Flash Solid State Disk Write Endurance in Database Environments

Table 4: E-Disk® Endurance at 100 GB Erase/Write per day using a Block Size of 8KB



Table 5: E-Disk® Endurance at 100 GB Erase/Write per day using a Block Size of 16KB

Table 6: E-Disk® Endurance at 100 GB Erase/Write per day using a Block Size of 64KB

Based on the tables above, you will notice that endurance values for 16KB and 64KB block sizes are the same and that the smaller the block size, the more write/endurance cycles are consumed. This is because an optimum number of writes to Flash chips is achieved when the I/O block size is a multiple of the physical block size, which in this case is 16KB. For I/O block sizes smaller than the E-Disk® physical block size (i.e. 1KB, 2KB, 4KB, 8KB), every I/O corresponds to a write in a 16KB physical block on the E-Disk® Flash drive.

Given that most random OLTP applications have block sizes smaller than 16K, one might suggest that sequential type applications would have longer endurance figures than OLTP applications. This is only the case, however, because the E-Disk® write cache is disabled. With write cache enabled, this difference between block sizes may not be as much. Also, future E-Disk® implementations might have variable physical block sizes where an optimum number of writes may be achieved even with differing I/O block sizes.

BiTMICRO Erase/Write Endurance Verification and Testing

An important item to point out regarding write endurance limits is that the endurance figures derived in the tables are based on a 1,000,000 erase/write endurance threshold limit commonly specified by most major Flash memory chip manufacturers, including that of BiTMICRO's. Furthermore, a minute portion (much less than a thousandth) of the sample population would turn into a bad block when the 1 million write cycle threshold is reached for that block.

To verify the manufacturer's technical specifications, BiTMICRO conducted accelerated Erase/Write endurance verification and testing involving 10 blocks each on 16 random samples of 256 Mbit Flash chips on a 3S40 E-Disk® that is configured to have a 2.6 MB storage capacity and with "cache disabled." Since the 24 hours per day and 7 days a week testing that began in November 2000, no error has surfaced during the first 10 million Erase/Write cycles on each of the 160 blocks.

To verify the Flash memory stability at extreme temperatures after every half a million Erase/Writes on each block, the sample 3S40 E-Disk® was subjected to power cycling (boot-ups) and temperature cycling every 2 hours for 24 hours at -42.5 °C and +87.5 °C extreme operational temperatures and baking at -57.5 °C and +127.5 °C extreme non-operational temperatures for additional 24 hours on each case. Under these stressful test conditions, only one (1) correctible error so far was encountered on one (1) of the 160 random samples of blocks after 15 million Erase/Writes on that block. These tests indicate that E-Disk® write endurance can be more than 15 times the values derived in Tables 2 to 6.

OLTP Environment Example

For a more real-world scenario, let us take the case of a government agency in Texas running a mid-size OLTP application. They implemented their OLTP application using an Oracle9i database on an eight-processor Dell 6000 Server running Windows Advanced Server, directly connected to an external Fibre Array. With more than 2 TB of data stored, they process about two million transactions per day.

General Hardware Setup:

Page: 1 | 2 | 3 | 4 | NEXT