Greedy is Good: On Service Tree Placement for In-Network Stream Processing

Question

1. What are the contributions mentioned in the paper "Greedy is good: on service tree placement for in-network stream processing" ?

2. What have the authors stated for future works in "Greedy is good: on service tree placement for in-network stream processing" ?

3. What is the goal of operator placement in distributed computing?

4. What is the way to use the greedy algorithm?

Accepted Answer

This paper is concerned with reducing communication costs when executing distributed user tasks in a sensor network.. In this paper, the authors take a fresh look at what is generally considered a simple but poor performance approach for service placement, namely the greedy algorithm.. The authors prove that a modified greedy algorithm is guaranteed to have cost at most 8 times the optimum placement.. Simulations suggest that in practice the greedy algorithm finds a low cost placement.. Furthermore, the cost of repairing a greedy placement decreases rapidly as a function of the proximity of the services to be aggregated.

Accepted Answer

This motivates further exploration into the possibility of a new paradigm in which routing structures are defined in the network, in real time, and can evolve, adapt, and change as the task at hand changes.

Accepted Answer

In the distributed computing context, the goal of operator placement is to minimize latency due to computation and communication.

Accepted Answer

When the network is dynamic, the greedy algorithm is a fast, simple, distributed, and efficient alternative with both provable guarantees and easy adaptivity.

Accepted Answer

The communications have energy, bandwidth and reliability constraints, which introduce a cost for using communication along an edge per unit of data sent.

Accepted Answer

A change of weights in the communication graph or a change in the topology or data rates of the service graph could potentially trigger an entire recalculation for the function C. (3)

Accepted Answer

One of the key challenges of service placement in sensor networks is the ability to adapt to changes in the communication and service graphs without expending too much energy.

Accepted Answer

Once B receives the median value message from D, it can anticipate the time to wait for other nodes since weights are symmetrical.

Accepted Answer

The problem of placing modules within a network has recently gained increasing interest in sensor networks due to the trend of interacting with an entire sensor network as one entity through, for example, database queries [5, 16], macroprogramming [13, 2], or service composition [20, 9].

Accepted Answer

The operator placement problem, also called the module assignment or task embedding problem, for distributed tasks with precedence constraints, is one of the classical distributed computing problems [4, 14, 19, 3].

Accepted Answer

The significant reduction in the size of the flooding zone union as the proximity of the children nodes increases, emphasizes that the performance of the distributed greedy algorithm is dependent on the proximity of the children nodes.

Accepted Answer

To collect performance statistics, the authors run each algorithm on 32 randomly generated networks in each topology category and compute the ratio of the total cost over the optimal cost.

Accepted Answer

Since these operators can take an arbitrary number of operands and return only one number, they can significantly reduce network traffic when placed strategically along the paths that data are routed to the end user.

Accepted Answer

Since these aggregators take an arbitrary number of inputs and produce only one output of the same data type, their data reduction rate equals the inverse of the number of their children in the TAG tree.

Accepted Answer

Although not used in their simulations, the unweighted median cost could also serve as a lower bound on the amount of communication that must be expended in finding this median (assuming the only way to determine distances is through direct communication between nodes).

Accepted Answer

The ratio rp = dop dip denotes the data reduction rate at service p, defined as the data out of p divided by the data fromall children into p.

Greedy is Good: On Service Tree Placement for In-Network Stream Processing

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the contributions mentioned in the paper "Greedy is good: on service tree placement for in-network stream processing" ?

2. What have the authors stated for future works in "Greedy is good: on service tree placement for in-network stream processing" ?

3. What is the goal of operator placement in distributed computing?

4. What is the way to use the greedy algorithm?

5. What are the costs of using communication along an edge?

6. What could trigger an entire recalculation for the function C?

7. What is the key challenge of service placement in sensor networks?

8. How does B anticipate the time to wait for other nodes?

9. What is the purpose of the problem of placing modules within a network?

10. What is the definition of the problem of module assignment?

11. What is the effect of the inverse on the performance of the distributed greedy algorithm?

12. How many networks are used to compute the ratio of the total cost over the optimal cost?

13. What can be done to reduce network traffic?

14. What is the inverse of the number of their children in the TAG tree?

15. What is the cost of the unweighted median?

16. What is the ratio rp = dop dip?

Figures

Citations

Efficient Operator Placement for Distributed Data Stream Processing Applications

Distributed Operator Placement and Data Caching in Large-Scale Sensor Networks

Greedy is Good: On Service Tree Placement for In-Network Stream Processing

Energy-Efficient Task Mapping for Data-Driven Sensor Network Macroprogramming

Energy-Efficient Task Mapping for Data-Driven Sensor Network Macroprogramming

References

Multiprocessor scheduling with communication delays

Adaptive and decentralized operator placement for in-network query processing

IrisNet: an internet-scale architecture for multimedia sensors

Distributed algorithms for finding centers and medians in networks

Approximating Min-sum Set Cover

Related Papers (5)

Distributed Operator Placement and Data Caching in Large-Scale Sensor Networks

Network-Aware Operator Placement for Stream-Processing Systems

Energy-Efficient Task Mapping for Data-Driven Sensor Network Macroprogramming

Computers and Intractability: A Guide to the Theory of NP-Completeness

Load balancing and stability issues in algorithms for service composition