Fastly 对大文件分发的改善

仔细看了下，3.x 里文档是这么写的：

Deliver the object to the client directly without fetching the whole object into varnish. If this request is pass'ed it will not be stored in memory. As of Varnish Cache 3.0 the object will marked as busy as it is delivered so only client can access the object.

而 4.x 是：

Deliver the object to the client directly without fetching the whole object into varnish. If this request is pass'ed it will not be stored in memory.

没有想到之前居然是这样的：

To illustrate this, let’s look at a download example. If a 25MB application is being served from an origin over a connection that’s giving each client ~375kb/s, the download will take about 70 seconds. If that application was cached on Fastly without the Stream-on-Miss feature, then the first client to get a miss would have to wait 70 seconds while Fastly fetched it from the origin, and only then would they start downloading from our edge server.

@Livid 我也没有想到，另外 100M 的限制也是有点，可能因为早期 Varnish 只缓存到内存？

我之前测试各种反代程序就是为了观察这种情况：
/t/141198