1. 1 CS 268: Lecture 22 DHT Applications Ion Stoica Computer Science Division Department of Electrical Engineering and Computer Sciences University of California, Berkeley Berkeley, CA 94720-1776 (Presentation based on slides from Robert Morris and Sean Rhea)

2. 2 Outline  Cooperative File System (CFS)  Open DHT

3. 3 Target CFS Uses  Serving data with inexpensive hosts: -open-source distributions -off-site backups -tech report archive -efficient sharing of music node Internet node

4. 4 How to mirror open-source distributions?  Multiple independent distributions -Each has high peak load, low average  Individual servers are wasteful  Solution: aggregate -Option 1: single powerful server -Option 2: distributed service But how do you find the data?

5. 5 Design Challenges  Avoid hot spots  Spread storage burden evenly  Tolerate unreliable participants  Fetch speed comparable to whole-file TCP  Avoid O(#participants) algorithms -Centralized mechanisms [Napster], broadcasts [Gnutella]  CFS solves these challenges

6. 6 CFS Architecture  Each node is a client and a server  Clients can support different interfaces -File system interface -Music key-word search node clientserver node clientserver Internet

7. 7 Client-server interface  Files have unique names  Files are read-only (single writer, many readers)  Publishers split files into blocks  Clients check files for authenticity FS Clientserver Insert file f Lookup file f Insert block Lookup block node server node

8. 8 Server Structure DHash stores, balances, replicates, caches blocks DHash uses Chord [SIGCOMM 2001] to locate blocks DHash Chord Node 1Node 2 DHash Chord

9. 9 Chord Hashes a Block ID to its Successor N32 N10 N100 N80 N60 Circular ID Space Nodes and blocks have randomly distributed IDs Successor: node with next highest ID B33, B40, B52 B11, B30 B112, B120, …, B10 B65, B70 B100 Block ID Node ID

10. 10 DHash/Chord Interface  lookup() returns list with node IDs closer in ID space to block ID -Sorted, closest first server DHash Chord Lookup(blockID)List of finger table with

11. 11 DHash Uses Other Nodes to Locate Blocks N40 N10 N5 N20 N110 N99 N80 N50 N60 N68 Lookup(BlockID=45) 1. 2. 3.

12. 12 Storing Blocks  Long-term blocks are stored for a fixed time -Publishers need to refresh periodically  Cache uses LRU disk: cacheLong-term block storage

13. 13 Replicate blocks at r successors N40 N10 N5 N20 N110 N99 N80 N60 N50 Block 17 N68 Node IDs are SHA-1 of IP Address Ensures independent replica failure

14. 14 Lookups find replicas N40 N10 N5 N20 N110 N99 N80 N60 N50 Block 17 N68 1. 3. 2. 4. Lookup(BlockID=17) RPCs: 1. Lookup step 2. Get successor list 3. Failed block fetch 4. Block fetch

15. 15 First Live Successor Manages Replicas N40 N10 N5 N20 N110 N99 N80 N60 N50 Block 17 N68 Copy of 17 Node can locally determine that it is the first live successor

16. 16 DHash Copies to Caches Along Lookup Path N40 N10 N5 N20 N110 N99 N80 N60 Lookup(BlockID=45) N50 N68 1. 2. 3. 4. RPCs: 1. Chord lookup 2. Chord lookup 3. Block fetch 4. Send to cache

17. 17 Caching at Fingers Limits Load N32 Only O(log N) nodes have fingers pointing to N32 This limits the single-block load on N32

18. 18 Virtual Nodes Allow Heterogeneity  Hosts may differ in disk/net capacity  Hosts may advertise multiple IDs -Chosen as SHA-1(IP Address, index) -Each ID represents a “virtual node”  Host load proportional to # v.n.’s  Manually controlled Node A N60N10N101 Node B N5

19. 19 Why Blocks Instead of Files?  Cost: one lookup per block -Can tailor cost by choosing good block size  Benefit: load balance is simple -For large files -Storage cost of large files is spread out -Popular files are served in parallel

20. 20 Outline  Cooperative File System (CFS)  Open DHT

21. 21 Questions:  How many DHTs will there be?  Can all applications share one DHT?

22. 22 Benefits of Sharing a DHT  Amortizes costs across applications -Maintenance bandwidth, connection state, etc.  Facilitates “bootstrapping” of new applications -Working infrastructure already in place  Allows for statistical multiplexing of resources -Takes advantage of spare storage and bandwidth  Facilitates upgrading existing applications -“Share” DHT between application versions

23. 23 The DHT as a Service K V

24. 24 The DHT as a Service K V OpenDHT

25. 25 The DHT as a Service OpenDHT Clients

26. 26 The DHT as a Service OpenDHT

27. 27 The DHT as a Service OpenDHT What is this interface?

28. 28 It’s not lookup() lookup(k) k What does this node do with it? Challenges: 1.Distribution 2.Security

29. 29 How are DHTs Used? 1. Storage -CFS, UsenetDHT, PKI, etc. 2. Rendezvous -Simple: Chat, Instant Messenger -Load balanced: i3 -Multicast: RSS Aggregation, White Board -Anycast: Tapestry, Coral

30. 30 What about put/get?  Works easily for storage applications  Easy to share -No upcalls, so no code distribution or security complications  But does it work for rendezvous? -Chat? Sure: put(my-name, my-IP) -What about the others?

31. 31 Protecting Against Overuse  Must protect system resources against overuse -Resources include network, CPU, and disk -Network and CPU straightforward -Disk harder: usage persists long after requests  Hard to distinguish malice from eager usage -Don’t want to hurt eager users if utilization low  Number of active users changes over time -Quotas are inappropriate

32. 32 Fair Storage Allocation  Our solution: give each client a fair share -Will define “fairness” in a few slides  Limits strength of malicious clients -Only as powerful as they are numerous  Protect storage on each DHT node separately -Must protect each subrange of the key space -Rewards clients that balance their key choices

33. 33 The Problem of Starvation  Fair shares change over time -Decrease as system load increases time Client 1 arrives fills 50% of disk Client 2 arrives fills 40% of disk Client 3 arrives max share = 10% Starvation!

34. 34 Preventing Starvation  Simple fix: add time-to-live (TTL) to puts -put (key, value)  put (key, value, ttl)  Prevents long-term starvation -Eventually all puts will expire

35. 35 Preventing Starvation  Simple fix: add time-to-live (TTL) to puts -put (key, value)  put (key, value, ttl)  Prevents long-term starvation -Eventually all puts will expire  Can still get short term starvation time Client A arrives fills entire of disk Client B arrives asks for space Client A’s values start expiring B Starves

36. 36 Preventing Starvation  Stronger condition: Be able to accept r min bytes/sec new data at all times  This is non-trivial to arrange! Reserved for future puts. Slope = r min Candidate put TTL size Sum must be < max capacity time space max 0 now

37. 37 Preventing Starvation  Stronger condition: Be able to accept r min bytes/sec new data at all times  This is non-trivial to arrange! TTL size time space max 0 now TTL size time space max 0 now Violation!

38. 38 Preventing Starvation  Formalize graphical intuition: f(  ) = B(t now ) - D(t now, t now +  ) + r min   D(t now, t now +  ): aggregate size of puts expiring in the interval (t now, t now +  )  To accept put of size x and TTL l: f(  ) + x < C for all 0 ≤  < l  Can track the value of f efficiently with a tree -Leaves represent inflection points of f -Add put, shift time are O(log n), n = # of puts

39. 39 Fair Storage Allocation Per-client put queues Queue full: reject put Not full: enqueue put Select most under- represented Wait until can accept without violating r min Store and send accept message to client The Big Decision: Definition of “most under-represented”

40. 40 Defining “Most Under-Represented”  Not just sharing disk, but disk over time -1 byte put for 100s same as 100 byte put for 1s -So units are bytes  seconds, call them commitments  Equalize total commitments granted? -No: leads to starvation -A fills disk, B starts putting, A starves up to max TTL time Client A arrives fills entire of disk Client B arrives asks for space B catches up with A Now A Starves!

41. 41 Defining “Most Under-Represented”  Instead, equalize rate of commitments granted -Service granted to one client depends only on others putting “at same time” time Client A arrives fills entire of disk Client B arrives asks for space B catches up with A A & B share available rate

42. 42 Defining “Most Under-Represented”  Instead, equalize rate of commitments granted -Service granted to one client depends only on others putting “at same time”  Mechanism inspired by Start-time Fair Queuing -Have virtual time, v(t) -Each put gets a start time S(p c i ) and finish time F(p c i ) F(p c i ) = S(p c i ) + size(p c i )  ttl(p c i ) S(p c i ) = max(v(A(p c i )) - , F(p c i-1 )) v(t) = maximum start time of all accepted puts

43. 43 FST Performance