LWN.net Logo

mainframe -> minicomputer -> PC -> smartphone

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 11:02 UTC (Sat) by djao (subscriber, #4263)
In reply to: mainframe -> minicomputer -> PC -> smartphone by dlang
Parent article: Why We're Fighting for an Open Cloud (Linux.com)

Simultaneous access sucks hard on NFS or AFS. The old protocols definitely don't win here.

I don't know what you mean by running applications directly from the network. We're talking about storage protocols, no? Storage is not related to applications. Dropbox is not intended for remote hosting of applications and nobody uses it for that purpose. There are different products such as EC2 for remote execution of applications; they would be the proper basis for comparison. If you mean that NFS supports things such as root filesystem over NFS, this is a comparatively easy problem to solve. For example put your filesystem image in the cloud and point your hypervisor to it (there are other ways too). Besides, the market demand for this feature is tiny compared with the market for personal storage.

Sorry for having to guess what you mean, but you really should be clearer about what you mean. To me there's no question that cloud storage solves a lot of really hard, previously unsolved problems that normal people find useful to have solved. In the end, if the old stuff does A, and the new stuff does B, and B is much more useful than A, then that's a big improvement, and one should not trivialize this accomplishment as merely "different functionality".


(Log in to post comments)

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 20:19 UTC (Sat) by dlang (✭ supporter ✭, #313) [Link]

B is only "much more useful than A" if you don't need to do A and do need to do B

I'm not saying that B isn't useful, but to say that it's better than A requires that it can replace A.

Dropbox and equivalent have their place, but they are not a replacement for NFS/CIFS/AFS or other network filesystems.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 20:37 UTC (Sat) by djao (subscriber, #4263) [Link]

The number of people who need to do B far exceeds the number of people who need to do A. It is therefore completely fair and accurate to say that, in aggregate, population-wide, B is much more useful than A.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 20:41 UTC (Sat) by dlang (✭ supporter ✭, #313) [Link]

that depends on the value of using A vs B

if A is used in the datacenters that provide the B functionality to users, then B would not exist without A and that would make A far more useful.

there are far more cars on the roads than trucks, but to say that cars are more useful, and therefor we should eliminate trucks would very quickly result in no more cars on the road either.

that's why I say that they are different, not more or less useful

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 20:55 UTC (Sat) by djao (subscriber, #4263) [Link]

I'm just guessing here, but I'm quite sure that most cloud storage providers do not use NFS, CIFS etc. in any way whatsoever.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 21:25 UTC (Sat) by Cyberax (✭ supporter ✭, #52523) [Link]

IMO, NFS/CIFS and other 'old' network technologies are too naïve to be useful in the current world. They were all designed for LANs and simply don't work with WANs.

And we actually don't have good solutions for WANs right now. But it's clear that the good old idea of 'network transparency' is dead.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 21:35 UTC (Sat) by dlang (✭ supporter ✭, #313) [Link]

I'm not saying that they do work for WANs, but they still work quite well for LANs.

you would not use something like dropbox for a LAN filesystem

and that's my point, the solution for WAN and LAN are just different, what will work well for one will not work well with the other.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 22:27 UTC (Sat) by Cyberax (✭ supporter ✭, #52523) [Link]

Well, we actually use DropBox across LANs simply because it works. And it also works across WANs.

It's clearly not optimal, sure. But it works everywhere.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 23, 2012 23:48 UTC (Sat) by dlang (✭ supporter ✭, #313) [Link]

but you don't use dropbox to replace NFS type network filesystems, you use dropbox to do different (probably archival type) things instead.

dropbox is not a replacement for NFS type things, it's a different beast that does different things.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 24, 2012 0:04 UTC (Sun) by Cyberax (✭ supporter ✭, #52523) [Link]

We actually DO use it somewhat like NFS. Our molecular biologists write code on their laptops in a folder which is synced with a folder on our large server.

Most of time they run things locally, but if they need a lot of computing power then they can log into the cluster server and run things there. And results will be synced back to their laptops. Yes, there's a small lag in replication, but it's just a few seconds usually.

I'd argue that this is a superior solution to using NFS.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 24, 2012 0:18 UTC (Sun) by dlang (✭ supporter ✭, #313) [Link]

following your logic, git is a better filesystem that NFS or dropbox because it provides version control features.

but I don't think that any of the git developers would take you seriously if you started saying that git was a replacement for NFS.
Although, in some workflows (like yours), git would probably work at least as well as dropbox

you are using dropbox to archive data and to access that archive from multiple places.

that is vastly different from the network filesystem approach where the application and user treat the remote storage like it was local storage.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 24, 2012 0:27 UTC (Sun) by Cyberax (✭ supporter ✭, #52523) [Link]

Yes, git would be great for these kind of tasks if it could do fast automatic replication. I've tried to move our scientists to git, but they resist. Indeed, DropBox works just fine for them - they don't need to do explicit commits, pushes and pulls. You just change a file and it's changed on the server.

Oh, and they're using rsnapshot for backups. So git's ability to store history is not really a selling point.

>that is vastly different from the network filesystem approach where the application and user treat the remote storage like it was local storage
As I see it, there are several main reasons for remote storage:
1) To be able to access large storage. It's not really a problem now when a typical notebook has more space than a SAN 20 years ago.
2) To share something. Dropbox works just fine for it in most cases when you don't need locking and sub-millisecond coherency.
3) To access archived data.

So DropBox works great for a lot of people simply because it doesn't try to do everything.

mainframe -> minicomputer -> PC -> smartphone

Posted Jun 24, 2012 6:20 UTC (Sun) by spaetz (subscriber, #32870) [Link]

I've tried to get something like dropbox-via-git with a smallish python app:
https://gitorious.org/synkie

It's a console app, inspired by sparkleshare and dvcs-autosync. As it works for me, and was written by me, it might have many rough edges though.

Feedback and patches welcome.
spaetz

mainframe -> minicomputer -> PC -> smartphone

Posted Jul 1, 2012 2:55 UTC (Sun) by jnahmias (subscriber, #16282) [Link]

Sounds like you want git-annex: http://git-annex.branchable.com/design/assistant/

Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds