General discussion


RAID implementations

By editor's response ·
Tell us what you think about Mike Talon's advice on RAID implementations, as featured in the August 13th Disaster Recovery e-newsletter. Which level is best for your organization's recovery needs?

This conversation is currently closed to new comments.

Thread display: Collapse - | Expand +

All Comments

Collapse -

Nothing New Here

by clarence.hemeon In reply to RAID implementations

There is nothing new here.

Collapse -

Old news for some, but not others . . .

by MikeTalonNYC In reply to Nothing New Here

Agreed that RAID technology is nothing new to those who have lived in the IT world for some time. However the CFO who's never had to worry about anything outside his laptop can really use this information to justify the budget for next year's tech development =)

Mike Talon

Collapse -

good basics

by MMelb In reply to RAID implementations

Good primer on basic levels of RAID. Our organizations uses combos of RAID. I've got a server that uses RAID1 and RAID5 at the same time because some of the drives have different missions and require a different approach.

Collapse -

Good basic intro

by sussex pete In reply to RAID implementations

Most businesses have raid but it is coming into the SOHO market and this is the sort of intro that these people require, simple, clear and concise. I agree that there is nothing for the experienced techie but the person new to RAID will find it a great help.

Collapse -

News is news

by barb In reply to Good basic intro

I disagree that there is nothing for experienced persons. While it is raid at basic level - it also is very concise information to use for immediate training knowledge or to share with a boss or board as to why the other methods are or aren't n't suitable. It isn't that we can't come up with it but why reinvent the wheel, and rack our brains for the non-technical explanation.

Collapse -

Elementary my dear Watson

by draco vulgaris In reply to RAID implementations

This was hardly worth an article!

Instead, try answering a tough one, like how to duplicate the Merrill-Lynch trick (back on line four minutes after the WTC went down) without building, equipping, and staffing a duplicate data center! With hundreds of millions of dollars at stake, their budget must have been practically infinite; mine is not!

Collapse -

Right on the nose!

by MikeTalonNYC In reply to Elementary my dear Watson

Often in my dealings with clients I am asked to provide the impossible. "Give me absolute fault tolerance and high availability of all my systems with no budget."

I'll be the first to tell you it cannot be done.

I can give you Disaster Recovery - meaning that Recovery Time is rated in days, not minutes.

I can give you Data Protection, and depending on how much you have to spend that can be hours or mere seconds out of sync with the live data. It can be synchronous too, but we're talking small budget here.

I can even give you high-availability for your data-systems, but with severe limits imposed by bugetary concerns.

So you can have a DR plan, you can even have HA, but you can only have what you are able to budget for, just like anywhere else in the business world. If you want Enterprise Class DR and HA, you have to have an Enterprise Class infrastructure and budget to support it.

ML has the buget and infrastructure to support the level of High Availability they required to recovery from 9-11 in the time they did. Not every company can afford those systems, and my goal is to offer solutions that fit mulitple budgets, and multiple levels of DR needs.

Sorry for the marathon post, perhaps you'd like to see an article on this subject in a future column?

Mike Talon

Collapse -

RAID1 bad for heavily utilized systems?

by ChipN In reply to RAID implementations

Where did this come from? Certainly not benchmarks performed on modern systems! Read performance is usually better since the there are multiple drives servicing the read requests. And, write performance hardly suffers at all.

It is very easy to demonstrate that RAID 1 (mirroring) has almost no write performance impact on hardware mirroring implementations (typically 1% or less), and very little impact on software mirroring implementations (typically 3% or less). I would be happy to provide people with a simple tool to prove this for themselves on their own systems.

The biggest problem we have ever seen is on older or low-end systems that have very small caching controllers, or that use software mirroring on underpowered or overburdened processors. Those systems typically have performance issues to begin with, so any configuration change must be done with care.

RAID 5 typically has a 50%-100% performance degradation on writes (even when all disks are working), and is far from foolproof. I have seen at least 3 systems (different clients) that had RAID 5 failures that required a restore from tape (controllers, more than 1 disk, driver failure). RAID 5 seems to give people a false sense of security.

Anyway, emailme if you want the simple benchmark program (free). We have a white paper that discusses RAID for those interested ( - it provides a simple, common sense approach to disk configuration.

Chip Nickolett (

Collapse -

Writer's Response

by MikeTalonNYC In reply to RAID1 bad for heavily uti ...


While I agree RAID 5 is not foolproof (note I mention that other DR options be employed), it does offer advantages over not having any redundancy at all.

As for RAID 1 (mirroring) not causing slow-downs on heavily utilized systems, the controller must make two writes for every write processed, and two reads in most implementations. That's twice the I/O of other systems (though some create even more I/O overhead) and therefore less performance than a non-RAID system. Even Microsoft reccomends that heavily hammered areas of disk used by their most top-heavy software (i.e. Exchange 2000) be RAID 5, and NOT mirrored. The system info is on mirrored disks, but the logs are reccomended to exist on other RAID configured partitions.Mike Talon

Collapse -

Response to authors reply

by ChipN In reply to Writer's Response

Hi Mike,

Thank you for taking the time to respond.

Yes, there must be 2 writes, but only 1 write needs to occur for the program making the call to continue (providing it even waits for the physical sync). Different implementations of RAID 1 handle read requests in different manners. Years ago (mid-80's with VAX disk shadowing) the algorithm for reads was not the most efficient, but on most platforms you will see performance that equals or exceeds that of a single disk (i.e., non-RAID).
I have sent our benchmark to several people based on this thread. Maybe a few of them will weigh-in on their test results.

There is always disagreement when discussing RAID, which is why we run benchmarks and get hard data. That is the only way to optimize performance and understand the limits of a high-performance system.

Chip Nickolett

Related Discussions

Related Forums