Home > Paid Traffic Sources > Mobile

site loading speed from RUM is not pretty (12)


04-24-2015 05:17 PM #1 ryanfuse (Member)
site loading speed from RUM is not pretty

Hi, need some technical advise. I used Pingdom's RUM on one of my site currently serving South East Asia offer and the loading time is not pretty, about 2.4 second on average.

And the bulk of the slowdown is caused by redirect (1.9s!) and DOM (1.1s). Now this is confusing because I use Voluum as my tracker, so that part of redirecting should not be a problem.

So maybe redirect in this case means my "pretty" URL for my CDN. I created a CNAME alias to the CDN.

Other than these two redirect I have no other redirects.

I'm pretty sure DOM is caused by inefficient code because I use Webflow for that landing page so that's not such a big mystery.

By the way, I use EdgeCast as my CDN.

Another possibility is that Cedexis Radar is taking a long time to load, not sure if this is considered a redirect, but Pingdom website performance grade for redirect recommends I remove Cedexis radar.js and handshake.js

And finally, Cedexis is also reporting the same result, with only 25% of user loading the page at 2 second and lower, yikes!

Please advise, thanks.


04-24-2015 05:54 PM #2 rdigital100 (Member)

Who are you using for DNS? Is that causing the redirect slowdown because they don't have SEA point of presence?

Probably not the problem but always a good first thought to look into. i think route53 has pops in SEA, and i know cloudflare does.


04-25-2015 02:14 AM #3 ryanfuse (Member)

Quote Originally Posted by rdigital100 View Post
Who are you using for DNS? Is that causing the redirect slowdown because they don't have SEA point of presence?

Probably not the problem but always a good first thought to look into. i think route53 has pops in SEA, and i know cloudflare does.
I'm using DNS Made Easy, should not be a problem as the RUM on DNS is like 0.13 s


04-25-2015 06:31 AM #4 zeno (Administrator)

The DNS is not the issue. DNS resolution times become less and less important with volume due to ISP-level caching.

With Cedexis, their script is asynchronous and will not impact user load times and you can configure it to load quite late. Be aware its not the ideal RUM solution because it involves bandwidth/ping testing to other platforms as part of the quid pro quo nature of community-derived data.

What it will be is two things:

1) your server and Voluum's are not that close to users
2) your users have poor connectivity

It's pretty much as simple as that - don't expect turtle connections to load even a small page very quickly when all the servers that need to be contacted first are quite far away. The CDN serving the pages will likely be the strongest link in the chain since they will have a PoP in that country.


04-25-2015 08:06 AM #5 ryanfuse (Member)

Thanks Zeno,

Voluum server should be a problem, I'm targeting Malaysia and there's a Singapore server. Edgecast is similar, they have server in Singapore.

That leaves my users. I'll try to isolate wether this is a problem by hand coding a landing page.

On another maybe related note, I noticed there's a lot of 500 error on my CDN "hits" report, its the same with AWstat.

The deeper I dig, the more i am stuck at this level, trying to create a decent infrastructure to serve websites. Sigh.


04-25-2015 10:27 AM #6 caurmen (Administrator)

Very interested to hear your results from this. My suggestion would be the same as what you're doing, I think - hand-code a very simple lander with no external dependencies at all, host it on a Malaysian VPS, and see what the results are then.


04-25-2015 02:49 PM #7 ryanfuse (Member)

I hand coded a new lander, its about 7 times smaller in size, all using your guide Caurmen.

I then split tested it against the older lander - and so far the difference in CTR is about 2 to 3 times. (the new lander has higher CTR, not statistically significant)

As for RUM, the speed is the same with the old lander, I suspect this is because the new lander has not been propagated to all the EdgeCast nodes yet. We will have a clearer picture tomorrow i think.


04-25-2015 05:36 PM #8 rdigital100 (Member)

are you still using that server in singapore then?

make sure to run all images through compression down to the lowest you possibly can. We have always seen big improvements by putting that effort in to work on compression.

I use caesium on windows, heard good things about jpegmini as well.


04-27-2015 11:38 AM #9 ryanfuse (Member)

A bit surprised by the result i have now.

I am split testing 3 landers - the loading time is as follow 3.7s, 2.1s, and 2.8s

either i'm doing something wrong or EdgeCast sucks for Asia.

The smallest lander is performing at 2.8s, which is just 30kb, vs the other two lander at 200kb

I have no idea how long EdgeCast's nodes need to "warm up", but it has been 48 hours.

Rdigital100 - i use imageoptim - i'm on Mac.


04-27-2015 12:22 PM #10 caurmen (Administrator)

Try out MozJPG - it's pretty impressive these days for image size reduction.

Yeah, that does sound like EdgeCast do not win in your geo. Might be time to test out some other CDNs or a VPS in that specific geo.


04-27-2015 01:16 PM #11 zeno (Administrator)

1. What exactly is your setup with EdgeCast? Are you actually using EdgeCast, or someone who uses part of their network? They aren't an entry level CDN. I doubt this is due to their network/CDN being sub par. In fact its quite the opposite - Edgecast is at the forefront of CDN infrastructure and DNS. Content is usually pushed to the edge as a result of demand so you will have to drive traffic to actually get it there. This will depend on your load... but if driving hundreds of visits that will certainly push your content to the edge.

2. I suggest you show some actual RUM data and maybe some synthetics e.g. from Pingdom, Site24x7, CopperEgg, GTmetrix or whatever tickles your fancy so we can look at the waterfalls and figure out what's going on.


04-27-2015 02:48 PM #12 ryanfuse (Member)

1. Not i am not using them directly, i got their service from FutureHosting, maybe they're using a small part of EdgeCast. Maybe thats why the experience is underwhelming for me.

Yes i drove traffic to warm up the edge, for 48 hours, even when the traffic is not making me a profit just to test it out. Definitely more than hundreds, 952 visitors for today alone.

I'll test out MetaCDN and see how it goes.

2. Ok, I'll post some screenshots. Thanks

Click image for larger version. 

Name:	screenshot_26.png 
Views:	34 
Size:	191.4 KB 
ID:	6553

Click image for larger version. 

Name:	screenshot_27.png 
Views:	32 
Size:	106.9 KB 
ID:	6554

Ok, cache hit looks low, maybe another reason why the loading speed is low.
Click image for larger version. 

Name:	screenshot_28.png 
Views:	28 
Size:	84.7 KB 
ID:	6555

Click image for larger version. 

Name:	screenshot_34.png 
Views:	32 
Size:	134.7 KB 
ID:	6559

First lander - smallest size - i will call it "mysmall"

Click image for larger version. 

Name:	screenshot_29.png 
Views:	33 
Size:	170.5 KB 
ID:	6556

Ok something wrong DNS takes too long to load - take another shot

Click image for larger version. 

Name:	screenshot_33.png 
Views:	32 
Size:	135.5 KB 
ID:	6557

Ok another lander - created using webflow - i will call it "my"

Click image for larger version. 

Name:	screenshot_31.png 
Views:	37 
Size:	273.1 KB 
ID:	6558

There's another lander - but its basically same with "my" , just different language.

Note: all traffic comes from Malaysia split tested into 3 lander.

Thanks again for looking over the screenshots Zeno, I'll provide any data you need as I'm eager to learn.


Home > Paid Traffic Sources > Mobile