A New Cached Data Recovery Mechanism
by Ken Wood on September 28, 2010
So this may fall in to the category of “So what? All of the vendors solve this problem somehow and why should I care?” True, this is one of those insurance features that you don’t appreciate until it happens to your data center. But, for the technical minded and those that like a cool technology description, this is for you.
How do you design an enterprise-level cache backup system to protect mission critical data with high-volume industry standard components and keep it green? By letting Hitachi come up with the idea.
Traditionally, the way to protect data written to cache from a failure scenario was to install banks of batteries to keep various parts of the subsystem powered up sufficiently long enough to either sustain the data in cache (non-volatile memory system) until power returns, or power the entire storage array and de-stage all data to the disks before the batteries run out. The latter can take a significant amount of time depending on how much cache is installed, how much written data is in cache and the number of disks that need to be written to. The length of time and the type of design used will determine the amount of battery power that is required, i.e., how many batteries are required.
With the announcement of the Virtual Storage Platform (VSP), we’ve introduced a new system for protecting your cached data that provides enterprise levels of protection while requiring a minimum amount of battery power, which reduces the overall footprint and is environmentally friendly.
How the Cache Data Recovery Mechanism Works
This new protection and recovery mechanism, designed as part of the Data Cache modules, is used to backup cached write data that has not yet been de-staged to the hard disk drive or solid state disk (SSD) arrays in the case of power-fail scenarios, component failure or other unscheduled outages. Cached write data is still mirrored in the data cache on separate power boundaries (different Data Cache modules). However, instead of using battery power to sustain the Data Cache modules during a power failure, write data in cache is backed up to an onboard flash-based SSD device. That is, the mirrored copies in cache are also copied, so double the protection.
This same protection mechanism also is used to backup the state of the system. A difference between previous generation systems is there is no mirrored operational version of Control Data.There is only one copy of the operational Control Data as part of the VSP because there is also a backup of the Control Data on the first Data Cache module pair. This Control Data backup is not a mirror, but a backup, and stores the state of the operation Control Data in an optimized recoverable format. Also, since any data written to the Data Cache modules are mirrored, the backup of the Control Data is also mirrored. This allows for three copies, one operational state and two backup copies (mirrors) to exist within the system. Also, as part of writing to the Data Cache modules, all write data is backed up to onboard SSD devices. This includes the entire Control Data backup.
The “So What” Factor
There are several advantages to this new approach in protecting the state of the system and user data. First, this design leverages the economies of using selected, readily available high volume components combined with special purpose Hitachi-developed components and microcode to protect critical data in an enterprise-level system. The standards for protecting this critical data has already been set by Hitachi in the previous generations of storage systems, and exceeding this standard by leveraging high volume components is a testament to Hitachi’s innovation capabilities.
Secondly, by using SSD to backup internal and user data, we’ve made the architecture more environmentally friendly. The reduction in batteries and battery backup circuitry and the reduced footprint allows for increased storage and storage controller densities, and the decrease in power requirements for supporting such subsystems is drastically improved.
Again, “so what?” All vendors of enterprise-level storage systems protect memory-based data to some degree, right? Again, true, but there are different ways to accomplish this and some are more economically and environmentally expensive than others.
Comments (1)
Christophe Bertrand » Blog Archive » Game on! on 01 Oct 2010 at 9:23 am
[...] more information about our announcements here and blog postings from Hu, Claus, Michael, David, Ken and [...]



