Would Gluster help me with I/O limits of harddrives, if I chained a
lot of machines together?
If it would, it would replace that with.... network limits. I mean, seriously - if you distribute IO over 50 machines, your network needs to handle that. If you run into IO limits, not just storage area limits (i.e. terabyte) the proper solution is rather to invest into an IO focused solution. It is not like they do not exist.
Can "Gluster" serve FLV files without
any front-end server layers just using
the built-in HTTP protocol ?
No file protocol serves over HTTP ;)
I see Gluster like a mid size / storage oriented system.Yes, you get space and that without a lot of administration, and yes, you can use cheap machines - like putting big hard drives into every workstation and use them as file server type of thing.
But if you get LARGE (youtube) you may still need dedicated systems (even if they run Gluster), and if you run a LOT of IO, then a SAN style of infrastructure with content cluster serving front ends (i.e. partitioning the files so that not every server has to cache everything) is logically the only solution. Gluster is smart, it can not do magic. Caching on the actual serving server can be overloaded.
Moving IO from the discs to the network results in a LOT of very high requirements for the network. You may want to read up on Infiniband - it is actually mentioned in the Gluster documents. But that said, what is the problem with IO limits, if a cheap (USD 1000 range) Raid controller can handle about 200 SATA discs? You do not need a lot of machines (at higher costs) to get around IO limits. That will always be more expensive than dedicated boxes (as obviously you have to pay for more CPU, more RAM etc.). And you wont save any discs with Gluster - having fully redundant data storage is needed in both cases, with and without Gluster.
That said, using front end servers is advisable for a LOT of reasons. Even if Gluster can handle without, in a large installation, it would be utterly stupid to not use separated front end servers, preferably with some firewall in between.
Best Answer
I'm a big fan of iozone myself