1. rain technology (redundant array of independent nodes)
Presented by:
Ravikrishnan.P

2. contents RAID Flavors RAID 0 RAID 5 rain architecture Features of rain
rain platform
rain project goals
Distributed storage
Advantages

3. INTRODUCTION
The name of the original research project is RAIN which stand’s for Reliable Array of Independent Nodes. The RAIN technology originated in a research project at the California Institute of Technology (Caltech),in collaboration with NASA’s Jet Propulsion Laboratory
A component that stores data across distributed processors and retrieves it even if some of the processors fail

4. RAID flavors Commonly used ones: RAID 0 RAID 1 RAID 5 RAID 10
Other types used…but rarely: RAID 2,3,4,6,50……

5. RAID 0

6. RAID 1 RAID1 is ‘data mirroring’.
Two copies of the data are held on two physical disks, and the data is always identical.

7. RAID 5
“Distributed Parity” is the key word here.

8. Network Connections Application MPI/PVM TCP/IP RAIN Ethernet Myrinet
rain s/w architecture
Servernet:Error correction
ATM:aysnchronous transfer mode protocol
PVM:parallel virtual machine
MPI:message passing i/f
Network Connections
Application
MPI/PVM
TCP/IP
RAIN
Ethernet
Myrinet
ATM
Servernet

9. rain Network Connections Application MPI/PVM TCP/IP RAIN Ethernet
Myrinet
ATM
Servernet
rain
Global state sharing protocol
Can use in different platform
Provide group communication
ip manage
Ip assign
Ip release
Pool of ip

10. Myrinet NIC card
Myrinet Switch:10Gb/s

11. Features of rain  Communication
Bundled Interface
Link Monitoring
Fault Tolerant Interconnect Topology
 Communication
 Group Membership: Identifies healthy nodes that are participating in the cluster
 Data Storage

12. rain platform Heterogeneous network of nodes and switches
bus network
switch
Computing/Storage Nodes
Multiple Network Interfaces
Variety of types of Networks
node
node
node

13. Rain testbed Myrinet & Ethernet. Come by and see our RAIN demos!
Myrinet Switches
10 Pentium boxes
w/multiple NICs
Myrinet & Ethernet.
Come by and see our RAIN demos!

14. Proof of Concept: Video Server
Video client & server on every node. A B C D
switch
switch

15. DATA STORED IN DISKS
Insufficient storage to replicate all the data on each node. A B C D
switch
switch

16. Node Failure A B C D
switch
switch

17. Node Failure A B C D
switch
switch

18. Node Failure
Dynamically switch to another node. A B C D
switch
switch

19. k-of-n Code Erasure-correcting code: a d+c b d+a c a+b d b+c a b c d
recover data
from any k of n columns
b =
a+b a +
d =
d+c c

20. Link Failure A B C D
switch
switch

21. Link Failure A B C D
switch
switch

22. Link Failure Dynamically switch to another network path. switch switchB C D
switch
switch

23. Switch Failure A B C D
switch
switch

24. Switch Failure A B C D
switch
switch

25. Switch Failure Dynamically switch to another network path. switchB C D
switch
switch

26. Node Recovery A B C D
switch
switch

27. Node Recovery Continuous reconfiguration (e.g., load-balancing).
switch
switch

28. Fault-tolerant interconnect technologies
Goal
To connect computer nodes to a network of switches in order to maximize the network’s resistance to partitioning S C
How do you connect n nodes to a ring of n switches?

29. A ring of switches N S a naïve solution
Given degree-2 & 4, how do we connect them?
= Node
= Switch S N

30. A ring of switches N S
a naïve solution
= Node
= Switch S N

31. A ring of switches N N S S S a naïve solution N N easily partitioned S
= Node N N S S S
= Switch N

32. resistance to partitioning1 2 3 4 5 6 7 8
nodes on diagonals
With switches in a ring, make connections as non-local as possible.

33. resistance to partitioning1 2 3 4 5 6 7 8
nodes on diagonals
With switches in a ring, make connections as non-local as possible.

34. resistance to partitioning
degree-2 compute nodes,
degree-4 switches 1 4 6 8 2 8 2
nodes on diagonals 7 7 3
tolerates any 3 switch failures (optimal)
generalizes to arbitrary node/switch degrees. 3
Optimal in sense that for degree 2 & 4, no construction can tolerate more switch failures while avoiding partitioning.
Also, generalizes to other switch networks, in particular a fully connected network. 4 5 5

35. Point-to-Point ConnectivityA
node
node
node ?
Is the path from A to B up or down?
Network
Local view: what node A sees.
node
node
node B

36. connetivity Bi-directional communication.
Link is seen as up or down by each node.
Node A
Node B
Model is Bi-Direct.: up/down with respect to both sends & receives.
Mechanism we’ll use is pings & time-outs.
{U,D}
{U,D}
Each node sends out pings.
A node may time-out, deciding the link is down.

37. group membership
Consistent global view given local, point-to-point connectivity information
ABCD
ABCD B A D C
link/node failures
dynamic reconfiguration
Crucial for distributed services/computation: who’s alive? Who’s performing what?
ABCD
ABCD

38. group membership Token-Ring based Group Membership Protocol
Our solution: simple yet effective
Begin with an initial ordering of the nodes into a ring.

39. group membership Token-Ring based Group Membership Protocol
1: ABCD A B
Token carries:
group membership list
sequence number
Sequence number is updated at every node. D C

40. group membership Token-Ring based Group Membership Protocol
1: ABCD B 1 A
Token carries:
group membership list
sequence number
Sequence number is updated at every node. D C

41. group membership Token-Ring based Group Membership Protocol1 A 2
Token carries:
2: ABCD
group membership list
sequence number D C

42. group membership Token-Ring based Group Membership Protocol1 A B 2
Token carries:
group membership list
sequence number
3: ABCD D C 3

43. group membership Token-Ring based Group Membership Protocol1 A B 2
Token carries:
4: ABCD
group membership list
sequence number D C 4 3

44. group membership Token-Ring based Group Membership Protocol5 A B 2
Token carries:
group membership list
sequence number D C 4 3

45. group membership
Node or link fails: 5 A B 2 D C 4 3

46. group membership
Node or link fails: 5 A B D C 4 3

47. group membership
Node or link fails: ? 5 A B D C 4 3

48. group membership
Node or link fails: ? 5 A B D C 4 3

49. group membership Node or link fails: If a node is inaccessible,5 A
5: ACD
If a node is inaccessible,
it is excluded and bypassed. D C 4 3

50. group membership Node or link fails: If a node is inaccessible,5 A
If a node is inaccessible,
it is excluded and bypassed.
6: ACD D C 4 6

51. group membership Node or link fails: If a node is inaccessible,5 A B
If a node is inaccessible,
it is excluded and bypassed. D C 7 6

52. group membership Node or link fails: If a node is inaccessible,5 A B
If a node is inaccessible,
it is excluded and bypassed. D C 7 6

53. group membership
Node with token fails: 5 A B D C 7 6

54. group membership
Node with token fails: 5 A B D C 6

55. group membership
Node with token fails: ? B 5 A ? D C 6

56. group membership Node with token fails: ? If the token is lost,5 A
If the token is lost,
it is regenerated. ? D C 6

57. group membership Node with token fails: If the token is lost,5 A
If the token is lost,
it is regenerated. D C 6

58. group membership Node with token fails: If the token is lost,5 A
If the token is lost,
it is regenerated.
5: ACD
6: AD D C 6

59. group membership Node with token fails: If the token is lost,5 A
If the token is lost,
it is regenerated.
5: ACD
Highest sequence number
prevails.
6: AD D C 6

60. group membership Node with token fails: If the token is lost,7 A
If the token is lost,
it is regenerated. D C 6

61. group membership
Node recovers: B 7 A D C 6

62. group membership Node recovers: Recovering nodes are added. 7 6 B D C

63. group membership Node recovers: Recovering nodes are added. 7 6 B
7: ADC D C 6

64. group membership Node recovers: Recovering nodes are added. 7 8 6 B
8: ADC D C 8 6

65. group membership Node recovers: Recovering nodes are added. 7 8 9 B
9: ADC D C 8 9

66. group membership Node recovers: Recovering nodes are added. 10 8 9 B D

67. group membership
Features: A B 10
Dynamic reconfiguration D C 8 9

68. distributed storage
disk
disk
disk
disk

69. distributed storage Focus: reliability and performance. 1010 10 101 11
disk
disk
disk
disk
1010 10
101 11

70. Rain project : goals
Efficient, reliable distributed computing and storage systems:
key building blocks
Applications
Identify and Develop Key BB
Networks, Comm., Storage, API’s.
Reliability, performance, functionality.
Distributed solns. that work.
Storage
Communication
Networks

71. advantages There is no limit on the size of a RAIN cluster.
There is no concept of master-slave relation.
A RAIN cluster can tolerate multiple node failure.
This is highly efficiency in traffic management.
New node can be added into the cluster to participate in load sharing

72. Conclusion High Availability Video Server High Availability Web Server
Distributed Check pointing Mechanism

73. Thank you
Any queries ?