Mario,

Thanks for continuing a great discussion.

I would like to point out that there are plenty of RAID solutions where you can simultaneously pull 8 drives out of 64 and keep running without data loss.

Best regards,
Steve Todd

Steve,

delighted to see you post here.

Could you please give an example or two? Just to make sure we are not skewed on semantics.

All the best

Mario

This artical lacks explanation on what is clustered storage.

Perhaps this article will help understanding clustered storage:
http://www.infoworld.com/article/07/02/26/09TCintro_1.html

Feel free to post again if more questions remain

Mario, geat spirit but RAID 6 can do this as well. You are incorrectly assuming that it only protects against two disk failures, which is the usual implementation but not the complete one. See this article: http://www.symantec.com/stn/articles/article_detail.jsp?articleid=raid_6_principles_38044. Keep up the good work, I read it regularly.

Mike,

Thank you for pointing out that very interesting paper. Reading it made me realize how much math I have forgotten.

Anyway, I don't dispute the theory, but shouldn't we compare actual implementations?

If there are RAID systems that implement resilience beyond a two drives failure I must have missed them.

Cheers.

Mario

RAID will not go away on individual use machines. In the gaming world, it is used to reduce the disk read/write time below physical actuality. I would wager that will be true in any small computing environment (cost, speed) for quite some time.

OK, maybe there is more to this than meets the eye but the article referenced above (http://www.infoworld.com/article/07/02/26/09TCintro_1.html) is basically an iSCSI implementation of virtual storage (which has been around for YEARS). iSCSI does not have a lock on virtual storage technology. It can be implemented on SCSI, Fibre Channel or even SATA and IDE drives (with the proper software).

Hi Mario,

I'll base my examples on EMC products because that's what I know.

If I have a 64 disk CLARiiON system, I can fail 8 disks and still keep running. Of course this depends on "which" disks fail. Same came be said of Symm.

My hunch Mario is that you meant to say that all 64 disks are part of the same RAID "unit".

Keep up the interesting posts,
Steve

Mario,
what makes you think that Equallogic (DELL) has node resiliance in their soltution? They do not. In fact they are not clustering at all. Their solution is merely volume spanning across RAID 5 or RAID 10 arrays. This is RAID 50 not synchronous replication. They call this storage peers not cluster nodes. BIG DIFFERENCE.

This is very misleading. RAID is behind Isilon, panasas and many other clustered files systems. Sun uses ZFS which has it's own raid like qualities imbedded in the FS. So this is not a death blow to RAID, but a totally different packaging that will allow RAID to move onto the next round

Mario,

Look no further, LeftHand’s Network RAID is here. LeftHand Networks delivers true scale-out clustering with Network RAID. A storage cluster can be configured to not only survive multiple disk failures in a RAID set, but also multiple disk array failures and even multiple site failures without losing access to your volumes. Imagine application volumes spanning 3 sites and 2 sites go down and your applications don’t miss a beat. And when the sites or nodes come back online they are incrementally rebuilt with no manual intervention required for failover and failback. Yes, LeftHand has been shipping this capability for years. And Network RAID can be deployed over traditional hardware RAID levels, offering numerous options for data protection and availability.

See “What Does Data Protection Mean to You?” at:

http://lefthandnetworks.typepad.com/virtual_view/

I think that RAID will be around for a long, long, long time, and it's unkind not to acknowledge the major step forward in storage capability it provides. Mass storage systems that virtualize above the level of the individual device, and do so reliably, can now be taken for granted.

Having said that, it's not a bad thing to visualize what could do the job better. One thing that has always struck me about RAID is the inflexibility of the array configurations. The requirement for matched hardware is not a particularly good fit for a hardware environment that evolves so quickly.

We all know the drill. Years after implementation a drive fails. So what, the array keeps running, right? Sure, but you've lost partial, or total, redundancy. You really must replace that failed unit. It's irresponsible not to.

Now you either have to buy a comparatively antiquated drive at premium prices (let's ignore the effect of maintenance agreements for the moment), or you buy a new(er) drive and the array ignores all the extra capacity and features of the new drive.

There was a company, several years ago in the dot com era, that offered Internet hosted virtual storage. The theory was that people would rent space on their systems to this company (kind of like subletting an apartment). The company then resold virtualized blocks of storage to anyone who wanted it. Everything was encrypted, multiply redundant, and as a loosely coupled system, didn't care if the components of the virtual storage blocks were mismatched. The redundant storage even theoretically enabled parallel I/O for much greater throughput.

It seems to me that there is an opportunity for some vendor to accomplish this kind of thing inside the corporate firewall. Sun's ZFS seems to have some attributes of such a system (although I have only seen parts of that system in action).

Brian,

thank you for that thoughtful comment. I regret if I was unkind to RAID, a technology that has been the pillar of storage for so many years.

Having said that, all too often customers are lead to believe that there are only two choices, RAID or nothing.

I guess my point was that there are other options that offer more reliability and better performance along with an attractive TCO.