TRAP-Array: A Disk Array
Architecture Providing Timely
Recovery to Any Point-in-time
Authors: Qing Yang,Weijun Xiao,Jin Ren
University of Rhode Island
Presented By: Anuradharthi T
6/5/2008
1
Outline






Introduction
Background
Related Work
TRAP- 4 Architecture
Results
Conclusion
6/5/2008
2
Introduction
RAID Architecture
- Most prominent architecture advance in disk I/O
systems
- In use for more than two decades
- Types of RAID:


6/5/2008
RAID1- provides 2N data redundancy to protect data
RAID3, 4 & 5- store data in parity stripes across
multiple disks to improve space
efficiency & performance
3
RAID Architecture

Advantages:
- Can recover data from more than one disk failure
- Improves data reliability

Disadvantages:
- Recovery not possible when damaged data are not
confined to 1 or 2 disks
- Accounts for 60% to 80% of data losses
- Examples of such damages are: software defects, virus
attacks, power failure or site failure
6/5/2008
4
Solution
Timely Recovery to Any Point-in-time(TRAP)
- Keep log of all previous versions of changed data
blocks in time sequence
- Utilizes a fast & simple encoding scheme
=> less space
- Provides faster data recovery to any-point-in-time due to
drastically smaller amount of storage space used
=>improved performance
- Thus achieves an optimal space & performance
characteristics
6/5/2008
5
Background
Recovery of data in real world is measured by ‘2’
key parameters:
1. RPO (Recovery Point Objective)
- measures maximum acceptable age of data at time of
outage
2. RTO (Recovery Time Objective)
- maximum acceptable length of time to resume normal
data processing operations after an outage
6/5/2008
6
Classification of storage architecture


Storage architectures are capable of recovering
data upon an outage
Based on the 2 key parameters




6/5/2008
TRAP-1
TRAP-2
TRAP-3
TRAP-4
7
Related work

TRAP-1




TRAP-2





Uses periodical backups & snapshots
Time consuming & degrades application performance
Data transferred to tapes or disk for backup
Performs file versioning that records a history of changes to files
Versioning has to be done manually
Have controllable RTO & RPO
File system dependent
TRAP-3



6/5/2008
Keep a log of changed data for each data block in a time sequence
(time stamps)
Continuous Data Protection (CDP)
Huge amount of storage space required
8
TRAP-4 Architecture
Keeps a log of parities as a result of each write on the block
 Suppose a host writes into data block with logic address Ai
that belongs to a data stripe (A1,A2….Ab….An)
 RAID controller perform the parity calculation as follows:
Where, PT(k) – new parity for corresponding stripe
Ai(k) – new data for data block Ai
Ai(k-1) – old data of data block Ai
PT(k-1) – old parity of the stripe
6/5/2008
9
Optimizing the parity





P’T(k)=
is appended to the parity log stored
in the TRAP disk after a simple encoding
Only 5% to 20% of bits inside a data block actually change
on a write operation
Parity P’T(k) reflects the exact changes at bit level of new
write operation on the existing block
As a result, this parity block contains mostly zeros with a
very small portion of bit stream that is nonzero
Thus it can be easily encoded to a small size parity block to
be appended to the parity log reducing the amount of
storage space
6/5/2008
10
Block Diagram of TRAP-4 design
6/5/2008
11
Recovery on outage


Consider the parity log corresponding to a data block, Ai
after a series of write operations.
Log contains P’T(k) P’T(k-1) …. P’T(2) P’T(1) with timestamps

T(k), T(k-1), …. T(2) and T(1) associated with parities
When an outage occurs at time t1, and we would like to
recover data as at time t0 (t0 < t1)

Note that for all l = 1,2, … r
6/5/2008
12
System Architecture of TRAP-4
6/5/2008
13
Experimental Setup
6/5/2008
14
File System micro benchmarks
6/5/2008
15
Data size comparison
6/5/2008
16
Data Reduction ratio
6/5/2008
17
Retrieval time comparison
6/5/2008
18
Conclusions




A new disk array architecture capable of providing
timely recovery to any point-in-time for user data stored
in array
Up to 2 orders of magnitude improvements in terms of
storage efficiency
Has quick recovery time
Provides continuous data protection capability
6/5/2008
19
Thank You
6/5/2008
20