Written by Maile Ohye as the website, Jeremy Lilley as the Googlebot
Googlebot -- what a dreamboat. It's like he knows us <head>, <body>, and soul. He's probably not looking for anything exclusive; he sees billions of other sites (though we share our data with other bots as well :), but tonight we'll really get to know each other as website and crawler.
I know, it's never good to over-analyze a first date. We're going to get to know Googlebot a bit more slowly, in a series of posts:
- Our first date (tonight!): Headers Googlebot sends, file formats he "notices," whether it's better to compress data
- Judging his response: Response codes (301s, 302s), how he handles redirects and If-Modified-Since
- Next steps: Following links, having him crawl faster or slower (so he doesn't come on too strong)
And tonight is just the first date...
***************
Googlebot: ACK
Website: Googlebot, you're here!
Googlebot: I am.
GET / HTTP/1.1
Host: example.com
Connection: Keep-alive
Accept: */*
From: googlebot(at)googlebot.com
User-Agent: Mozilla/5.0 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)
Accept-Encoding: gzip,deflate
Website: Those headers are so flashy! Would you crawl with the same headers if my site were in the U.S., Asia or Europe? Do you ever use different headers?
Googlebot: My headers are typically consistent world-wide. I'm trying to see what a page looks like for the default language and settings for the site. Sometimes the User-Agent is different, for instance AdSense fetches use "Mediapartners-Google":
User-Agent: Mediapartners-Google
Or for image search:
User-Agent: Googlebot-Image/1.0
Wireless fetches often have carrier-specific user agents, whereas Google Reader RSS fetches include extra info such as number of subscribers.
I usually avoid cookies (so no "Cookie:" header) since I don't want the content affected too much by session-specific info. And, if a server uses a session id in a dynamic URL rather than a cookie, I can usually figure this out, so that I don't end up crawling your same page a million times with a million different session ids.
Website: I'm very complex. I have many file types. Your headers say "Accept: */*". Do you index all URLs or are certain file extensions automatically filtered?
Googlebot: That depends on what I'm looking for.
If I'm indexing for regular web search, and I see links to MP3s and videos, I probably won't download those. Similarly, if I see a JPG, I will treat it differently than an HTML or PDF link. For instance, JPG is much less likely to change frequently than HTML, so I will check the JPG for changes less often to save bandwidth. Meanwhile, if I'm looking for links as Google Scholar, I'm going to be far more interested in the PDF article than the JPG file. Downloading doodles (like JPGs) and videos of skateboarding dogs is distracting for a scholar—do you agree?
Website: Yes, they can be distracting. I'm in awe of your dedication. I love doodles (JPGs) and find them hard to resist.
Googlebot: Me, too; I'm not always so scholarly. When I crawl for image search, I'm very interested in JPGs. And for news, I'm mostly looking at HTML and nearby images.
There are also plenty of extensions (exe, dll, zip, dmg...), that tend to be big and less useful for a search engine.
Website: If you saw my URL, http://www.example.com/page1.LOL111, would you (whimper whimper) reject it just because it contains an unknown file extension?
Googlebot: Website, let me give a bit more background. After actually downloading a file, I use the Content-Type header to check whether it really is HTML, an image, text, or something else. If it's a special data type like a PDF file, Word document, or Excel spreadsheet, I'll make sure it's in the valid format and extract the text content. Maybe it has a virus; you never know. If the document or data type is really garbled, there's usually not much to do besides discard the content.
So, if I'm crawling http://www.example.com/page1.LOL111 with an unknown file extension, it's likely that I would start to download it. If I can't figure out the content type from the header, or it's a format that we don't index (e.g. mp3), then it'll be put aside. Otherwise, we proceed indexing the file.
Website: My apologies for scrutinizing your style, Googlebot, but I noticed your Accept-Encoding headers say:
Accept-Encoding: gzip,deflate
Can you explain these headers to me?
Googlebot: Sure. All major search engines and web browsers support gzip compression for content to save bandwidth. Other entries that you might see here include "x-gzip" (the same as "gzip"), "deflate" (which we also support), and "identity" (none).
Website: Can you talk more about file compression and "Accept-Encoding: gzip,deflate"? Many of my URLs consist of big Flash files and stunning images, not just HTML. Would it help you to crawl faster if I compressed my larger files?
Googlebot: There's not a simple answer to this question. First of all, many file formats, such as swf (Flash), jpg, png, gif, and pdf are already compressed (there are also specialized Flash optimizers).
Website: Perhaps I've been compressing my Flash files and I didn't even know? I'm obviously very efficient.
Googlebot: Both Apache and IIS have options to enable gzip and deflate compression, though there's a CPU cost involved for the bandwidth saved. Typically, it's only enabled for easily compressible text HTML/CSS/PHP content. And it only gets used if the user's browser or I (a search engine crawler) allow it. Personally, I prefer "gzip" over "deflate". Gzip is a slightly more robust encoding — there is consistently a checksum and a full header, giving me less guess-work than with deflate. Otherwise they're very similar compression algorithms.
If you have some spare CPU on your servers, it might be worth experimenting with compression (links:
Apache,
IIS). But, if you're serving dynamic content and your servers are already heavily CPU loaded, you might want to hold off.
Googlebot: Ah yes; meeting the parents, the robots.txt. I've met plenty of crazy ones. Some are really just HTML error pages rather than valid robots.txt. Some have infinite redirects all over the place, maybe to totally unrelated sites, while others are just huge and have thousands of different URLs listed individually. Here's one unfortunate pattern. The site is normally eager for me to crawl:
User-Agent: *
Allow: /
Then, during a peak time with high user traffic, the site switches the robots.txt to something restrictive:
# Can you go away for a while? I'll let you back
# again in the future. Really, I promise!
User-Agent: *
Disallow: /
The problem with the above robots.txt file-swapping is that once I see the restrictive robots.txt, I may have to start throwing away content I've already crawled in the index. And then I have to recrawl a lot of content once I'm allowed to hit the site again. At least a 503 response code would've been temporary.
I typically only re-check robots.txt once a day (otherwise on many virtual hosting sites, I'd be spending a large fraction of my fetches just getting robots.txt, and no date wants to "meet the parents" that often). For webmasters, trying to control crawl rate through robots.txt swapping usually backfires. It's better to
set the rate to "slower" in Webmaster Tools.
Googlebot: Website, thanks for all of your questions, you've been wonderful, but I'm going to have to say "FIN, my love."
Website: Oh, Googlebot... ACK/FIN. :)
***************
