3 ms·
So, I didn't downvote you, and I work at AppNeta. In fact, I spent a lot of time arguing for us to switch from a per-request pricing model to a per-host model.
by trjordan 12y ago
So, I didn't downvote you, and I work at AppNeta. In fact, I spent a lot of time arguing for us to switch from a per-request pricing model to a per-host model. Reasons we did it:
1) Per-request is unpredictable. Paying 2x because you got a traffic spike and had to auto-scale sucks. Paying 10x because you got a traffic spike and handled it without autoscaling sucks more. Getting capped at day 20 in the month because we didn't want variable pricing sucks the most.
2) Estimating cost is nigh-on impossible. If we charge per metric, quick, what will it cost per month?
3) The price we picked per host is, to some extent, based on the amount of data we collect and store. We store every traced request for 90 days, and do analysis on all of that as it comes it. The AWS bill to pay for all that analysis and storage isn't trivial.
The dinosaurs of the early 2000s got around this by saying "just pay for the license, and store it yourself!" And you ended up with customers that had 2 hours of data retention, because storing it all was expensive.
I guess the point I'm trying to make is, we're in a different league of data collection than something like Pingdom. There are serious costs associated with all that -- we're not colluding to screw you, we promise :)
Edit: Check out Twitter Zipkin and Etsy CrossStitch if you want open-source. The basic collection tech exists, but it's a pain to run yourself. At least, that's what we hear from customers.
- kolev 12y agoPer-host is also not well-suited for elastic computing. Pricing should be billed for host-hours instead, if you ask me. I'm familiar with zipkin, but it's an overkill for most projects; Etsy CrossStitch - although I follow Etsy's projects, I don't think I've seen it. Anyway, I think I've exchanges comments with you before in here, but the issue is that you guys have some large enterprise clients, and usually with companies like yours, pricing totally gets shifted towards that model. It's really too expensive to pay more for monitoring (although I greatly value the data), when it exceeds the cost of the server itself. At least, business people can hardly justify something with clear costs like hardware (cloud or not) to something without 100% clear benefit to them (as they don't use it; it's for the developers, who's job should be to write code that just works and doesn't require all these expensive babysitting tools, if you see my point). This is not to attack you or anything, but my experiences across multiple organizations. Worst case (at least with your competitor New Relic) is that you pay for internal dev servers as well. Come on, give me a break - how can you compare live with dev servers, really? This is the abuse that I'm talking about. Maybe it's a different situation with you, I don't know, but it's really unfair and a frustration for us.