homepage Welcome to WebmasterWorld Guest from 54.204.249.184
register, free tools, login, search, pro membership, help, library, announcements, recent posts, open posts,
Become a Pro Member
Home / Forums Index / Google / Google News Archive
Forum Library, Charter, Moderator: open

Google News Archive Forum

This 61 message thread spans 3 pages: < < 61 ( 1 [2] 3 > >     
Different serps from 2 PCs on same network
phpmaven




msg:213263
 8:21 pm on Aug 11, 2003 (gmt 0)

I have a network setup where all PCs are accessing the Internet though 1 PC as a gateway. Machine A which is the gateway PC constantly gets different serps than machine B which is inside the network.

Now I'm not a TCPIP guru, but I would think that all traffic coming from my network would look like it was coming from the same IP. Why would the 2 PCs consistently get serps from 2 different datacenters?

This one's got me kinda perplexed. It's not that big of a deal. It's just kinda annoying when I tell me partner "hey, we're up to #10" and he says "It's at #14 on my PC”.

Any clues?

phpmaven

 

SlyOldDog




msg:213293
 1:26 pm on Aug 13, 2003 (gmt 0)

maybe try a search on hotbot with google selected and see how that looks?

mcavic




msg:213294
 2:37 pm on Aug 13, 2003 (gmt 0)

Load balancing does not explain this

Yeah it does. Again, use an ip address in the URL and you'll get identical results.

[edited by: mcavic at 3:13 pm (utc) on Aug. 13, 2003]

phpmaven




msg:213295
 2:51 pm on Aug 13, 2003 (gmt 0)

Wow, I didn't realize there would be so much interest in this topic.

There have been some interesting attempts at explaining this as a load balancing issue, but I'm still not convinced that that is what's happening.

Just to clarify, I am on a cable modem with a fixed IP, no proxy. I just did a couple of tests on my network that seem to debunk the theory that this is load balancing.

First of, I did an nslookup on both machines and proved that they were both getting the same IP for google.com. I also ran a packet sniffer on my tcpip traffic and can see clearly that all of the traffic between both PCs and Google is being sent to and received from the exact same IP.

Someone mentioned cookies, but I eliminated that as a possibility early on by disabling cookies on both machines.

Now even though the packets are coming back to me from the same IP, it's obvious that they are being served up from different servers. Then I am back to my original theory that Google is sniffing the user agent and serving up different results accordingly.

I guess if I really get ambitious I can take a close look at the packets coming back to each PC from the same IP and see if they are basically identical in content and the slightly different browsers on each machine are interpreting the results differently as someone suggested. I think that one is a bit of a stretch. If we were talking about Opera vs IE, then that might be a theory. But IE6.0.2600.000 vs IE6.0.2800.1106? I don't think so.

I'll post back if I come up with anything definitive after I do a bit more packet sniffing.

Thank you for all of your responses.

phpmaven

globay




msg:213296
 3:06 pm on Aug 13, 2003 (gmt 0)

that is really strange!

I guess your os and browser use the same language, am I right?

trillianjedi




msg:213297
 3:41 pm on Aug 13, 2003 (gmt 0)

Wow, I didn't realize there would be so much interest in this topic.

Well it's an interesting thread!

Just to satisfy mcavic's curiousity (and mine), could you please try putting an IP address for the URL in each browser (the same one obviously) rather than "www.google.com" and let us know the results?

TJ

hetzeld




msg:213298
 3:59 pm on Aug 13, 2003 (gmt 0)

Sounds interesting!
This is definitely the kind of question deserving a technical answer from GoogleGuy.
Hi GoogleGuy, where are you hiding? (if I may ask ;) )

Dan

phpmaven




msg:213299
 4:09 pm on Aug 13, 2003 (gmt 0)

trillianjedi,

I actually tried that this morning. Still got different results.

One other thing I just thought of. I'm going to monkey around with the user agent being sent from each machine and see what happens.

phpmaven

phpmaven




msg:213300
 4:11 pm on Aug 13, 2003 (gmt 0)

globay,

You are correct. Both have identical language settings.

phpmaven

sun818




msg:213301
 4:50 pm on Aug 13, 2003 (gmt 0)

Why don't you flush the DNS on both workstations and compare the search results. I get the same results, but if you sticky me your search terms, perhaps I can play around with. Anyway, after I flushed my DNS, the IP associated with google.com changed. Both workstations probably resolve to a different IP for google.com thus retrieving different results.


C:\>ping google.com

Pinging google.com [216.239.51.100] with 32 bytes of data:

Reply from 216.239.51.100: bytes=32 time=80ms TTL=51
Reply from 216.239.51.100: bytes=32 time=80ms TTL=51
Reply from 216.239.51.100: bytes=32 time=80ms TTL=51
Reply from 216.239.51.100: bytes=32 time=80ms TTL=51

Ping statistics for 216.239.51.100:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 80ms, Maximum = 80ms, Average = 80ms

C:\>ipconfig /flushdns

Windows 2000 IP Configuration

Successfully flushed the DNS Resolver Cache.

C:\>ping google.com

Pinging google.com [216.239.53.100] with 32 bytes of data:

Reply from 216.239.53.100: bytes=32 time=20ms TTL=48
Reply from 216.239.53.100: bytes=32 time=20ms TTL=48
Reply from 216.239.53.100: bytes=32 time=20ms TTL=48
Reply from 216.239.53.100: bytes=32 time=30ms TTL=48

Ping statistics for 216.239.53.100:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 20ms, Maximum = 30ms, Average = 22ms


wkitty42




msg:213302
 4:56 pm on Aug 13, 2003 (gmt 0)

phpmaven,

since you are so convinced that it is the useragent string that is being used, why not just go edit it on one or both machines and make it identicle?

searching the registry for "compatible; MSIE" sans quotes should locate it for you...

on a win98se box, i located one in HKCU\software\microsoft\windows\currentversion\internetsettings containing "Mozilla/4.0 (compatible; MSIE 6.0; Win32)"...

if nothing else, you could set up something like proxomitron to filter out the UA string and set it to something else... proxomitron can be set up on the main machine and all browsing filtered thru it or it can be set up on each machine... i'm sure there are other options to altering the UA string, too...

hetzeld




msg:213303
 5:13 pm on Aug 13, 2003 (gmt 0)

> since you are so convinced that it is the useragent string that is being used, why not just go edit it on one or both machines and make it identicle?

Set it to GoogleBot and see if they like it better :)

Dan

phpmaven




msg:213304
 5:18 pm on Aug 13, 2003 (gmt 0)

sun818,

I have already determined that google.com resolves to the same IP on both machines and have used the IP in the address bar on both machines. But I also just tried your suggestion which made no difference.

wkitty42,

That is exactly what I'm going to try a little later and see what happens.

phpmaven

nick420




msg:213305
 5:19 pm on Aug 13, 2003 (gmt 0)

I have noticed consistently different results for SERPs, number of indexed pages and backlinks since last Friday across all datacenters. Isn't it possible that instead of Google updating each datacenter one at a time, they are now almost randomly updating each server individually? By this I mean that with thousands of servers at each datacenter, until this "dance" is complete, it is very possible that even querying the same datacenter you will receive different results (as a result of load balancing within datacenters)?

While this doesn’t necessarily explain the consistently inconsistent results, it does decrease the odds of two PCs displaying the same results from 1-in-10 to 1-in-5500 (assuming you are querying the same datacenter, 1-in-55,000 if you are not). While I understand these numbers or odds aren’t accurate as there aren’t 55,000 different indexes on Google’s 55,000 servers, I think you can get the point I’m trying to make.

Yidaki




msg:213306
 5:43 pm on Aug 13, 2003 (gmt 0)

<edited>All quatsch - nonsense. Browser history cache fooled me again. Sorry!</editet>

[edited by: Yidaki at 6:21 pm (utc) on Aug. 13, 2003]

Sharper




msg:213307
 6:00 pm on Aug 13, 2003 (gmt 0)

This effect is most likely a result of load-balancing of three combined kinds.

Some more background:
Google has literally thousands of little (what are essentially stripped down) PCs serving requests. In one of the data centers I'm also in, their cage contains over 100 racks of their servers, with each rack having 4 servers on one shelf per U. That's over 150 servers per rack, over 1500 servers in their VA datacenter.

They do three things that combined could cause different results as described above.

1. The index data isn't stored on the servers. The index data is on central clustered storage devices that act as the filesystems for each individual server. There are multiple devices with multiple copies for redundancy and performance. Different groups of servers would use different file storage, but be able to fail-over to use another set if needed.

2. They use some DNS load-balancing, but mostly to the outside of their datacenters. An IP address for google.com is never assigned to a server itself, it is assigned to a clustered set of load-balancers. Those load-balancers (essentially working as network switches with some additional code on them) direct the traffic from there to and from the individual servers. The individual servers have a different set of IPs that the outside world doesn't see and that the load-balancers know.

3. DNS load balancing for "main" IPs, as described above.

The way those first two items above could cause this affect would be if the indexes for the file storage devices aren't in sync with each other, combined with the way the load balancing works.

This is almost certainly the case as portions of each index are updated. They could take some extensive timing steps to minimize this affect, but there really isn't much reason for them to go to that sort of trouble and effort.

Contrary to some of the above posted assumptions, high-end load balancers don't balance traffic based on source and destination IP address. If that was the case, AOL, proxying behind a few addresses, would overload the ones they were directed at. You can see how that would be a bad thing, eh? The source IP address doesn't matter, except as a piece of data to be used in the load balancing software. Generally the load balancer will use _at least_ the IP address, plus the source port to split requests between servers. Even on a proxied or nat'd network, the source port is almost always going to be different. Google could also use details from the request itself (browser, URL requested, etc...) to add into the load balancing decisions, but because that would mean sacrificing speed and LB cpu(have to look at a higher level in and more of the packet), they are extremely unlikely to be using anything more than source IP and port.

The other wild card is that they may or may not be using "sticky" connections to keep the same person connected to the same server for a certain period of time before balancing them again. There are good arguments for them to use either one (sticky lowers the transaction cost of setting up new connections and also makes it _easier_ to use http 1.1 instead of 1.0 and https instead of http if they desire, while non-sticky load balances a little better and speeds up recovery time if a server fails and shouldn't get traffic anymore), but from the posts above, it sounds like they are using some form of stickyness.

In summary, they'll probably load balance you based on DNS, IP and port to a specific server and that server may or may not have the same storage as another connection from a different combination of DNS, IP and port, however the number of possible storage devices is much lower than the number of possible servers and the percent of difference between storage device's indexes is complete guesswork for those of us who don't actually work for Google.

Then of course, if you throw in geo-targeting and languages, etc... it just gets that much more complicated. :)

claus




msg:213308
 6:09 pm on Aug 13, 2003 (gmt 0)

>> msie redirects to google.de

I've experienced the same language-confusion although it's OT in this thread. I'm in Denmark, not in Germany and my English browser accepts English only. My (fixed) IP is owned by a company that has nothing to do with Germany. Further, i've set my Google language preferences to English.

In my case it says "Go to Google Deutschland" on the G front page. Strange.

Yidaki




msg:213309
 6:12 pm on Aug 13, 2003 (gmt 0)

<edited>All quatsch - nonsense. Browser history cache fooled me again. Sorry!</editet>

[edited by: Yidaki at 6:20 pm (utc) on Aug. 13, 2003]

Sharper




msg:213310
 6:20 pm on Aug 13, 2003 (gmt 0)

Yidaki,

If you are getting a 302 redirect, that's coming from the web server your request went to, not the load balancer. I doubt they do geo-targeting or language at the load balancer network device level. I presume that their web server (for language at least) or DNS (possible for some very rough geo-targeting, but most likely web servers as well) is handling that, especially since it should be a one-time handoff.

I would think your different browsers (different source ports for each browser, of course) can handle or be handled by Google's web servers differently for language depending on exact implementation and code details.

Yidaki




msg:213311
 6:23 pm on Aug 13, 2003 (gmt 0)

Yep, sorry Sharper, i just edited / deleted my previous posts. I've been fooled by the browser history. How embarrassing. :/

RossWal




msg:213312
 6:40 pm on Aug 13, 2003 (gmt 0)

From the user's perspective this is B-A-D bad. I found a site selling something I wanted over the weekend. I posted an inquiry and haven't gotten a response. So I want to call them. Go back to google and no can find. They were number 3 on Saturday. Don't remember the name and thought I was loosing my mind until I remembered what's going on with google. I frequently don't bookmark things because I rely on being able to retrace my steps to get back.

Ross

Camster




msg:213313
 6:57 pm on Aug 13, 2003 (gmt 0)

This one might be too obvious to mention, but have you checked to make sure that the Google preferences are identical for both browsers?

[google.com ]

The interface language, search language and safesearch settings can all make a difference in the SERPs.

TheTruth




msg:213314
 7:57 pm on Aug 13, 2003 (gmt 0)

Hello everybody:
I'm surprised that you are all so surprised at this. I'm used to seeing different GRs [google ranking =) ] on the same PC when repeating a query 5 minutes apart from one another. Currently I have two sites with fluctuating PRs. My first site returns a PR of 3 half the time and PR of 1 the other half. My other site has PR of 4 or 3 depending on how G feels. I'm not sure about all the load balancing stuff but logically it has to be that not all G machines are housing the same data (like you guys mentioned).

I would suspect that eventually they will all catch up to the latest crawl info and GRs will become more stable for a given site. But then again this won't competely happen because it is always crawling and updates for other sites are also continuous.

In my case I think my PRs are on their way down because they used to be the higher ones =T!

hetzeld




msg:213315
 8:11 pm on Aug 13, 2003 (gmt 0)

> The interface language, search language and safesearch settings can all make a difference in the SERPs.

Good point camster! So obvious that most of us completely forgot it ... :(

Dan

globay




msg:213316
 8:25 pm on Aug 13, 2003 (gmt 0)

As far as I know, search preferences are safed in a cookie. So if you have deleted your cookies, it won't have an impact on the search results. I guess, it's more likely a problem with local / ISP caching.

projectphp




msg:213317
 3:05 am on Aug 14, 2003 (gmt 0)

I want to point out a statistical error. The chance of two computers using the same Goiogle data centre, given that there are 10, is actually 10/100 which is one in ten.

How come? Well, there are 100 combinations (10 * 10) of which 10 are the same, i.e. 1-1,2-2,3-3,4-4,5-5,6-6,7-7,8-8,9-9,10-10. 10/100 === 1/10.

Another way of looking at it is, comp X is at some datacenter, what is the chance Y will pick the same one? 1 in 10.

Again, it matters not which one is picked. Small point, but nonetheless important. 99% chance of being different is a lot different than 90%.

hetzeld




msg:213318
 5:13 am on Aug 14, 2003 (gmt 0)

Thanks projectphp,

Agreed, this had ben spotted by Globay in msg#20 ...
It's kind of a "mental typo" due to my lack of english fluency :)

Dan

Some_Bloke




msg:213319
 8:16 am on Aug 14, 2003 (gmt 0)

Looks like I found the source of my problem. I just changed the number of results shown per page. It makes quite a difference.

(I'm still comparing using the "dance tool" for convenience.)

When it is set to return 10, 15, 20 or 25 results per page, the serps remains constant.

At 30 results per page, position drops by 2.

At 50 per page it rises by 2 again.

But at 100 per page it drops by 7.

I tried this several times and the results appear consistent and repeatable.

Dave

Some_Bloke




msg:213320
 9:28 am on Aug 14, 2003 (gmt 0)

Yet more checking shows it is definitely the results per page. By changing this setting on the Google preferences (on their main page) it seems to have an effect on the indented second pages (but not all) of some sites listed in serps.

I tried with various vague words such as apples, bananas etc.(am I allowed to mention that here?) and it keeps doing the same.

So you can now get higher or lower ranking dependant upon people's personal settings.

Dave

hetzeld




msg:213321
 12:42 pm on Aug 14, 2003 (gmt 0)

Some_Bloke,

Let's imagine you have #3 for a given search term.
If your main competitors (#1 and #2) have another result showing in any position between 11 and 20 and the visitor has selected to show 20 results per page or more, those supplementary results will show up, indented right after the #1 and #2 position, pushing you downwards by 2 positions.

So... yes the position depends on user's settings, as your page would show up in #3 (for 10 results/page) or #5(for 20 or more) :)

Dan

Some_Bloke




msg:213322
 2:38 pm on Aug 14, 2003 (gmt 0)

hetzeld:

So is that how it's always been? - I didn't realise.

Thanks for the explanation, it makes sense now.

Dave

hetzeld




msg:213323
 4:40 pm on Aug 14, 2003 (gmt 0)

Some_Bloke,

AFAIK, at least for the last 18 months. I wasn't spending much time trying to improve SERPs before that... ;)

Dan

This 61 message thread spans 3 pages: < < 61 ( 1 [2] 3 > >
Global Options:
 top home search open messages active posts  
 

Home / Forums Index / Google / Google News Archive
rss feed

All trademarks and copyrights held by respective owners. Member comments are owned by the poster.
Home ¦ Free Tools ¦ Terms of Service ¦ Privacy Policy ¦ Report Problem ¦ About ¦ Library ¦ Newsletter
WebmasterWorld is a Developer Shed Community owned by Jim Boykin.
© Webmaster World 1996-2014 all rights reserved