The Five Minute Rule and HANA

Posted by Robert Klopp on September 17, 2012

More by this author

In 1987, and again in 1997, Jim Gray and Gianfranco Putzolu published famous papers (referenced here) that suggested a Five Minute Rule for managing data in memory. Simply put, the rule is based on the suggestion that it costs more to wait for data to be fetched from disk than it costs to keep data in memory so we can determine how often can you fetch data from disk before it makes economic sense to just keep it in-memory.

Since the number of CPU cycles per second is increasing and the cost of processors is decreasing, as is the cost of memory and of storage, the break-even point changes over time. In 1997 the break-even point was five minutes for a 4KB block, hence the label of the five minute rule. In 2009 these numbers were revisited by Goetz Graefe. He found that for a 4KB block of data, the point where the cost of memory equals the cost of SATA disk storage is even for any data accessed within 90 minutes.

Today, the break-even point is a little lower at about an hour (see here) as disks are faster and memory, although improving in price/performance, has not dropped as fast in price.

Let me point out the implications for HANA. If you have a table, no matter how large, that is touched by a query at least once every 55 minutes, it is less expensive, in raw hardware costs, to keep it in memory than to read it from disk. It does not matter if the table is 200TB; according to the Gray/Putzolu formula if it is frequently accessed it is less expensive to store it in memory.

If the data is compressed then the time it is cost-effective to keep it in-memory goes up with the compression. A database with 2X compression should stay in-memory for 2 hours with 3X compression for 3 hours and so on. Since we have seen compression over 10X with HANA you can see that in many cases any data that is accessed daily can be cost-effectively stored in-memory.

While we often talk about the ROI from HANA coming business benefit… there is also a case to be made for efficiently using IT hardware resources to manage IT costs. The in-memory architecture of HANA provides a most efficient use of hardware and the famous Five Minute Rule postulated by Gray and Putzolu demonstrates the cost-effectiveness perfectly.

Note: the original technical note and the Five Minute Rule was devised in part by Jim Gray, one of the pioneers in database systems. You can read more here.

VN:F [1.9.22_1171]
Average User Rating
Rating: 0.0/5 (0 votes cast)