Hacker News Reader

Comment by gte910h | Hacker News Reader

The constant instances seem like a bid to take over the heroku type space.

There were a lot of complaints from people that had infrequently used (new) web apps, and so there would be a substantial delay to spin up an instance when the app did get hit. This makes your site seem "slow" even though it really isn't, which is a big problem if you are trying to build some momentum, or show off your site to potential investors. People were resorting to some less than ideal ways for keeping their instances warm, more information on the app engine for java group: http://code.google.com/p/googleappengine/issues/detail?id=24...

I believe you can use a feature called "Always On" to avoid the cold start problem. I noticed such an option in my billing settings for $0.30 / day.

The always on feature in december fixed half the issue (cold start)

This fixed the other half (long running processes).

Oh I've used some of the "less than ideal" things during demos before. It truly was an underserved need within the ecosystem.

That is only a problem for Java apps (and for Python apps using things like Django that were never really suited to AppEngine).

None of my AppEngine apps ever had any problem with 'cold' startup speed, and now with Go startup speed will be even less of an issue.

Backends are not primarily intended for user-facing traffic

I believe they are more like Tasks on Steroids than a way of having something like a little persistant VPS for your app.

I take it that you're talking about the new Backends:

Backends are special App Engine instances that have no request deadlines, higher memory and CPU limits, and persistent state across requests. They are started automatically by App Engine and can run continously for long periods. Each backend instance has a unique URL to use for requests, and you can load-balance requests across multiple instances.

Backends seem overpriced to me; their default backend is a long-lasting process running at 1.2Ghz allowed to use 256MB of RAM and it's priced $0.16/hour ($115/month).

They can shutdown/restart/relocate your backend whenever they want so what they suggest to keep your service live is "Configuring more backend instances than are normally required to handle your traffic patterns" (yeah that's at least another $115/month fee).

At linode that kind of money will give you 4G of ram, cpu power in the same ball park and loads of storage and transfer. It seems that Backends - and the GAE in general - makes most sense when you view the system as a whole, and work with systems of a certain size. Tasks goes in to queues and backends are started and stopped as needed - stuff like that. If all you need is to serve a long running process, linode and the like is more economical.

If you're actually using them as a backend, I'm not sure you couldn't start them up and shut them down.

But yes, they are expensive.

Reply on news.ycombinator.com