WebPagetest Forums
CloudFlare accuracy of tests - Printable Version

+- WebPagetest Forums (https://www.webpagetest.org/forums)
+-- Forum: Web Performance (/forumdisplay.php?fid=3)
+--- Forum: Discuss Test Results (/forumdisplay.php?fid=4)
+--- Thread: CloudFlare accuracy of tests (/showthread.php?tid=931)

CloudFlare accuracy of tests - grahamginsberg - 07-15-2011 10:57 PM

I have installed the free version of CloudFlare just to see how it performs.

Before installing I did several tests: NZ, London, Paris, TX just to have the data

http://www.webpagetest.org/result/110715_FM_12AKK/ is a test done on a dynamic page after CloudFlare installed - no page speed difference noticed before and after Cloudflare

I spoke to CloudFlare yesterday and they said they were descussing with webpagetest.org to more accuraty show their speeds.

Can you shed some light on this because I'm about to dump it as I see no benefit right now.

RE: CloudFlare accuracy of tests - piratos - 07-15-2011 11:27 PM

There are some big differences in real speed between browsers.
With ie8 http://www.webpagetest.org/result/110715_74_12ASZ/

Pagespeed 1.11 beta says 91 - but 5 errors, 2 warnings and many rules you should do something.

Instead of a cdn you should use html5 appcache, that works in latest Browsers ( ff3.5 =>, safari, opera, chrome but not in ie (in ie10 it works).

RE: CloudFlare accuracy of tests - pmeenan - 07-15-2011 11:40 PM

html5 app cache requires a serious and complete rewrite of most sites/apps. Don't get into it unless you REALLY know what you are doing but when implemented correctly the gains can be great for repeat visitors (doesn't do anything for first visits though).

CDN and App Cache are orthagonal and can co-exist.

RE: CloudFlare accuracy of tests - pmeenan - 07-16-2011 12:07 AM

Sorry, back to your original question - I have been talking with them about several different things but there is nothing related to the accuracy of the measurements that I'm aware of.

Looking at the waterfall, most of the resources are served from different domains than your own so a CDN won't help much for that page.

RE: CloudFlare accuracy of tests - husdalcom - 07-16-2011 07:19 PM

Graham, your story sounds familiar. I was also running my site through CloudFlare and I too noticed some odd things when testing with tools.pingdom.com. Most resources would load superfast, but some would lag seriously behind. Visually my site appeared faster, but not when testing with Pingdom.

When asking CloudFlare about this I got more or less the same answer as you did; that there was something wrong with Pingdom, not CloudFlare.

However, if speed and performance matter to you, I wouldn't ditch CloudFlare. That said, there are alternatives, e.g. Incapsula, and if security is what matters most, Incapsula is better than CloudFlare, based on my own reckoning: http://www.husdal.com/2011/07/01/incapsula-versus-cloudflare/ As to performance they're about the same.

RE: CloudFlare accuracy of tests - grahamginsberg - 07-17-2011 02:48 AM

Thanks for the info. They definitely said that webpagetest was not showing the speeds accurately and that user experience was the best way to judge.

Unfortunately, thats not very scientific and I cannot be calling random people across the globe for their opinion. LOL

Anyway, I removed Cloudflare because it appeared to be more like a caching/security issue than a pure speed issue, which matters to me most. They have a boost (beta) feature available, but it messed up my tagged widget, so I disabled that.

With many pages that only get landed on once, the only solution is precaching, which is not a good idea for 20,000 pages that change almost daily.

My best solution for speed has been good ole smartopimizer. A tricky fella to get going, but works great for caching CSS and JS on the backend. Works through your htaccess.

@piratos: Not much I can do about those errors because it a dynamic page fed to my site for which I have very little control.

I may reconsider Cloudflare when they improve themselves and make their 'purge cache' a LOT easier to get to. It should be right on their homepage.

RE: CloudFlare accuracy of tests - grahamginsberg - 07-17-2011 03:51 AM

HERE ARE RESULTS - Two websites have the same exact style/feed of dynamic links
naples-fl-real-estate doesn't use anything but the normal hosting - NO PAGE CACHING
jupiterflhomes uses PAID VER OF CLOUDFLARE
Conclusion: Even though cloudflare does offer pages slightly faster, it is inconsistent, and in one case, the Sydney test, wouldn't load the page no matter what.

WEBPAGE TESTED: http://www.naples-fl-real-estate.com/idx/mls-211505399-14580_meravi_dr_bonita_springs_fl_34135

Yotta = 51 ; Gobal Average = 5.07 ; Global Reachability = 1.3 sec

WEBPAGE TESTED: http://jupiterflhomes.com/idx/mls-r3211338-31_bermuda_lake_dr_palm_beach_gardens_fl_33418

Yotta = 61 ; Gobal Average = 5.07 ; Global Reachability = 0.5 sec

RE: CloudFlare accuracy of tests - husdalcom - 07-17-2011 03:46 PM

Hi Graham and thanks for adding more details. Again, this sounds familiar. I too felt that CloudFlare was giving inconsistent results, at least according to my monitoring tools. Going over my historical and recent Yotta records just now, my picture is this:

Only server cache (11 weeks of data)
Time to title: 477 ms
Time to First Paint: 2110 ms
Time to Display: 3830 ms
Time to Interact: 4630 ms

CloudFlare + cache (6 weeks of data)
Time to title: 438 ms
Time to First Paint: 1450 ms
Time to Display: 1870 ms
Time to Interact: 2820 ms

Incapsula + cache (3 weeks of data)
Time to title: 410 ms
Time to First Paint: 1180 ms
Time to Display: 1540 ms
Time to Interact: 2310 ms

Conclusion: Both CloudFlare and Incapsula improve my site speed considerably, with Incapsula being perhaps an inkling better...when measuring. Visually, from my location, my site did seem a bit "snappier" (as CloudFlare likes to call it) when using CloudFlare, but I also noticed that while the text appears instantly, some images take a little longer to load with CloudFlare than with Incapsula.

That said, the main reason why I abandoned CloudFlare in the end were too many false positives of site being offline when it really wasn't...it's just that CloudFlare couldn't connect, perhaps consistent with your loading timeout?

RE: CloudFlare accuracy of tests - grahamginsberg - 07-19-2011 01:49 AM

Well I tried several times to get that page to test in NZ and nothing, so that worries me.

I'm looking for a new caching solution using hypercache but I'll start a new thread for that.