+ 12/09 meeting 張皓翔 陳俞任

+ Outline Synology DS207+ review Storage Virtualization - introduction Project Division Implementation Simulation Problems to be solved or discovered Reference

+ Synology DS207+ CPUARM 926EJ-S IP Userroot/admin Passwordcsie504 Firmware latest updatedDSM Modules installedarm-bootstrap, ipkg, snmp, mrtg, moc Service enabledssh, samba, itunes server, ftp NAS GPL source code can be download from

+ Storage Virtualization [5][6] Provides a logical, abstracted view of physical storage devices Provides a way for many users or applications to access storage without being concerned with where or how that storage is physically located or managed Enables the physical storage in an environment to be shared across multiple application servers, and physical devices behind the virtualization layer to be viewed and managed as if they were one large storage pool with no physical boundaries.

+ Venders of Storage Virtualization Software based Datacore EMC FalconStor Gresham Enterprise Storage Microsoft Hardware based EMC HP IBM NetApp StoreAge Sun HDS

+ Product specification It often concerns about property in the following table An overview article can be visited on [7] (2006)[7] Maximum Capacity Drive/Storage Support Maximum Transactions CostMaximum Throughput Largest Virtual Volume I/O CachingSecurity Features Implementatio n Reporting and Logging Maximum Disks Maximum Virtual Volumes Provisioning Support Snapshot Support Services Supported

+ Implementation approach Host-based Using a software layer above the physical device driver intercepts the I/O requests performing the meta-data lookup and I/O redirection. LVM in Linux, LDM in windows Our project is host-based Storage device-based A primary storage controller provides the virtualization services and allows the direct attachment of other storage controllers. Network-based When talking about Storage virtualization it is most commonly thought of as a network based device using fiber channel networks connected as a SAN The virtualization device sits in the SAN and provides the layer of abstraction between the hosts performing the I/O and the storage controllers providing the storage capacity.

+ In-band vs. out-of-band (network- based) In-band Symmetric virtualization devices actually sit in the data path between host and storage. All I/O requests and their data pass through the device. Out-of-band Asymmetric virtualization devices, sometimes called meta-data servers. These devices only perform the meta-data mapping functions.

+ Project Division Storage Virtualization Mechanism (power/ performance) Simulation File System Trace Collection Device Performance / power model Implementation Virtual Device Creation Power Control Need discuss more details with ACARD

Implementation Model Linux TCP/IPFile System Printer…SambaBT Virtualization A file-level, host-based, in-band storage virtualization NAS System Storage Devices

Memo.(11/21 meeting) Ramdisk Write 快 可做 write buffer ，但 size 越大 data lost 的機率越高 BT Download disk usage 高 sata disk is inappropriate Virtualization + power management BT download data 放在最省電的 storage device 省電需同時考慮寫入速度與耗電功率 Mechanism monitor BT download directory characteristics: file size, download speed, usage time, disk utilization

+ Virtual Device Creation Writing a device driver for our virtual device Our driver is between VFS and true hard disk or ramdisk device driver LVM[8] could be a reference

+ Linux LVM [8]

+ Power control Power control in Linux ACPI [10] [11] APM [12] Based on BIOS, older than ACPI ACPI [13] ACPI (Advanced Configuration and Power Interface) is an open industry specification co-developed by HP, Intel, Microsoft, Phoenix, and Toshiba. It establishes industry-standard interfaces enabling OS-directed configuration, power management, and thermal management of mobile, desktop, and server platforms.

+ Experiment to do… Test disk and ramdisk performance, build a performance table Use LVM to combine disk and ramdisk to a logical device, test its performance, and check if it works Previous result could be our implementation baseline

Virtualization Layer Simulation Model RamDisk Flash Memory HardDisk Input (Configurations, File-level trace ) Input Output (Performance Numbers) Output Caching Prefetching Power Consumption

J. Kim, J. M. Kim, S. H. Noh, S. L. Min, and Y. Cho. A space-efficient flash translation layer for compact- flash systems. IEEE Transactions on Consumer Electronics, 48(2):366–375, May MediaAccess Time ReadWriteErase DRAM60ns (2B) 2.56 μ s (512B) 60ns (2B) 2.56 μ s (512B) - NOR Flash150ns (1B) 14.4 μ s (512B) 211 μ s (1B) 3.52ms (512B) 1.2s (16KB) NAND Flash10.2 μ s (1B) 35.9(512B) 201 μ s (1B) 226 μ s (512B) 2ms (16KB) Disk12.4ms (512B) (average) 12.4ms (512B) (average) - MediaAccess Time ReadWriteErase DRAM60ns (2B) 2.56 μ s (512B) 60ns (2B) 2.56 μ s (512B) - NOR Flash150ns (1B) 14.4 μ s (512B) 211 μ s (1B) 3.52ms (512B) 1.2s (16KB) NAND Flash10.2 μ s (1B) 35.9 μ s (512B) 201 μ s (1B) 226 μ s (512B) 2ms (16KB) Disk12.4ms (512B) (average) 12.4ms (512B) (average) - 15X 100X 400X50X Storage Performance [9]

Storage Power Consumption Samsung 2.5in SATA SSD Fujitsu 2.5in SATA Standard HDD Seagate 2.5in PATA Extended Temp HDD Read/Write0.41W/0.41W1.9W/1.9W1.9W/1.7W Idle0.32W0.60W0.99W Standby0.24W0.13W0.26W Sleep0.24W0.13W0.26W

+ File system trace collection Auspex File System Traces [1] -timestamp- -type- -action- FID: -fid- OFF: -offset- SIZE: -size- HOST: -host- 1992, too old Other trace toolkits LTTng [2] SystemTap [3] Dtrace [4]

+ SystemTap script Trace Time Spent in Read and Write for Files Obtained from We’ll modify this script to watches each open, close, read, and write system calls on the system Our testing platform is ubuntu on VMware, and have problems…

+ Device Performance / Power model DiskSim[14] is an efficient, accurate, highly-configurable disk system simulator originally developed at the University of Michigan and enhanced at CMU to support research into various aspects of storage subsystem architecture. Microsoft has a SSD extension for it [15] Power model extension [16] 因為 disksim 未來可能變成論文的比較基準，我想利用未來兩個 禮拜的時間好好研究一下 disksim ，並在聖誕節左右提出初版的 simulation model 運作流程 Configuration

+ To Do List Collect file system trace with LTTng and systemtap, determine which is better. Reading book “Storage Virtualization: Technologies for Simplifying Data Storage and Management “ by Tom Clark, Addison Wesley, 2005“Storage Virtualization: Technologies for Simplifying Data Storage and Management “ Study some white papers from website bellow ml ml Virtualization.html Virtualization.html

+ Reference [1] [2] [3] [4] [5] [6] _Specific_Implemenationshttp://en.wikipedia.org/wiki/Storage_virtualization_- _Specific_Implemenations [7] ,00.html ,00.html [8]

+ Reference [9] J. Kim, J. M. Kim, S. H. Noh, S. L. Min, and Y. Cho. A space- efficient flash translation layer for compact-flash systems. IEEE Transactions on Consumer Electronics, 48(2):366–375, May [10] Linux ACPI Project [11] Linux ACPI HowTo [12] linux 电源管理 from IBM [13]

+ Reference [14] [15] b41019e2-1d2b-44d8-b512-ba35ab814cd4/Details.aspx b41019e2-1d2b-44d8-b512-ba35ab814cd4/Details.aspx [16] power extension – dempsey: Modeling-Hard-Disk- Power-Consumption