Mystic Journey
Table 2: DCP Feedbacks from the multi perspectives
Reliable Energy-Aware SSD based RAID-6 System
(FAST 2012, Feburary 14-17, San Jose, U.S.A)
Mehdi Pirahandeh,
Deok-Hwan Kim*
Dept. of Electronic Engineering, Inha University, Incheon, South Korea,, * Corresponding Author
Motivation and goals
 Challenges of SSD markets deal with
 Breakthroughs in terms of energy consumption, reliability, performance of SSDs
 Achieve optimal energy consumption for HDD, SSD and large storage systems
 Dynamic voltage measurement
 Auto power management
 Compiler directed energy optimization
 Energy flow of sequential/random read-write operation using proposed model
 Data pages are segmented into large chunks
 Use power switching of SSDs after writing of each chunk is done.
 After reading-writeing of current chunk in SSD Sj, IO operation of next chunk
is performed in SSD Sj+1
 Measure SSD reliability and choose the parity SSD Sj+4 and Sj+5
with less utilization level.
 Read operation needs to access four disks from Sj to Sj+3 sequentially
and skips two parity SSDs
 Performance and energy use of RAID systems with various type SSDs
 Reliability system and some criteria
 Cost for replication based schemes are expensive
 Erasure codes considering power management
Need to measure repair transition rate
 Average data loss during N iterations of IO operations on critically exposed sectors
 Lower utilization level produces lower failure rates
 When one disk fails, 1st parity SSD is “active” and 2nd parity SSD is in “sleep”
When two disk failure occurs, both SSDs are in “active” mode
 Write operation requires accessing six SSDs using power switching modes
Table 1: pseudo code of
reliable energy-fault
aware algorithm
 Goals
 Proposes an energy aware algorithm
 Model which enable SSDs to increase the performance and decrease the level of
power consumption
Proposed RAID System
Fig.1 Overall structure of reliable energy-aware SSD based RAID System
Preliminary Result
 Estimate power consumption manually for each SSD
 Multiple power modes - waken(1.2w), active(2.4w), sleep(0.06w), idle(0.5w) and off
 Average delay time between switching power mode smaller than that of traditional model
 Dynamic SSD scheduler can activate each disk adaptively and minimize time of waken
mode of SSDs.
Average data loss is calculated based on the Markov model and utilization level
 SSD energy consumption for busy sever-like workload
 Core layer in the RAID-6 controller
 Encoding and decoding scheduler
 Random and sequential read and write modules
with an internal failure detector using erasure codes
 Traces are updated using the log generator
 Reliability-aware layer
 Procedures for prediction of SSD reliability
 Use erasure codes generated from the core layer
 The estimation period is 2 sec.
 60% energy saving for read operation and 48% for write operation compared to the
general model
 We can see that the proposed model significantly reduces the energy consumption by
controlling the power switch of SSDs adaptively.
Table 2: SSD energy consumption for busy sever-like workload
 Energy-aware layer
 Pre-processing procedure
 Initialize the reliability measurement and utilization level
 SSD selector and pre-processing procedure.
 Set the status of selected SSD power mode to idle-sleep-active via imported traces
 Dynamic SSD scheduler
 Visualize the statistics of SSD energy consumption
 Update the power mode into idle, sleep or active.
 Host interface layer, flash translation layer and NAND flash chips.
Fig.2 energy flow of sequential/random read-write operations
Power Modes
Read operation
Write operation
Read operation
Write operation
 Improved
approach for periodic estimating the energy consumption of SSDs
The reliability estimation considered to enhance the energy efficiency on the SSD
based RAID-6 system
 A layered architecture for reliable energy-aware RAID-6 system
 Reduce energy consumption of parallel access of SSDs
Dynamically switching of SSD power modes among active, idle, waken and sleep
 Allowing one of parity SSDs in safe zone to be sleep mode
 Segmenting data pages into large chunks
 Using the power switching of SSDs after writing or reading of each chunk is done.
 Reduce energy consumption during active mode
 Activating SSDs sequentially and minimizing delay time of power switching of SSDs
 Avoiding repetitive accessing of same SSD by using large chunks
 Increasing number of disks in idle mode
This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science
and Technology (2011-0004114) and in part by the Ministry of Knowledge Economy(MKE) and Korea Institute for Advancement in Technology (KIAT) through the
Workforce Development Program in Strategic Technology.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF