5 ms·

I have been thinking of enabling this feature. Aside from this issue, does it make visible difference in terms of performance?

vladr 9y ago
It's been hard to measure since turning on Argo coincided with a larger rewrite of the backend. Overall performance is up but the real culprit might have been more of the optimizations on the origin.
- foobarbazetc 9y ago
  You guys should really retest with it off. Might save a bunch of money. :)
  - vladr 9y ago
    Already turned it off and will compare the numbers with the two weeks or so that it was on. Didn't have much of a choice as I did not want to continue racking up a larger bill from the abuse traffic being blocked.
    - foobarbazetc 9y ago
      Just to be clear: you're being billed for ingress? That's... odd.
      - mst 9y ago
        jgrahamc's already in the thread going "this isn't right, I'll be looking into it".
        foobarbazetc 9y ago
        Yeah but he's talking about blocked/filtered traffic. No CDN that I have used (and we've used Panther Networks, CDNetworks, Cotendo, EdgeCast, Akamai and Fastly) has ever charged us for ingress. Usually you pay for egress from edges and whatever is going between the edge and origin. So it would be good to get that clarified so we know.
foobarbazetc 9y ago
It really depends. If your origin has poor upstream connectivity then it might. We're on our second Argo experiment. So far it's actually made little difference for some locations and significantly decreased performance in others. Definitely not worth paying for... for us. You can just enable it, test your performance metrics and disable it.
- vladr 9y ago
  Also with the minor but fun benefit of paying $5 each time you toggle the on/off switch.
  - foobarbazetc 9y ago
    They fixed that bug after we (and others?) reported it. It now properly pro-ratas the $5 and even credits back the remainder of the month if you cancel it from the Billing tab. And let's you re-enable it. :)
- foobarbazetc 9y ago
  Replying to myself here but a pet peeve of mine (and all CDNs do this to some extent) is showing "improved" performance on something like... the Pingdom website tester (?) which is a one shot page and asset fetch. There's really no valid metric that can be derived from a one shot test. Just because the 2nd run with the magic on is "faster" doesn't mean anything. Hundreds of variables could have changed between runs. Most people use monitoring systems that run synthetics every minute and those are reasonable for an "overview" of trends, but they also don't show you the worst case times that are happening between runs. If you start getting more granular and monitoring at 5 or 10 second intervals you get to see more real world response times. On something like CloudFlare the averages (without Argo) are really good, but the max and standard deviation is very high. Hard to beat for $200/month but when you start paying per Gb you should start testing other providers.
mark242 9y ago
For a large ecom site hosted primarily in the US, for our European customers it cut TTFB almost in half.
- foobarbazetc 9y ago
  How are you testing this? I really wonder why we see the opposite result. Is the site on AWS? Maybe that's the reason for this.
- aisofteng 9y ago
  TTFB = time to first byte

vladr 9y ago

It's been hard to measure since turning on Argo coincided with a larger rewrite of the backend. Overall performance is up but the real culprit might have been more of the optimizations on the origin.

foobarbazetc 9y ago
You guys should really retest with it off. Might save a bunch of money. :)
- vladr 9y ago
  Already turned it off and will compare the numbers with the two weeks or so that it was on. Didn't have much of a choice as I did not want to continue racking up a larger bill from the abuse traffic being blocked.
  - foobarbazetc 9y ago
    Just to be clear: you're being billed for ingress? That's... odd.
    - mst 9y ago
      jgrahamc's already in the thread going "this isn't right, I'll be looking into it".
      - foobarbazetc 9y ago
        Yeah but he's talking about blocked/filtered traffic. No CDN that I have used (and we've used Panther Networks, CDNetworks, Cotendo, EdgeCast, Akamai and Fastly) has ever charged us for ingress. Usually you pay for egress from edges and whatever is going between the edge and origin. So it would be good to get that clarified so we know.

foobarbazetc 9y ago

It really depends. If your origin has poor upstream connectivity then it might. We're on our second Argo experiment. So far it's actually made little difference for some locations and significantly decreased performance in others. Definitely not worth paying for... for us. You can just enable it, test your performance metrics and disable it.

vladr 9y ago
Also with the minor but fun benefit of paying $5 each time you toggle the on/off switch.
- foobarbazetc 9y ago
  They fixed that bug after we (and others?) reported it. It now properly pro-ratas the $5 and even credits back the remainder of the month if you cancel it from the Billing tab. And let's you re-enable it. :)
foobarbazetc 9y ago
Replying to myself here but a pet peeve of mine (and all CDNs do this to some extent) is showing "improved" performance on something like... the Pingdom website tester (?) which is a one shot page and asset fetch. There's really no valid metric that can be derived from a one shot test. Just because the 2nd run with the magic on is "faster" doesn't mean anything. Hundreds of variables could have changed between runs. Most people use monitoring systems that run synthetics every minute and those are reasonable for an "overview" of trends, but they also don't show you the worst case times that are happening between runs. If you start getting more granular and monitoring at 5 or 10 second intervals you get to see more real world response times. On something like CloudFlare the averages (without Argo) are really good, but the max and standard deviation is very high. Hard to beat for $200/month but when you start paying per Gb you should start testing other providers.

mark242 9y ago

For a large ecom site hosted primarily in the US, for our European customers it cut TTFB almost in half.

foobarbazetc 9y ago
How are you testing this? I really wonder why we see the opposite result. Is the site on AWS? Maybe that's the reason for this.
aisofteng 9y ago
TTFB = time to first byte