TY - JOUR
T1 - Evaluating performance degradation due to block size increase in 3D NAND-based SSDs
AU - Shin, Ilhoon
N1 - Publisher Copyright:
© International Research Publication House.
PY - 2019
Y1 - 2019
N2 - Recently, 3D NAND has been proposed that dramatically improves the integration density of NAND flash memory. 3D NAND has advantages over conventional 2D NAND in that it reliably stores data, improves the integration density, and speeds up data writing, but there is a potential risk that the NAND block size increases significantly. When the block size increases, the number of valid pages of the victim block increases during garbage collection, which increases the latency of the garbage collection. As a result, the SSD performance may deteriorate. Therefore, in this study, the performance degradation due to the increase of the block size in the 3D NAND-based SSD is analysed and the following conclusions are drawn. First, the average performance is not degraded much. The average response time is maximally increased up to 7.1% in fin1 trace. This is because the latency of individual garbage collection is longer, but the total number of garbage collections is reduced. Second, the decrease in the tail performance is relatively large. fin1 trace have an increase of about 4.8x tail latency, and other traces have a longer latency of 5.6%-20.2%. Therefore, when using 3D NAND, we should focus on improving the tail performance rather than the average performance, and it is expected that the performance enhancement through clustering hot data and cold data separately is relatively large in 3D NAND. In other words, the block size increase of 3D NAND can be an opportunity for improving SSD performance.
AB - Recently, 3D NAND has been proposed that dramatically improves the integration density of NAND flash memory. 3D NAND has advantages over conventional 2D NAND in that it reliably stores data, improves the integration density, and speeds up data writing, but there is a potential risk that the NAND block size increases significantly. When the block size increases, the number of valid pages of the victim block increases during garbage collection, which increases the latency of the garbage collection. As a result, the SSD performance may deteriorate. Therefore, in this study, the performance degradation due to the increase of the block size in the 3D NAND-based SSD is analysed and the following conclusions are drawn. First, the average performance is not degraded much. The average response time is maximally increased up to 7.1% in fin1 trace. This is because the latency of individual garbage collection is longer, but the total number of garbage collections is reduced. Second, the decrease in the tail performance is relatively large. fin1 trace have an increase of about 4.8x tail latency, and other traces have a longer latency of 5.6%-20.2%. Therefore, when using 3D NAND, we should focus on improving the tail performance rather than the average performance, and it is expected that the performance enhancement through clustering hot data and cold data separately is relatively large in 3D NAND. In other words, the block size increase of 3D NAND can be an opportunity for improving SSD performance.
KW - 3D NAND
KW - Block size
KW - Garbage collection
KW - SSD
UR - https://www.scopus.com/pages/publications/85063637658
M3 - Article
AN - SCOPUS:85063637658
SN - 0974-3154
VL - 12
SP - 119
EP - 123
JO - International Journal of Engineering Research and Technology
JF - International Journal of Engineering Research and Technology
IS - 1
ER -