Cloudflare raises TTFB by a huge ammount

[tech12]

Hi there,

We run our website on a colocation rack in Amsterdam. We have several Windows 2019 servers on a load balance, very well optimized.

We have noticed that our TTFB reported by gtmetrix, webpagespeedtest and other similar tools, is very large. More specifically:

For page A via Cloudflare: TTFB: 966ms
For page A locally on the server: Waiting (TTFB): 27ms

For page B via CloudFlare: TTFB: 964ms
For page B locally on the server: TTFB: 177ms

For page C via CloudFlare: TTFB: 356ms
For page C locally on the server: TTFB: 145ms

and the list goes on. The difference in some cases is huge !

We are on the Professional plan on Cloudflare.

Any ideas ?

Thanks
Alex
I

[M4rt1n]

@tech12

locally on the server

If you request your site from the server it actually get served by you actually just meassure no TTFB but TTG (Time To Generate)! So you actually compare these two times with each other:

1. How long your Server needs to build the page and would be able to serve itself internally
2. real TTFB which (also) gets routed through CloudFlare

This does not make any sense and yes it is very well known that CloudFlare increases the TTFB by about 10ms when proxying requests which have to be handled by your origin server. Cloudflare actually speeds up the PageLoad and SiteLoad and all your static assets. EVERY proxy service will actually slow down TTFB by sime ms and CloudFlare actually is doing this very good by just adding (for me) about 10ms to the TTFB.

@tech12

The difference in some cases is huge !

The difference is huge as you measure two completely different things. You compare “internal TTFB” (TTG) with real TTFB.

Also:

@tech12

For page A via Cloudflare: TTFB: 966ms
For page A locally on the server: Waiting (TTFB): 27ms

written down informations by yourself do not count and noone will really look into this unless you can provide “real” benchmarks which proves this. This actually is very easy by console with curl or by testing with WebPageTest, but you would have to provide us with both:

1. test while beeing proxied
2. test while bypassing CloudFlare

[MGS]

This is most likely to be related with:

High latency from some EU servers to cloudflare servers

Yes here we have an issue affecting West Europe. Since 2 April there is a latency increase (+100ms) between Cloudflare and OVH datacenters in europe except Poland. Please check carefully the 3rd graph on the OVH smokeping page: http://lil2-gra.smokeping.ovh.net/smokeping?target=ANYCAST.AS13335-3 . You’ll see clearly the problem. This should be fixed asap as it must be affecting all customers using CloudFlare and OVH together. [net-ovh-cloudflare]

We are waiting for some information as this is affecting a lot of customers for sure.