More Codes Never Enough - PowerPoint PPT Presentation

1 / 23
About This Presentation
Title:

More Codes Never Enough

Description:

generalize to tolerate more than triple failures. STAR. slope 0, 1, -1 ... per node 2880 byte, XOR-based RS implementation from J. Blomer. Bit-Decoding. 15 ... – PowerPoint PPT presentation

Number of Views:13
Avg rating:3.0/5.0
Slides: 24
Provided by: cheng74
Category:
Tags: codes | enough | more | never | triple

less

Transcript and Presenter's Notes

Title: More Codes Never Enough


1
More Codes Never Enough
2
EVENODD Code
  • Basics of EVENODD code
  • each storage node as a single column
  • of data nodes k p (prime)
  • of total nodes n p2
  • encoding and decoding
  • pure XOR operations
  • MDS property (r 2)
  • tolerate any 2 node failures


3
EVENODD Code Encoding
parity I









  • Parity node I
  • Simple horizontal parity
  • Parity node II
  • Diagonal parity with adjuster complement

4
EVENODD Code Encoding
parity I
parity II









adjuster
  • Parity node I
  • Simple horizontal parity
  • Parity node II
  • Diagonal parity with adjuster complement

5
EVENODD Encoding
  • Numerical example

parity
data
0 0 0 1 0
1 1 0 0 0
0 1 0 0 0
1 1 0 1 1
1

6
EVENODD Code Decoding
node failures

















adjuster
  • Zig-Zag decode algorithm
  • Recover adjuster
  • Find a start point
  • Decode iteratively

7
r 3
8
STAR Code
parity III








  • Basics of STAR code
  • Extension of EVENODD code
  • EVENODD code 1 additional parity node
  • An efficient MDS code
  • Tolerating up to 3 node failures (r 3)
  • Encoding is straightforward

9
STAR Code Decoding
node failures






  • Decode algorithm needs to handle any 3 node
    failures
  • Special cases can be handled easily (parity
    failures)
  • e.g. parity node III among the 3 failures ? exact
    EVENODD decode
  • Difficult part is to deal with 3 information node
    failures
  • Key to efficient decoding

10
STAR Code Decoding (cont.)
node failures










  • In the 2nd column, the sum of any pair of cells
    with stride 3 can be recovered.
  • Starting with the last cell (zero), all cells in
    the 2nd column can then be recovered.
  • The remaining problem is to recover 2 node
    failures
  • apply EVENODD decoding

11
Comparison with Extended EVENODD Code
  • Similarities
  • pure XOR-based
  • (k3, k) MDS
  • Differences
  • Extended EVENODD
  • slope 0, 1, 2
  • generalize to tolerate more than triple failures
  • STAR
  • slope 0, 1, -1
  • geometric symmetry ? faster decoding

12
Decoding Complexity
STAR vs. Extended EVENODD
13
Decoding Performance
  • per node 2880 byte, XOR-based RS implementation
    from J. Blomer

14
Bit-Decoding
15
Bit-Decoding of EVENODD
16
Bit-Decoding of EVENODD
17
Bit-Decoding of EVENODD
18
Bit-Decoding of EVENODD
19
Optimal Updates
20
More on EVENODD
  • Encoding Complexity
  • Decoding Complexity
  • Update Complexity

21
Update Complexity
  • EVENODD 3 2/p
  • Lower Bound?

22
Update Complexity
  • EVENODD 3 2/p
  • Lower Bound 2 1/p
  • Gap 49

23
EVENODD-2?
  • Update Complexity 2 1/p
  • May be extended to r 3
  • r gt 4?
Write a Comment
User Comments (0)
About PowerShow.com