However, updated ‘MAID 2.0’ technologies are now available that provide varying levels of disk-drive ‘spin down’, and include intelligence that adjusts MAID levels based upon LUN access history. These improvements expand the reach of MAID considerably, and bring its power saving benefits into more ‘mainstream’ secondary (and primary) storage infrastructures.

MAID 1.0

Massive Arrays of Idle Disks (MAID) is an adaptation of the RAID acronym, with “Idle” replacing “Inexpensive” (or “Independent”). The concept has traditionally been to simply turn off (or stop spinning) disk drives that weren’t being accessed, since drives spin and consume power even when not in use. Of course, it’s not quite that simple. Since MAID is a drive-level operation, a MAID LUN would encompass an entire RAID group. This means that either a large volume of appropriate data must be available for the MAID LUN or a lot of MAID storage would go unused.

At the data management level, the frequency of data access, as well as its acceptable access time would determine whether it was an appropriate candidate for MAID. The original, or ‘MAID 1.0’ was a binary process, either the drive was spinning and ready for data transfer, or in a ‘sleep’ mode and unavailable. This fact likely hurt the overall adoption of MAID as it reduced the amount of data that IT was comfortable assigning to this technology.

IT, erring on the conservative side, probably kept the majority of data out of the MAID LUNs, due to uncertainty about its access requirements. Also, if data that was put on MAID and eventually accessed more frequently than anticipated, it could reduce the overall cost savings, as an entire LUN would get spun up to serve a small amount of data. Consequently, MAID was relegated to mostly backup, DR  and deep archive applications, where access requirements were well known and/or delays with MAID data retrieval were less of an issue.

MAID for non-backup data

The 80/20 rule, as it’s applied to primary (non-backup) disk storage, indicates there should be plenty of candidates for MAID outside of backup. Historically, only a fraction of the data stored on primary storage systems is frequently accessed, in theory, making all of these less-active data sets good MAID candidates. In practice however, it’s a different matter, since most current traditional MAID technology presents only one reduced power (spin down) level. Unlike older archive or backup data, these primary data sets typically have variable access histories. With the large access time difference between MAID and regular storage, the consequences of putting these data sets on MAID could be long wait times. With this ‘all or nothing’ choice, IT managers usually chose ‘nothing’ and restricted a large percentage of potential candidates from MAID LUNs.

What’s needed is a more granular implementation of MAID technology so that retrieval performance isn’t such an issue. If there were additional levels between the current ‘spinning’ and ‘spun down’ states, with corresponding access times, the risk of putting data onto MAID would be reduced. The result could be considerably more data classified as appropriate for MAID, including secondary and some primary data, instead of just backups, DR sites or deep archives.

MAID 2.0

Newer MAID technologies, like AutoMAID from Nexsan, represent this more granular application of the spin down process with varying response times. The exact numbers depend on the manufacturer, but typically, these more sophisticated MAID systems offer one or two levels between the full spinning mode and the fully spun down mode. Typical MAID levels could be:

Level 0 is the regular hard drive full-spin mode, the baseline with full power consumption and the shortest data access time.

Level 1 has disk read/write heads parked, or unloaded. Power usage is typically 15%-20% less than Level 0 and access time is still less than a second on the first I/O and full speed on all subsequent I/Os.

Level 2, in addition to parking disk heads, slows the platters 30-50% from full speed. This produces a comparable percentage of power savings over Level 0, but puts access time in the 15 second range on the first I/O and full speed on all subsequent I/Os.

Level 3 is the traditional MAID condition, where disk platters stop spinning, heads are parked and the drive is put into a ‘sleep mode’, although it usually remains powered on. The power consumption of Level 3 is typically 60%-70% less than Level 0 and access time is 30-45 seconds on the first I/O and full speed on all subsequent I/Os.

In addition to these new MAID levels, next generation MAID systems can also adjust the specific level that’s applied to each MAID LUN based upon its access history, with separate thresholds set for each level. This means that implementation consists of simply moving data to the MAID LUN and essentially turning it on. The controller then changes the MAID level the longer the data remains inactive.

With these new levels and dynamic controller a significant percentage of secondary storage and some primary storage could most likely be put on MAID and produce real savings. For example, data that’s active during the business day and inactive at night, like user directories, could be spun down when users go home and accumulate several hours of ‘down time’ each evening. Since the process is totally automated, weekend and holiday down time would be increased without IT involvement.

MAID, as it was originally developed was a good idea that never really caught on. A major reason was that too little data was appropriate for MAID, primarily based on IT’s comfort level. Too many applications couldn’t tolerate the wait of up to a couple of minutes to access data as drives spun up and IT wasn’t willing to take the chance these data would be recalled. This lack of adoption relegated MAID to the ‘fringes’ of data storage and generated rumors of its immanent demise.

‘MAID 2.0’ represents an evolution that should help bring this technology to the front of the data center. MAID technology has improved and now includes multiple levels of operation, each with different power usage and access times and the intelligence to adjust these levels dynamically. This granularity gives IT managers a system that’s applicable to a greater percentage of stored data and the automation that makes it as simple to implement as a traditional disk array. Instead of just backups, DR or deep archive, MAID is now an appropriate technology for data currently residing on ‘secondary storage’ and for many primary applications as well. Technologies like AutoMAID from Nexsan enable MAID to meet its original expectations and provide IT with the energy savings to address green initiatives without giving up performance.

Eric Slack, Senior Analyst

This Article Sponsored by Nexsan