What might cause that - is that CloudFlare or my webhosting?
[/quote]Create a page on your site that has no images, isn’t on CF and has only “Hello There” text, and test it.
You can also turn CF off and test your site - this can be educational in showing you CF really doesn’t do, jack for you.
Even though it’s just a savings of 3kb, I would still grab thethird image shown here, rename it to match what’s in the file system, and upload to overwrite.
You’re looking pretty darn good on the only test parameter I use, Dulles VA on fast cable connection, IE 10. Here’s that result from just now.
That lone F grade is there only because of the image I suggested you change out, above. Do that and this F becomes a A grade.
Don’t focus too much on the results from a single test - run a bunch of them and see if the slow DNS is common or if you just happened to catch an outlier. A single dropped packet can add 2 seconds to DNS for example. Generally CloudFlare is usually good about DNS performance but their DNS servers also get DOS attacked a lot (usually with no visible impact since they deal with it regularly) but if you catch a test right when a particularly large attack is going on you may also see some impact.
If you’re not already, you can follow their system status twitter account which notifies about any known issues: https://twitter.com/CloudFlareSys
[quote=“Anton_Chigurh, post:21, topic:8647”]
You’re looking pretty darn good on the only test parameter I use, Dulles VA on fast cable connection, IE 10. Here’s that result from just now.[/quote]
That’s true, results from LA today are also good: test result.
Thanks, that’s true. My webpagetest results seem to have improved from that location, so probably just a fluke (for example, Moscow here from today).
[quote=“Anton_Chigurh, post:21, topic:8647”]
You can also turn CF off and test your site - this can be educational in showing you CF really doesn’t do, jack for you.[/quote]
As an update, I upgraded previous weekend to CloudFlare Pro, not only since this would get better performance but also for the security. So far I have the impression that my first byte times have dropped, perhaps due that paying customers get a priority in a data centre over free customers.
Besides that, I also changed the following:
Changed cache expire headers from htaccess for HTML to multiple weeks so that more content is served from a CF data centre;
Enabled SPDY,
Enabled mirage 2 and image optimization from CF. This has dropped my grade due to how mirage works.
If I look at Pingdom’s RUM data, performance last week was much, much poorer:
Average page load time 3.9s (501 pageviews),
Average page load has increased for various countries compared to the weeks before: US (154 pageviews, now 4.0s), Australia (17 pageviews, now 35.0s!), UK (16 pageviews, now 3.9s).
Only a few countries with more than 10 pageviews have an average page load below 2 seconds this week: Hong Kong (0.78s), Germany (1.1s), Russia (0.93s), Italy (1.1s), Netherlands (0.40s), Ukraine (1.2s).
I don’t understand this: why would page views from Ukraine almost be four times as quick than from the US, given that the latter quite probably has superior internet infrastructure?
Furthermore, if I look at the average page load times as collected by Pingdom RUM, performance has deteriorated recently (I did not change the site layout; graph shows data from the last month. ):
Edit: Do you guys know how reliable Pingdom’s RUM data is? I’ve tried to replicate the poor results from Australia, but I still get a load time less than 3 seconds (test result), which is nowhere near the poor results Pingdom suggest.
Edit: Do you guys know how reliable Pingdom’s RUM data is? I’ve tried to replicate the poor results from Australia, but I still get a load time less than 3 seconds (test result), which is nowhere near the poor results Pingdom suggest.
[/quote]All I know is, multiple tests from multiple sites just muddy the waters. WPT is real browsers on real connections, it’s not theoretical at all and it is objective and scientific. It’s not trying to sell you anything and it doesn’t ask for your email address so it can spam you later. And I test my sites against only one parameter in WPT - fact cable connection, IE10 browser in Virginia. Then after I have them grading well on that, I test using WPT from other locations, other browsers and other connections just to see how they do with those. But I make my adjustments, based on only the one set of test parameters.
I think of it this way: Do you have 4-5 scales in your bathroom, weighing yourself on each one then just cherry picking the number you like? Of course not. That wouldn’t be scientific and it would just muddy the waters.
Remember my most important things about website optimization: 1.) There is a point of diminishing returns re: the amount of time and effort you’re putting in vs. results. 2.) The goal is to achieve a nice balance between form and function. You want to look good, but load fast. Never sacrifice performance for looks. Find that balance. And finally 3.) Keep checking your site, especially after adding anything. See if the addition(s) mess up your grades. And above all, keep it simple.
(Note: I chose the one test parameter in WPT - the IE 10 cable connection out of Dulles VA - because it is the most consistent and reliable of all the other locations and connections available in WPT. I used to test scattershot, as Patrick suggests, but then learned better. I make all adjustments based on the one parameter, then check other parameters just out of curiosity. But I never make any adjustments, based on those tests. And I only, ever, use ONE site. WPT.)
I don’t know why people think CloudFlare helps. It demonstrably does not. Think about it - it still relies ON YOUR HOST and it cannot ever, make that host machine any faster than it is. And when it is down, your site is down. It’s a SHAM. I used it for two years before figuring this out.
Edit: Do you guys know how reliable Pingdom’s RUM data is? I’ve tried to replicate the poor results from Australia, but I still get a load time less than 3 seconds (test result), which is nowhere near the poor results Pingdom suggest.
[/quote]All I know is, multiple tests from multiple sites just muddy the waters.[/quote]
I’m inclined to agree after more looking into this.
For example, a test today from Dulles shows 0,5s load time and from Sydney 0,4s. That are quite difference numbers than Pingdom suggests.
Furthermore, Google Analytics’ page timing tells the average load time last week was 2,19s, and not 3,9s as Pingdom says. Pingdom says their timing is better than Analytics’, but the results from Analytics are much more in-line with WPT results and what I experience myself when browsing the website.
More specifically, more than half of the pageviews are below 1.5 seconds according to Analytics:
[attachment=378]
Thanks for sharing that approach Anton. How do you deal with the incidental nature of WPT’s test? For example, if you have made a website change, do you run a test every 10 minutes?
[quote=“Anton_Chigurh, post:24, topic:8647”]
Remember my most important things about website optimization: 1.) There is a point of diminishing returns re: the amount of time and effort you’re putting in vs. results. 2.) The goal is to achieve a nice balance between form and function. You want to look good, but load fast. Never sacrifice performance for looks. Find that balance. And finally 3.) Keep checking your site, especially after adding anything. See if the addition(s) mess up your grades. And above all, keep it simple.[/quote]
Good points, thanks.
[quote=“Anton_Chigurh, post:24, topic:8647”]
I don’t know why people think CloudFlare helps. It demonstrably does not. Think about it - it still relies ON YOUR HOST and it cannot ever, make that host machine any faster than it is. And when it is down, your site is down. It’s a SHAM. I used it for two years before figuring this out.[/quote]
For dynamic websites, where CF still relies on the host, I think you’re right. But with caching everything while being on shared hosting, I think it helps. For me it’s an easier and more cost-efficient option than using a dedicated server for a small website.
[quote=“JMTC, post:25, topic:8647”]
Thanks for sharing that approach Anton. How do you deal with the incidental nature of WPT’s test? For example, if you have made a website change, do you run a test every 10 minutes?[/quote]It’s not as incidental as you might think. It’s just real browsers that clear cache each time. After making a change the results are immediate, on the next test you run after the change.
[quote]For dynamic websites, where CF still relies on the host, I think you’re right. But with caching everything while being on shared hosting, I think it helps. For me it’s an easier and more cost-efficient option than using a dedicated server for a small website.
[/quote]My common sense rule of thumb is: if your site is not optimized, CF doesn’t help you. It can’t possibly. And if it is, you don’t need CF. And leveraging browser cache of static content is a snap for a small site on shared hosting, (or any site) just put my magic code for this in the .htaccess file. CF and other CDNs though, interfere with this and make it ineffective.
CF itself doesn’t cache anything really - otherwise you wouldn’t get their “we’re sorry” splash page every time something goes wrong. You would get a cached version of your site and it never seems to happen. You’ve never seen that little CF message, “Cached version not available?” Why? Why is it not available? Why, when my host server is down, does CF go down too and with no cache? Because CF is a fraud.
