Queue - July/August 2008

The Five-minute Rule
20 Years Later
and How Flash Memory Changes the Rules

interval is about inversely proportional to the record size. Gray and Putzolu gave one hour for 100-byte records and two minutes for 4-KB pages.

The five-minute rule was reviewed and renewed 10 years later in 1997.² Lots of prices and performance parameters had changed (e.g., the price of RAM had tumbled from $5,000 to $15 per megabyte). Nonetheless, the break-even interval for 4-KB pages was still around five minutes. The first purpose of this article is to review the five-minute rule after another 10 years.

Of course, both previous papers acknowledged that prices and performance vary among technologies and devices at any point in time (e.g., RAM for mainframes versus minicomputers, SCSI versus IDE disks, etc.). Therefore, interested readers are invited to reevaluate the appropriate formulas for their environments and equipment. The values used here (in table 1) are meant to be typical for 2007 technologies rather than universally accurate.

In addition to quantitative changes in prices and performance, qualitative changes already under way will affect the software and hardware architectures of servers and, in particular, database systems. Database software will change radically with the advent of new technologies: virtualization with hardware and software support, as well as higher utilization goals for physical machines; many-core processors and transactional memory supported both in programming environments and hardware; ³ deployment in containers housing thousands of processors and many terabytes of data; ⁴ and flash memory

TABLE
1Prices and Performance of Flash and Disks

Price and capacity
Transfer bandwidth
Access latency
Active power
Idle power
Sleep power

$3 for 8x64 Mbit ---- ---- ---- ---- ----

$999 for 32 GB 66 MB/s API 0.1 ms 1 W 0.1 W 0.1 W

that fills the gap between traditional RAM and traditional rotating disks.

Flash memory falls between traditional RAM and persistent mass storage based on rotating disks in terms of acquisition cost, access latency, transfer bandwidth, spatial density, power consumption, and cooling costs. ⁵ Table 1 and some derived metrics in table 2 illustrate this point.

Given that the number of CPU instructions possible during the time required for one disk I/O has steadily increased, an intermediate memory in the storage hierarchy is desirable. Flash memory seems to be a highly probable candidate, as has been observed many times by now.

Many architecture details remain to be worked out. For example, in the hardware architecture, will flash memory be accessible via a DIMM slot, a SATA (serial ATA) disk interface, or yet another hardware interface? Given the effort and delay in defining a new hardware interface, adaptations of existing interfaces are likely.

A major question is whether flash memory is considered a special part of main memory or a special part of persistent storage. Asked differently: if a system includes 1-GB traditional RAM, 8-GB flash memory, and a 250-GB traditional disk, does the software treat it as 250 GB of persistent storage and a 9-GB buffer pool, or as 258 GB of persistent storage and a 1-GB buffer pool? The second purpose of this article is to answer this question and, in fact, to argue for different answers in file systems and database systems.

Many design decisions depend on the answer to this question. For example, if flash memory is part of the buffer pool, pages must be considered “dirty” if their contents differ from the equivalent page in persistent storage. Synchronizing the file system or checkpointing a database must force disk writes in those cases. If flash memory is part of persistent storage, these write operations are not required.

Designers of operating systems and file systems will
want to use flash memory
as an extended buffer pool
(extended RAM), whereas
database systems will ben-
efit from flash memory as
an extended disk (extended
persistent storage). Mul-
tiple aspects of file systems
and database systems
consistently favor these
two designs.