sporc: group collaboration using untrusted cloud resources
DESCRIPTION
SPORC: Group Collaboration using Untrusted Cloud Resources. Presented by: Nipun Sehrawat. Cloud Deployment: Pros & Cons. Focus on user-facing applications Real-time collaboration Concurrent access Anywhere-available Is Cloud suitable? - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/1.jpg)
1
SPORC: Group Collaboration using Untrusted Cloud Resources
Presented by:Nipun Sehrawat
![Page 2: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/2.jpg)
2
Cloud Deployment: Pros & Cons
• Focus on user-facing applications– Real-time collaboration– Concurrent access– Anywhere-available
• Is Cloud suitable?Availability, scalability, elasticity, global-accessibilityTrust the cloud providers!
![Page 3: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/3.jpg)
3
Overall Goals
• Deploy user-facing apps on a cloud– Real-time collaboration– Offline operations– Generic servers
• Deal with untrusted servers– Data confidentiality– Detect misbehavior– Recovery
![Page 4: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/4.jpg)
4
Design Overview
Server
Client 1 Client N
Server’s role:- Global order of update
messages- Store (encrypted)
history
Update message = “diff” against a particular state in the history
![Page 5: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/5.jpg)
5
Problem 1: Consistency
Server
Client 1 Client N
Inconsistent State!Diff applied to wrong base file.
Optimistic Update
![Page 6: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/6.jpg)
6
Solution: Operational Transformation
• OT allows synchronization of states– No locks– Optimistic local updates
Client 1 Client 2
ABCDE ABCDEdelete(2) ACDE delete(4) ABCE
delete(2)delete(4) ACD ACE
![Page 7: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/7.jpg)
7
Solution: Operational Transformation
• OT allows synchronization of states– No locks– Optimistic local updates
Client 1 Client 2
ABCDE ABCDEdelete(2) ACDE delete(4) ABCE
delete(2)delete(3) ACE ACE
Transform a foreign ops past any new local ops
![Page 8: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/8.jpg)
8
Problem 2: Malicious Server
Server
Client 1 Client N
- Different Views of the state!- Histories “Forked”
![Page 9: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/9.jpg)
9
Solution: Fork* Consistency
• Clients send:Update Message + Hash (chain) of its history
• Receiving clients compare with their hash, fork if mismatch.
• Can forks go undetected?
![Page 10: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/10.jpg)
10
Fork* Consistency
Client 1 Client 2
1
2
3
1
2
3
1
2
3 4
1
2
3 5
Some other client generates OP 4 and 5.Server forks states of client 1 and client 2
Fork detected as soon as these clients interactwith some other client
1
2
3 4 5Correct history:
![Page 11: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/11.jpg)
11
Is Fork* Consistency good enough?
• What about banking operations?
Account information maintained on two machines
Hack the server to fork their states
Withdraw $10K from both machines
Balance in account : $10K
![Page 12: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/12.jpg)
12
Detailed Design
Invariant 1: Local Coherence
![Page 13: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/13.jpg)
13
Detailed Design
Invariant 2: Fork* consistency
![Page 14: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/14.jpg)
14
Detailed Design
Invariant 3: Client-Order preservation
![Page 15: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/15.jpg)
15
Detailed Design
Committed History = Base FileUpdate message = Diff wrt a Committed History
![Page 16: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/16.jpg)
16
Membership Management
New Shared KeyShared Key (for efficiency)
- Old key needed to decrypt updates encrypted using it (new clients)
- Key shared with the new client- New client generates current state from Ops stores at server
![Page 17: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/17.jpg)
17
Evaluation
• Two configurations:– Unloaded: Single client sending updates– Loaded: Every client sending updates– One server, four machines with multiple clients– Experiments on LAN
• Metrics:– Latency: “In-flight” time– Server throughput– Client time-to-join
![Page 18: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/18.jpg)
18
Latency
- “Server processing” increases as broadcast to more clients
- Client overhead nearly constant
- “Client queuing”, “Client processing” increase
- “Server Processing” also increases
![Page 19: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/19.jpg)
19
Server Throughput
More payload => More processing/packet
More payload => Processing overhead/byte decreases
![Page 20: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/20.jpg)
20
Related Work
• Venus:– Only for key-value stores– Requires a “core set” of clients to be online– Membership can’t be revoked
• Depot:– Applications-logic for conflict resolution– Client eq. to server, can also tolerate faulty clients
![Page 21: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/21.jpg)
21
Discussion
• What are a typical requirement of the targeted apps?– Latency?– Bandwidth?
![Page 22: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/22.jpg)
22
Discussion
• Is the server needed at all?– Limited role:
• Assign increasing sequence number to updates – clients receive updates in the same order (TCP used).
• Store history.– An alternative architecture?– Timestamp server + P2P communication
• Demerits?• Do all clients receive updates in the same order?
![Page 23: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/23.jpg)
23
Discussion
• How can server attack availability of the service?
• Any remedies?
![Page 24: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/24.jpg)
24
Discussion
• A new client has to apply all the operations to get current state. Can we do any better?– Check-pointing
![Page 25: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/25.jpg)
25
Thank you!
Questions?
![Page 26: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/26.jpg)
26
Client Time-to-Join
Text-Editor: OT is expensive
2000 pending ops
Key-Value Store: Conflicts only when keys match
More history => More diffs to apply
![Page 27: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/27.jpg)
27
Towards RobustDistributed Systems
PODC 2000 keynoteProf. Eric Brewer
Presented by:Nipun Sehrawat
![Page 28: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/28.jpg)
28
Do “classic” Distributed Systems work?
• “Working” DS like WWW, Napster etc are not “classic” DS.– No RPC– No modularity– No distributed objects
![Page 29: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/29.jpg)
29
Three issues in DS
• Where is the state?
• Consistency vs Availability
• Understanding Boundaries
![Page 30: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/30.jpg)
30
Where is the state?
• Classic DS:
• Data centers – consistency + availability
• Other locations: primarily for caching state.– Desktops, proxies
• DS shall acknowledge location distinctions
DataComputation
![Page 31: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/31.jpg)
31
Consistency vs Availability
• Classic DS: focused on ACID transaction semantics– A: Operation performed on all replicas or none– C: After an operation, all replicas end-up in same
state
What can be trade off for better performance and availability?
ACID makes application dev. easier
![Page 32: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/32.jpg)
32
Consistency vs Availability
• New paradigm: BASE– Basically Available – Soft-state – Eventually Consistent
![Page 33: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/33.jpg)
33
Brewer’s Conjecture
• Impossible for a web-service to provide following three guarantees simultaneously:– Availability – Non-failing node shall respond to every
received request– Consistency – Replicas updated as if all on a single
node– Partition Tolerance
• CAP theorem: Can have atmost two of these at a time
![Page 34: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/34.jpg)
34
CAP Theorem
ConsistencyAvailability
Tolerance to n/w Partitions
- Single-site clustered database- Can user 2-Phase Commit
![Page 35: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/35.jpg)
35
CAP Theorem
ConsistencyAvailability
Tolerance to n/w Partitions
- No replication or Quorum- Don’t reply if response misses from a data-node- Distributed databases
![Page 36: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/36.jpg)
36
CAP Theorem
ConsistencyAvailability
Tolerance to n/w Partitions
- Easiest!- Just reply with available data- DNS
![Page 37: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/37.jpg)
37
The DQ Principle
• For a given node: (Data/query) * (Query/sec) is constant
• What can a fault affect?
![Page 38: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/38.jpg)
38
Discussion
• How does SPORC tries to address:– Availability?– Consistency?– Partitions?
• Can a wide-area distributed database have both high consistency & availability?
• How can we make up for the lost consistency?
![Page 39: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/39.jpg)
Cumulus: Filesystem backup on Cloud
Michael Vrable, Stefan Savage, Geoffrey Voelker
![Page 40: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/40.jpg)
Background
![Page 41: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/41.jpg)
Background
![Page 42: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/42.jpg)
Backup on Cloud
• Simple
• Reliable
• Easy off-site storage
![Page 43: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/43.jpg)
What are the monetary cost of backup on cloud?
![Page 44: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/44.jpg)
Thin vs Thick cloudThick cloud: Highly integrated service
oGreater performance and EfficiencyoReduced flexibilityoLocking customers
Thin cloud: Build on top of generic layer
oMinimal interfaceoPortableoAdditional overhead
![Page 45: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/45.jpg)
Thin vs Thick cloudWhat is the right choice?
What is the penalty of using a thin cloud service?
Can one build competitive product economy around a cloud of abstract commodity resources or do underlying technical reasons ultimately favor an integrated service-oriented infrastructure?
![Page 46: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/46.jpg)
Cumulus
Filesystem backup on cloud designed under a thin cloud
Motivation:Explore thin-cloud vs thick-cloud trade-off in the context of backup
![Page 47: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/47.jpg)
Cumulus
• Cloud based backup systemoReliableoEasy offsite storage
• Portable
• Tuned for cost
![Page 48: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/48.jpg)
Prior work
![Page 49: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/49.jpg)
Design
• Storage Server Interface• Snapshot format• Storage Segment• Sub-file incremental• Segment Cleaning• Restoring
![Page 50: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/50.jpg)
Storage Server Interface
• Minimal interface: get, put, list, delete
• Operates on entire files
• Write once storage model
![Page 51: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/51.jpg)
Snapshot format Snapshot Descriptor
Date: Day1Root: A/0Segments: A B
![Page 52: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/52.jpg)
Snapshot format
![Page 53: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/53.jpg)
Snapshot format
What are the benefits of this format?
![Page 54: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/54.jpg)
Storage Segment
![Page 55: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/55.jpg)
Storage Segment
What are the benefits of grouping data together into segments?
Is there any downside to grouping data into segments?
![Page 56: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/56.jpg)
Sub-File incrementals
Is the design of Cumulus suitable for incremental changes?
![Page 57: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/57.jpg)
Removing old snapshot
![Page 58: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/58.jpg)
Segment Cleaning
![Page 59: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/59.jpg)
Segment Cleaning
![Page 60: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/60.jpg)
Segment Cleaning
![Page 61: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/61.jpg)
Segment Cleaning
![Page 62: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/62.jpg)
Segment Cleaning
How do we decide when to clean a segment?
What other approaches can be taken to reduce space wastage?
![Page 63: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/63.jpg)
Alternative Scheme
If a new snapshot is refers to an object in less utilized segment, create copies of that data.
![Page 64: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/64.jpg)
Segment Cleaning• Clean based on utilization of a segment
outilization threshold = 0; no cleaningoutilization threshold = 1; aggressive cleaning with 100% utilization
• What is the right threshold?
![Page 65: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/65.jpg)
Evaluation• Trace based simulation
o A fileserver trace : research group settingo A user trace : home setting
• Simulator models the process of backing upo Tracks storage, network and segment overheado Single and multi snapshot scenario
• Optimal baselineo No more data is stored or transferred than neededo Transfers entire content of the file if any part
changeso Stores only one copy if multiple files have same
hash
![Page 66: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/66.jpg)
Is cleaning necessary?
Optimal => No space wastage
![Page 67: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/67.jpg)
Average Daily storage plot
• Larger segment size => larger overhead• Too little cleaning => larger overhead• Too aggressive cleaning => larger overhead
![Page 68: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/68.jpg)
Cleaning PoliciesBreakdown of storage overhead when using 16 MB
![Page 69: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/69.jpg)
Average daily upload
• Larger segment size => larger n/w overhead• Increasing threshold => increases n/w overhead
![Page 70: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/70.jpg)
Restoring from Backup• Complete restore
oExtract all filesoGet the root snapshot descriptoroDownload all segments referenced
• Partial restoreoRecover small number of filesoDownload segments with metadata of snapshot
oLocate segments containing file data
![Page 71: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/71.jpg)
Evaluation
What is the penalty of using a thin cloud service?
What are the monetary cost of using remote backup?
What are the additional benefits?
How does this prototype perform?
![Page 72: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/72.jpg)
Paying for remote backup
![Page 73: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/73.jpg)
Prototype evaluation• Prototype built on Amazon S3• Cost comes to $0.19 / per month for
1.26 GB of data storage• Parameters can be tuned to minimize
cost• Comparing it to existing tools : Brackup
and Jungle DiskoBoth will have much higher operation cost
• Thick cloud solutionsoMozy: integrated solutiono$4 + 0.5 / per GB per month
![Page 74: SPORC: Group Collaboration using Untrusted Cloud Resources](https://reader036.vdocuments.us/reader036/viewer/2022062411/568168d5550346895ddfca57/html5/thumbnails/74.jpg)
Thank you!