11

So, given the fact that you can only get hard drives of about 160GB now for the cheapest price, I find i have a ton of extra drive space on workstations spread throughout my office. I'd like to utilize this space as a redundant RAID-like array of some sort for low-priority files and backups.

I have about 100 workstations that need only 40GB of disk space to operate, which leaves about 120GB of spare space which i have partitioned off. I'm wondering if there is some way to use iSCSI or similar to mount those partitions in some kind of RAID-like array. I mean, that's something like 12.5TB of disk storage at my disposal which won't be used by the workstations (they're locked down and don't allow users to fiddle with it).

Also, the workstations are always on, so that's not an issue, but the system should be tolerant of occasional downtime due to reboots for various reasons (Windows, you know).

Are there any off-the-shelf solutions for this? What about open source? I'd rather not roll my own if I don't have to.

My server environment is ESX 3.5 with mostly Windows hosts. SQL Server, etc... workstations are all Windows.

2
  • You're saying...just for clarification...that you want a network-distributed storage system on your workstations? I don't think I've heard of that being used before. I know it's possible...from what I've read...on Linux, but didn't see it with Windows. Also you'd lose data if one workstation loses a drive. Aug 14, 2009 at 23:18
  • Thus my thinking that it would be raid-like, thus having the ability to reconfigure itself.. say keep 10 workstations as "hot spares". If one dies, it takes over one of the hot spares, rebuilds the array and you're bat in business. Aug 14, 2009 at 23:58

6 Answers 6

5

There are a few P2P backup systems out there. I was just reading about Tahoe, but there are others.

There is also GlusterFS, which can create a big storage pool out of a lot of commodity nodes. It's UNIX only, so to use it you'd have to do something odd, like deploy a virtual linux appliance running on each Windows machine and reshare it from one or two parent servers using Samba to make it look like a windows file server.

2
  • Wow.. Tahoe looks like what i'm looking for... i don't need all the encryption, but hey... can't hurt... and you only need 3 out of 10 nodes to rebuild.. nice Aug 15, 2009 at 3:48
  • Tahoe really seems to be nice... Now I want to find a reason to use it ! :D
    – edomaur
    Aug 15, 2009 at 10:14
2

Well, it's certainly not a typical way of doing it, and I really wouldn't suggest it for...well, anything other than an experiment, but you might be able to get OpenAFS to do it.

1
  • That looks very interesting.. and seems pretty mature, even though they consider it experimental.. Was hoping for some kind of commercial "mature" solution.. but hey.. i only plan to use it for stuff i have in less easy to access locations (like tape backup). Aug 15, 2009 at 0:15
2

OK I cannot help you with the iSCSI RAID solution, but I can help with the iSCSI. I have started using a product called StarWind - Free from www.starwindsoftware.com. That you would install on the workstations to share up the space. Then you would use the Microsoft standard iSCSI initator

Basically it gives you an iSCSI target. It says it only works for Windows 2003, but it might work for Vista or XP.

I have been using the StarWind software for a while and it looks almost too simple to be an enterprise product, but it does a very good job of offering space. I was able to successfully attach an ESX 3i host to to a Starwind iSCSI target which was a Dell GX110 with an IDE drive. Then I was able to install and run Windows 2003 on that LUN

I guess that you would use another piece of software to aggregate all those expose iSCSI luns into a iSCSI RAID.

1

Vembu Storegrid supports this functionality.

Flud is an upcoming option for this as well, and they have a list of similar systems.

3
  • Flud seems to only support Linux, but Storegrid seems interesting. Thanks. Nov 20, 2009 at 7:21
  • Also, the last release of Flud was January 2008, almost 2 years ago.. looks like a deadish project to me. Nov 20, 2009 at 7:24
  • I wasn't aware of that, but I mainly added that link for the list of other options.
    – menko
    Nov 20, 2009 at 10:11
0

Put a distributed compiler or perhaps a distributed 3D renderer on them and use the space for working set files? ^^

0

Cou can use CrashPlan Backup to use the nodes as backup targets.

1
  • There are plenty of backup systems that can use file shares as targets. What I need is something that can tolerate a small number of nodes not being available at any given time. The Tahoe system eas mentioned seems ideal for that. Aug 15, 2009 at 21:39

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .