In-Depth

Revivio's Point-in-Time Mirror Splitting Replacement: A Closer Look

Revivio’s Time Addressable Storage is an important technology that also happens to work.

• By Jon William Toigo
• 01/20/2005

Just before the holidays, I had the opportunity to travel to the snowy North, visiting the headquarters of Revivio Inc. in Lexington, MA, just outside Boston. About a year and a half ago, this little start-up fired a shot heard round the world (Okay, maybe not around the whole world, but at least around EMC headquarters in Hopkinton, MA) by claiming that they had new technology that would reduce cost and improve performance in continuous data protection.

At the time, chief technology officer Mike Rowan and long-time industry insider Kirby Wadsworth, who now serves as vice president of marketing and business development at Revivio, were talking about a replacement for point-in-time mirror splitting called Time Addressable Storage (TAS). With TAS, you can offload a function that consumes a lot of expensive big iron disk, mirror splitting, to less expensive arrays. At the same time, you could eliminate the huge software licensing fees that the big iron guys charged for their on-board mirror-splitting software.

There was little doubt in my mind that this upstart company was on to something important. I dug into their concept and wrote a white paper about it because it just made a lot of sense. External confirmation of my view came from Nick Tabellion at SOFTEK, who characterized TAS, together with Avamar Technologies’ data compression technology (called Commonality Factoring), as the two most important storage technologies to hit the street in years. Given Tabellion’s pedigree as a co-patent holder on IBM Systems Managed Storage, the endorsement seemed important.

Over the following 18 months, I talked up Revivio’s technology at all the conferences I attended. When the company was seeking additional capital, they sent prospective investors to me and I gave them an earful.

I told the investors that many of my clients were misusing capacity on their most expensive arrays to make useless data copies. At predetermined intervals, they would stop mirroring, quiesce all applications and flush all caches, write all the data to the mirror, and remove the mirrored disk from further use. Next, they would select more expensive capacity to create a new mirror, and restart the process.

This method produced multiple copies of data considered to be “crash consistent” at the time that they were made. The problem, from my perspective, was that the entire strategy had more holes in it than my favorite Tampa Bay Buccaneers t-shirt. Not only was the process a terrible abuse of a $180/GB disk that could be accomplished just as readily by using a $3/GB SATA, not only did it join consumers at the hip to proprietary hardware, and not only did it line the pockets of the big iron vendor and enable him to overcharge for his gear, it was also based on a totally erroneous assumption: that the mirrors have clean data.

To tell the truth, most database corruption events are not detected within 24 (or even 48) hours. By the time corruption is detected, corrupt data already exists on your point-in-time mirror-split volumes. You end up having to restore a very old copy of data from tape, assuming your vendor hasn’t convinced you that expensive mirror splitting is all you need.

Revivio, I said, had a better idea. Why not write block-level changes to an external disk—the cheaper the better? Just keep a running log of all the changes, then restore to a certain point in time when you have a failure. Revivio’s strategy is aided by the crash-recovery capabilities that have become the norm in database products from Oracle and others. DB software crash recovery can get you back to a certain point, and Revivio’s logs can get you closer to the point before the damage was done.

Plus, you don't need to be wedded to some proprietary hardware and you can reduce the overall cost of the continuous data protection solution. This is always good in my book.

The investors were persuaded, but to be honest, I felt a bit odd. I had seen the demo unit and read the engineering specs, but I had never actually seen the final product in operation. I phoned Kirby and Mike and told them that I couldn’t say anything more about their product until I saw it do what all of the specs said it would.

So, in late November 2004, I made the trek to Lexington to see if I was full of feathers. I was delighted to find that the prototype had yielded a production solution that was as promised. We failed a large 6 TB Oracle database deliberately, and restored it successfully, in about six minutes, to within microseconds of its state at time of failure. I was impressed, and you will be too.

TAS does not eliminate mirroring. You still need one copy of mirrored data for use in doing your restores. TAS just gives you better granularity when you go to restore data from the copy.

Admittedly, I didn’t see how TAS operated against other databases. A technologist for a large media company recently told me that he was testing a database journaling product from another vendor because Revivio didn’t support his preferred DBMS—or didn’t when he was surveying the market a year ago. So, do your homework to see whether the technology will work with your environment before purchasing a license.

For those who are using Revivio TAS or other point-in-time mirror-splitting replacement technologies, I’d appreciate hearing from you about your experience with these products. I’d also welcome input from traditionalists who believe they are still deriving some sort of value from a proprietary mirror-splitting play. Send your comments to [email protected].