses from the server. I wondered how much additional traffic this was adding to the network compared to search requests done prior to deployment of Google Instant. I also couldn’t find a disclosure from Google about this topic.
Why could this be a potential problem for IT managers, if end-users are using this setting?
In most environments, this issue may be moot. Search queries take up very little time on the network, which puts the chances of congesting a critical link low. However, I could see an issue if traffic prioritization is not enforced (via QoS or other means) and if network utilization between the Internet cloud and your corporate network is increasing. Worst-case would be spikes of search queries across a large user base suffocating other business-related traffic — though these would be short-lived as result sets are retrieved quickly, and would be more likely to occur if network utilization is already critical, say >90% on average.
What practices might help IT managers avoid this type of additional response/request traffic?
Specifically, the only way to totally avoid the additional traffic is to disable the Instant search feature. But, I’m not advocating that. My focus is to disclose the additional increase in traffic that is likely to occur if most users continue using search as they had been. For IT managers, implementing QoS or other such methods is a first step in mitigating the risk of mission-critical and business-critical traffic from being delayed or dropped during periods of high volume.
How could IT managers adjust end-user behavior to help avoid the increase in traffic?
This is one of the times where users, many of whom are not geeks or tech-savvy, need to be educated about the efficient use of online tools. They need to know how the new Instant search works, be aware that typing fewer items into the search bar may now yield the desired result, and understand that if they are annoyed or dislike the constant refreshing, that this option can be turned off, thereby saving a few bytes on the network. If IT management chooses to provide this information to users, they need to be careful how it is done. For example, a memo from IT management to users about “how to search” could be seen in a more negative light by users, especially by tech-savvy persons, and the desired message is lost in frustration.
Najera included technical disclosures for repeatability of the experiment below.
Environment:
- Windows 7 Professional
- 6 GB RAM
- Intel Core2 Duo P8800
Software:
- Wireshark 1.4.0
- Mozilla Firefox 3.6.10, preferences set to clear cache and cookies when closing program.
Methodology:
- Take preliminary packet capture to find what IP addresses are returned for “www.google.com” from DNS.
- Set Wireshark capture filter to “net 74.125.227.0/24”.
- When testing Google Instant, begin packet capture prior to entering “www.google.com” in Firefox.
- When testing Google “Classic”, load Google, change search preferences, acknowledge the preference change, then start the packet capture prior to entering the search query.
- Close Firefox after each test.
Actual order of testing:
- Instant — no typos
- (1 min separation)
- Instant — with typo and correction
- (2 min separation)
- Classic — no typos
- (2 min separation)
- Classic — with typo and correction
- (23 min separation – realized I needed more samples)
- Instant — no typos
- (1 min separation)
- Instant — no typos
- (1 min separation)
- Instant — with typo and correction
- (2 min separation)
- Instant — with typo and correction
- (1 min separation)
- Classic — no typos
- (1 min separation)
- Classic — no typos
- (1 min separation)
- Classic — with typo and correction
- (1 min separation)
- Classic — with typo and correction
Have you used Google Instant? Do you prefer it to Google Classic? What would you like to see change about the updated feature? Do you think network performance could be impacted? Please leave a comment here or let me know directly via e-mail at Denise.Dubie@ca.com.
Do you Tweet? Follow Denise Dubie on Twitter here.
No comments yet.