[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: GlusterFS



On 07/29/2013 02:19 PM, Diego Spinola Castro wrote:
Hi, guys.
I can't disagree file persistence is the scalability botneck.

It is certainly an issue, I'm not sure it's the only one.

Despite
there're solutions like S3, no better than just write a file on a dir
and make it visible for all gears.

Correct.

Wondering how you guys are planning to solve it and if anyone has
intentions to start a workforce to test GlusterFS and build some
experimental implementation for Openshift.

There's a number of things being looked at internally, including GlusterFS. The biggest problem we've got right now is performance of those solutions. In the best case scenario of what I've tested you see a slow down of about 4x over local disk, in the worst case it's somewhere in the realm of 20x. What that means is that if you were to untar the Linux Kernel it takes about 30 seconds or so on a local disk, with persistent shared storage in the form of a file system it would take between 2minutes and 10 minutes. There are use cases that this wouldn't matter for, but in the general case it's a rather large pill to swallow.

There are, unsurprisingly, other ways to achieve data persistency, the way I'd recommend right now is to toss those files into something like mongodb and deal with the data there. I'll be the first to admit, it's not a perfect solution, but it's something that works now. The most serious issue though, is that your app has to be written to take advantage of it, and a lot of off the shelf apps don't.

AFAIK  some issues would be solved:

That would need to be solved?

MCS labels via fuse client:
Figure out how allow gears to write on Gluster mountpoint (application
wide, domain wide)

Ideas ?

there are a whole slew of issues beyond this, particularly when you take the idea to the next logical conclusion of being able to share the storage across multiple gears.

Right now I would recommend using another mechanism to get that data persistency, sql database, nosql database, etc. I agree S3 isn't really an option, and filesystems aren't quite there yet as something I'd recommend.

If you, or anyone else, starts running any of this down on their own I'd love to collaborate on it, so drop me an e-mail or find me on irc (if you are on freenode I'm warthog9)

- John "Warthog9" Hawley


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]