How much data are you really expecting to store? Unless you are looking at tens or terabytes or more I would consider something simpler. I myself have built a very redundant 9 TB NAS for our streaming servers. I looked at GlusterFS, et al but opted for a simpler arrangement--a distributed file system only made sense of we needed more than 20TB or so.
Do you know what your usage patters will be like? E.g.: Most activity on a small set of files is a lot easier to deal with than having activity distributed fairly equality.
Networking:
You say bandwidth is not an issue. But still...1 Gpbs is not that much when streaming video. If you have access to 10GoE networking (cards, CAT6, switches, Internet feed) then great. I've done a lot of LACP stuff to get our infrastructure beyond 1GoE and in hind sight wish we had just paid the up front price for 10GoE from the start.
Aim for CDR-based pricing. If you have the traffic, your costs will be pennies per GB
Consider hosting your servers in the same building as a major Internet Exchange. This potentially will let you off load traffic to ISPs for free. (Well, excluding the one-off costs.)
Hardware:
I've no heavy experience with FMS, but do have experience with our own proprietary streaming servers and Wowza for live streaming. A modern dual socket server with 6 cores per CPU can saturate a 10GoE NIC, depending on the server software.
Use high NICs. They are not that expensive.
Server software:
I would highly recommend going with a Linux-based systemas *nix accells at networking related stuff.
If the streams will all be on demand, and you really want to maximize efficiency, then consider running lighttpd with mod_264_streaming and do "HTTP pseudo streaming".
There are other exotic options that may be an option, like Erlyvideo, a Erlang-based streaming server. It looks like a promising platform to get more streams out of less hardware.
Do proper load testing to validate your server and OS configuration. There lots of defaults in the all of this stuff that could artificially limit your output.
The way we scale is by buying cheap commodity hardware and (probably) build the infrastructure ourselves using varieties of open source software.
Us too. It is amazing what a small team can do with commodity hardware and open source software!
发布评论
评论（1）
存储：
网络：
硬件：
服务器软件：
我们也是。一个小团队可以利用商用硬件和开源软件做出如此巨大的贡献，这真是令人惊奇！
Storage:
Networking:
Hardware:
Server software:
Us too. It is amazing what a small team can do with commodity hardware and open source software!