[http://roland.entierement.nu/blog/2012/01/15/looking-for-the-ultimate-distributed-filesystem.html] Roland, I had just a few comments on your characterization of Ceph that I thought I should share. :) "Availability/redundancy 1:" Saying Ceph "works" on the net is a bit of a stretch. It will probably not fail, but Ceph expects LAN-like latency and bandwidth throughout its design. "Availability/redundancy 2:" Ceph automatically balances all its data as new servers are added, old servers are removed, or current servers fail. If there's documentation somewhere that implies otherwise, it's in need of a great deal of work! This note is important in your context — if you host data on your personal computer and it gets turned off; Ceph will try and place other replicas elsewhere, and will go through a potentially-expensive sync operation every time you turn it on. (In practice it probably won't be too expensive; and it's very cheap for dedicated storage servers; but you might not appreciate it scanning directories every time you log in.) "Performance:" Ceph does allow configuration of data location, but it's important to understand that each replica of a file (/file chunk) is kept updated synchronously. So if you're doing writes then it's still going to go over your internet connection — it'll just go to a lot of other places too. Similarly, by default Ceph will only read off one of those servers (the primary), which is not necessarily the server closest to you. Some configuration of this is possible; whether it's enough for what you're seeking will depend on your computing environment. "Scalability:" The configuration you need is pretty limited, and is being reduced each time we work on it, so probably at some point all it will need is to be pointed at the monitor nodes…but yeah, there is configuration right now. In general I think there's a communications gap over what a "distributed" filesystem is (because the word is used very differently by different projects). Ceph is distributed in the sense that it doesn't have a Single Point of Failure, and the system's intelligence is spread across all the nodes; it is not distributed in the sense of being intended for use over the internet. (In contrast [to the best of my understanding], Tahoe-LAFS is distributed in both senses; and XtreemFS is distributed in the second but only partly in the first.) Indeed, the features you like in each project are largely correlated with their target use cases — Ceph is intended for use across a data center, or possibly across a fast WAN; Tahoe-LAFS is intended for secure long-term storage over the internet; XtreemFS is intended (best I can tell) for sharing research data over the internet, but not for frequently-updated personal data in several locations. >From your use case I think that Ceph is not the solution you are looking for. Tahoe-LAFS is certainly closer (though I'm unfortunately not familiar enough with the particulars of each of these projects to say for sure). :) You might also want to check out Andrew FS; which does not distribute authority but I think is designed for use cases a little closer to what you're after. -Greg -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
