| Posted by SagoDrew, 12-23-2010, 07:36 AM | |
Hey guys,
I know there is alot of debate about cloud computing, and how a virtual environment may be the best solution for RAID over IP, but has anyone ever gotten some kind of solution like this working?
The goal: offer RAID1 or RAID5 to systems that would not normally support it, or do not have room for the spare for parity or striping.
Expandability would be a key factor in this too if anyone has any input on a working solution.
|
| Posted by (Stephen), 12-23-2010, 07:47 AM | |
Why do the RAID portion over IP, why not leave that to the backend SAN where the extra parity stays off the network? Seems much easier to me to ensure SAN redundancy than 'RAID' over IP.
I'd love to see the debate over this though
|
| Posted by NoSupportLinuxHostin, 12-23-2010, 10:22 AM | |
I agree with Stephen that a SAN based solution would make a lot more sense. Actually, either a SAN or a NAS would work. For best performance, you would want the SAN or NAS to be a local box if it was for primary storage. It is generally not something you would offer for people to use over a typical WAN connection, unless it was just a backup solution.
|
| Posted by OpenInternet-Vince, 12-23-2010, 11:56 AM | |
There is drbd you could look into, but that is limited to 2 nodes.
|
| Posted by GusBouras, 12-23-2010, 01:05 PM | |
could I do this?
node1 ->->->- node2 ->->->->->->->- node3
sync.... when done, sync.....
|
| Posted by nwmcsween, 12-24-2010, 05:41 AM | |
There is no such thing as "RAID over IP". What you want is a distributed file system, current distributed file systems are listed on wikipedia.
|
| Posted by MikeDVB, 12-24-2010, 05:42 AM | |
What is it exactly that you're wanting to achieve?
|
| Posted by SagoDrew, 12-26-2010, 07:10 AM | |
Thanks for all the followups everyone. I'm trying to get a better grasp of cloud computing as possible redundancy options I can make available to clients as an alternative to having a duplicate server ready to swap out if any hardware is damaged. It would be nice to throw a new drive or blank server in and just sync up from a SAN of some sort.
|
| Posted by (Stephen), 12-26-2010, 08:10 AM | |
Don't Sync off a SAN, run off a SAN, but just remember, no matter how redundant a SAN design, it is not a replacement for proper backups.
|
| Posted by RapidRick, 12-28-2010, 09:53 PM | |
As posted, there is no such thing as raid over IP as far as I can imagine.
What you want is storage without a SPOF. There are many ways to do this. Some methods are open source and others costs 100k+ from vendors.
You can do something as simple as two iscsi iniators on the server from different server targets, then raid1 them. That would be a hellish rebuild though. You want to keep your data redundancy of the server in my opinion.
I think your real goal is diskless servers.
|
| Posted by pubcrawler, 12-28-2010, 10:45 PM | |
There are RAID over IP solutions.
http://freshmeat.net/projects/chironfs/
http://freshmeat.net/projects/moosefs
Distributed file systems have such functionality... That's what those are. Many others out there.
ATA-over-Ethernet, iSCSI and other ethernet solutions are out there also.
SAN's are promoted widely, but are a single point of failure, rather exotic, locked into a single vendor. They also are limited on expansion. Compare a SAN to a cluster of cheap Intel boxes with a few disks on each. Really can be dirt cheap storage nodes with common off the shelf lowly components versus proprietary most costly stuff.
Having said all that, all storage solutions like this really have a glaring problem and that is the connectivity between your front ends and the storage. Much of the connectivity is just Gbit ethernet. Sure you can bond 2 or 4 together. Yes they make faster ethernet NICs but are $$$ per machine.
Many SAN's are using other forms of connectivity to the storage such as fiber channel.
|
| Posted by RapidRick, 12-28-2010, 11:09 PM | |
Both of those use fuse, you are not going to be getting great performance from my experience with anything running on top of fuse.
ataoe/fc/iscsi is more geared towards the transport layer. They can be used to solve the problem but are not direct solutions. I think what he wants is a single block device which has no spof. FC is great but redundant FC is expensive and complicated (think multiple fc switches).
I think drbd exported over iscsi is your best bet. Drbd is a bitch to fix under the gun when things go wrong, but with enough experience with it, it is the best option that I have come across.
|
| Posted by pubcrawler, 12-28-2010, 11:37 PM | |
FUSE overhead is indeed an issue and varies depending on the solution author(s). FUSE does work and is good generic filesystem conduit.
Question is what sort of performance is really needed? If that's a requirement steers solutions further
SPOF is a complicated issue with distributed filesystems. In theory when properly implemented and where you have ample storage nodes there is no SPOF.
Distributed filesystems are where to look for a solution. There are enormous installations of most of these. Government and computing labs are massively scaling most of theses.
Search freshmeat.net for distributed file system solutions and root through them. Big pile of homework
|
| Posted by SagoDrew, 12-31-2010, 07:40 AM | |
Thanks everyone for the input. I've searched through quite a few hosts companies, and have not found this level of redundancy offered without having to buy an enterprise level server that can manage this redundancy itself... is it simply never cost-effective?
|
| Posted by pubcrawler, 12-31-2010, 07:54 AM | |
Many hosting companies are lucky to even offer RAID as an option. Things like NAS, SAN, external storage, distributed storage, etc. are all way beyond their economics typically. Too much of the hosting industry is lowest common denominator which equates to selling very little product for very little income. Like the industry habit of shoving 1000 sites on a single server.
To answer your question, a distributed filesystem isn't exotic and it isn't costly either. At minimum you need to have full control of two servers minimum and ideally more. Many of these so called hosting companies have a rack full of servers. Many real sites operate on four or more servers. The gear is sitting there to create a distributed solution, even if it's just bonded Gbit NIC based to start.
Hard to beat the price of the software, which like most open source is free --- at least to get started -- other than your time.
The only way this might prove beyond your finances is if you are accustomed to massively shared hosting and are just moving into your own dedicated server or first colo unit.
|
|
 Add to Favourites
 Print this Article
|
