Andre, For large files like that, you'll find that Riak CS is a much better fit. Riak is not currently designed to handle files above ~ 10MB. Riak CS should support the Accept-Ranges header just like S3. If you find a case where that's not true, please let us know. It's a bug.
A couple more thoughts inline: On Jun 11, 2013, at 4:55 AM, Andre Lohmann <lohmann.an...@googlemail.com> wrote: > Hi, I'm in the need of a comodity hardware driven CDN for Video Streaming. > > The Files that are uploaded can be up to serveral Gig in size (each). I need > this system to be automatically replicated and horizontal skalable, to have a > round-robin on all Servers (for a simple loadbalancing algorithm). > > As far as I tested, riak does match the replicating and skalability needs. > > I allready did some testinstallations on some virtualservers (riak v 1.3.1, > riak cs I had problems with following the installation instructions on Ubuntu > 12.04, it didn't worked, but I will try some more). I noticed, that the > "accept-ranges: bytes" HTTP Header is not set. Also, compared to an apache > box, the range requests (initiated by VLC) didn't seem to work on riak (when > seeking to the mid of the video, all previous Data was downloaded by the > client, jumping backward always crashed). > > So now I have a few questions about my case and which seems to be the best > System (pure riak or riak cs)? Also some questions about how it internally > works? > > For Example, I have 5 Boxes with replication to 3 Copies (which means each > file is stored at least on 4 Boxes). When uploading a file to Box 1, this > will be copied to Boxes 2-4. When fetching this file from Box 5, this box > itself needs to fetch the the file internally from box 1-4. Does it select > the Box by load, randomly, or does it fetch parts by all boxes? Will the file > this way also get cached/replicated to Box 5 too? To be clear, a replication value of 3 means there are 3 replicas. Not 3 _additional_ replicas. > > Am I able to set a nginx Proxy in front of each riak box, to masquerade the > filename and still be able to handover the range requests to the riak box? I don't know offhand if there are any gotchas with nginx and Accept-Ranges, but don't foresee any issues here. > > while riak is totally save against failing boxes what about riak cs and the > stanchion instance? Riak CS is stateless. If a Riak CS box has failed/is-failing/slow, you can contact any other one in its place. Stanchion should be run on a singleton node, and replaced if it fails. You can also use 'traditional' HA techniques to make stanchion handle failover. Note, stanchion is only used for a small percentage of requests, like bucket creation/deletion. Normal object CRUD requests do _not_ require stanchion. So, if it fails, only a small amount of requests will be affected until it is back running again. Reid > > kind regards > > andre lohmann > _______________________________________________ > riak-users mailing list > riak-users@lists.basho.com > http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com _______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
