Presentation is loading. Please wait.

Presentation is loading. Please wait.

Reliable Data Storage using Reed Solomon Code Supervised by: Isaschar (Zigi) Walter Performed by: Ilan Rosenfeld, Moshe Karl Spring 2004 Part A Final Presentation.

Similar presentations


Presentation on theme: "Reliable Data Storage using Reed Solomon Code Supervised by: Isaschar (Zigi) Walter Performed by: Ilan Rosenfeld, Moshe Karl Spring 2004 Part A Final Presentation."— Presentation transcript:

1 Reliable Data Storage using Reed Solomon Code Supervised by: Isaschar (Zigi) Walter Performed by: Ilan Rosenfeld, Moshe Karl Spring 2004 Part A Final Presentation

2 Problem: Cosmic radiation in space causes bit-flips, and therefore valuable information could be lost. Solution: All data stored will first be encoded via a Reed Solomon Encoder. A Reed Solomon Decoder will be used when data retrieval is required.

3 Reed Solomon Encoder Reed Solomon Decoder Storage Device Hi res Hi freq. input data (possibly) corrupted data (hopefully) corrected data Basic Project Function

4 How will we implement? RocketIO Link “Storage” Board Memory “Master” Board RS Encoder RS Decoder CPU resources will be used at a minimum Satellite Computer & Logic Satellite Data Storage

5 First Semester Goals Studying the system-on-chip design process. Studying the EDK environment. Studying the blocks that build our system: –PowerPC –Reed Solomon Encoder/Decoder. –PLB IPIF and bus transaction protocol. –RocketIO Implementing the Reed Solomon cores as slaves on the bus (CPU still very involved in the flow). Sending a packet through RocketIO Reference Design

6 PLB PowerPC RS Decoder RS Encoder PLB2OPB Bridge OPB UART Hyperterminal on DIGLAB PC IPIF FIFO IPIF First Semester Goal Arch.

7 A Few Numbers… PLB Frequency is 100 MHz The Reed Solomon data width is 8-bit. Thus, if we could theoretically perform an 8-bit write each clock cycle, our basic architecture would reach a throughput of 0.8 Gbps.

8 System Blocks

9 IPIF The IPIF is (as its initials suggest) an interface between the bus and the IP. It takes care of the transaction protocols on the bus and simplifies access to the IP. It also enables special features such as S/W Resetting, User Logic address ranges, interrupts, Bursting, DMA support and more.

10 IPIF – continued

11 Bus Transactions Bursting enables us to transfer data with a higher throughput!

12 Reed Solomon cores When creating the cores using Xilinx CoreGen, the following parameters are needed: –k: number of symbols per data block (to be encoded) –n: total number of output symbols (original data + check symbols) –s: number of bits per symbol The Reed Solomon code can detect n-k symbol errors and correct (n-k)/2.

13 We have chosen k=239, n=255, s=8, which are similar to G. 709 standard. RS cores

14 RS Cores – Detailed

15 FIFO We have used several different FIFOs in our design, all taken from Xilinx CoreGen. We have made three FIFO types for our purposes, the “plb_in_fifo”, “plb_out_fifo” and the “plb_fifo”. For this we have chosen a 1024x8bit FIFO.

16 PLB OPB UART Encoding data flow Writing to encoder is done via IPIF. When there is no write request the encoder is in bypass mode. Data from the encoder is directly connected to a FIFO. The FIFO is read via IPIF to the PowerPC and sent through UART to PC. PowerPC RS Decoder RS Encoder Hyperterminal on DIGLAB PC IPIF FIFO IPIF PLB2OPB Bridge

17 Decoding data flow The data is put through IPIF to a FIFO at the decoder inputs. Using a block called FIFO-Helper, data is sent every clock cycle to the decoder. The decoder also has a FIFO at its outputs where data is taken from via IPIF for PowerPC and sent through UART to PC. PLB PowerPC RS Decoder RS Encoder PLB2OPB Bridge OPB UART Hyperterminal on DIGLAB PC IPIF FIFO IPIF

18 RocketIO RocketIO can transmit and receive serial data at rates ranging from 620 Mbps to 3.125 Gbps. At these rates, in order to assure sampling at correct times, each byte is coded into a 10 bit sequence with enough edges. The 10 bit scheme allows for special “K- Characters”, which are used for various tasks, including packet start and end signaling (SOP, EOP) and byte-synchronization (Comma).

19 DCR RocketIO Ref. Design PowerPC PLB RS Decoder RS Encoder PLB2OPB Bridge OPB UART Hyperterminal on DIGLAB PC IPIF FIFO IPIF Packet Processing Engine Data BRAM RocketIO Transceiver

20 Mindspeed Physical Media Attachment Xilinx Coding Sublayer RocketIO Transceiver Diagram loopback 8B/10B Encode FIFO TXDATA 16 bit Serializer Transmit Buffer TXN TXP 8B/10B Decode Elastic Buffer RXDATA 16 bit Deserializer Comma Detect Receive Buffer RXN RXP We will use one of the following: 125 MHz x 16 bit x 10b/8b = 2.5 Gbps 156.25 MHz x 16 bit x 10b/8b = 3.125 Gbps

21 RocketIO Ref. Design The reference design has several disadvantages. Packets must be prepared (in other words, encoded) and stored in memory first, and only then sent. There is too much CPU resource waste. PowerPC RS Encoder PowerPC Data BRAM Packet Processing Engine RocketIO Transceiver

22 Original Second Semester Goals Building a CPU offload unit that will be a master on the PLB, perhaps using the DMA capability of the IPIF. Using another development board to simulate a storage device. Performing fast and reliable data transfers between the two boards using the RocketIO ports.

23 Original Second Semester Goal System Diagram PLB Rocket I/O SDRAM Memory Controller PowerPC RS Decoder RS Encoder Second Development Board (simulating storage device) CPU offload unit PLB2OPB Bridge OPB UART, etc.

24 OPB Our New “Master” Board Architecture PLB PowerPC PLB2OPB Bridge UART Hyperterminal on DIGLAB PC RS Encoding Unit IPIF Outgoing Packet Delivery Unit Outgoing Memory RocketIO Transceiver RS Decoding Unit IPIF Incoming Packet Receiving Unit Incoming Memory

25 Advantages Of This Architecture The CPU responsibility is only to give simple read/write orders. The rest of the flow is done in hardware. There is no need for the “CPU offload unit”. The unit we will create can be treated as a black box, and be added multiple times for multiple storage devices in a modular fashion. A very similar box would be put in the “storage” board.

26 “Storage” Board Diagram OPB PLB PowerPC PLB2OPB Bridge UART Hyperterminal on DIGLAB PC IPIF Outgoing Packet Delivery Unit “Storage” Memory RocketIO Transceiver IPIF Incoming Packet Receiving Unit Large SDRAM

27 Second Semester Schedule Building the outgoing delivery unit and the incoming receiving unit – 6 weeks. Interfacing the units to PLB – 1 week. Testing the architecture on “Master” board with RocketIO loopback – 2 weeks. Creating similar units for “Storage” board – 2 weeks. Testing architecture on “Storage” board with RocketIO loopback – 1 week. Integration of the two boards and final debugging – 2 weeks.

28 Demonstration Through the UART interface, we will demonstrate the following: –Encoding of 239 bytes into a 255-byte Block. –Destroying several symbols according to a randomization seed from the user. –Decoding and correction of the original block.

29 Thank you


Download ppt "Reliable Data Storage using Reed Solomon Code Supervised by: Isaschar (Zigi) Walter Performed by: Ilan Rosenfeld, Moshe Karl Spring 2004 Part A Final Presentation."

Similar presentations


Ads by Google