Advanced Topics in Distributed Systems Fall 2011 Instructor: Costin Raiciu.

Advanced Topics in Distributed Systems

Fall 2011

Instructor: Costin Raiciu

We’ve gotten used togreat applications

Enabling Such Apps is Hard

• Apps– Process huge amounts of data– Are fast– Are reliable

• One machine is not enough– Limited reliability, speed

• Super computers are expensive

What Makes These Applications Tick?

Distributed Systems

• Cares about technology relating to distributed systems:– Networks– Virtual machines– Distributed filesystems– Distributed computation

• We care about details, not about products– Why?

This course…

Traditional Data Center Network Topology

…Racks of servers

Top of Rack Switches

Aggregation Switches

Core Switch

1Gbps

10Gbps

10Gbps

Fat Tree Topology [Fares et al., 2008; Clos, 1953]

Aggregation Switches

K Pods with K Switches

each

K=4

Racks of servers

1Gbps

1Gbps

• Many operating systems running on a single box

• Provides:– Isolation– Flexibility– Better utilization of the machine

Inside a Machine: Virtualization

How do we store data?

• Distributed filesystem– NFS: • UNIX-like semantics• Single server• Limited scalability

– Google File System• Optimized for large-batch writes and sequential reads• Tolerates inconsistency

How do we get work done?

• Map reduce– Apply the same function in parallel on different

data on many machines– Aggregate results

• Useful for:– Building big web-search indices– Processing large amounts of data (PB)

This is just a taster

Course outline

• Distributed Apps we care about– Distributed Computation (Map Reduce, Driad, Hadoop)– Distributed Filesystems (NFS and GFS)– Web search– Caching (Memcached)– Distributed Hash Tables (Chord, Dynamo)– NoSQL databases (BigTable, Cassandra)

• Infrastructure: networks– Topologies: FatTree, VL2, Bcube– Using capacity: Hedera, MPTCP– Performance Optimizations: Incast, DCTCP

Course outline [2]

• Infrastructure: OS abstractions– Virtual Machines (Xen, VMM)– Distributed memory (Ivy)

• Security– Information Leakage– Good Isolation vs. High Utilization (Seawall,

CloudPolice)

Course Admin

• Lectures:– 2 hours per week, Tuesday 8-10 EC102

• Lab classes: – 2 hours per week, Tuesday 10-12 EG106– Project discussions– Help with practical issues– Help with high level goals, theory

• Website: curs.cs.pub.ro– If you have problems, let me know

Grading

• Project: 5p– Groups of 3-4 students– 4 stages: to help you get the job done easily,

without last minute work over Christmas• Exam: 3p• Presentation (1h): 1p• Class participation: 1p

Presentation

• Select one topic before the end of October (list will be posted this week)– Presentation date is fixed– If you miss your presentation, you lose 2p

• Class participation– 2 papers presented per course by your colleagues– Read them before and take part in discussion

Exam

• Open book• Need to understand and think– not memorize

• Studying 3 days before the exam won’t work– You need to take part in classes and read-up

Projects

• Large scale data processing with MapReduce– We will use Apache Hadoop– We will run code on Amazon EC2 (and maybe on

local clusters)– Several datasets you can choose from

Datasets available

• Crawled set of HTML pages from .uk• Wikipedia Page Traffic Statistics• Apache Mail Archives• Million Song Dataset• M-Lab dataset: Network Path and Application Diagnosis

tool• Human genome• US Census databases• Freebase data dump

Stage 1

• Choose dataset to use • Select one/many questions to answer using the dataset• Write small Hadoop script to parse a subset of the data• Come up with a few simple graphs (e.g. dataset size,

histograms)• Start writing:• Introduction to your report, problem statement• Start the implementation and evaluation

– Size of dataset, time to do one pass, etc.

• Strict deadline [1p]: November 1st

Stage 2• How do we solve the problem?

– Review related work– Select potential approaches

• Discuss pros/cons

• Implementation and evaluation– Implement the code– Run experiments– Refine code and reiterate

• Goal: 70% of functionality should be implemented• Deadline [1p]: December 1st

– Output in report:• Implementation section • Early evaluation section

Stage 3

• Final implementation• Evaluation• What did we learn?• Deadline [1p]: December 21th

– In class project presentation: 10 mins

Stage 4

• Write-up– Polish report– Create a coherent story– Convince me that this is useful

• Deadline to hand-in final report: last day of semester (January 14th) [1p]