Danny Raz

IDMaps: a global internet host distance estimation service

There is an increasing need to quickly and efficiently learn network distances, in terms of metrics such as latency or bandwidth, between Internet hosts. For example, Internet content providers often place data and server mirrors throughout the Internet to improve access latency for clients, and it is necessary to direct clients to the nearest mirrors based on some distance metric in order to realize the benefit of mirrors. We suggest a scalable Internet-wide architecture, called IDMaps, which measures and disseminates distance information on the global Internet. Higher level services can collect such distance information to build a virtual distance map of the Internet and estimate the distance between any pair of IP addresses. We present our solutions to the measurement server placement and distance map construction problems in IDMaps. We show that IDMaps can indeed provide useful distance estimations to applications such as nearest mirror selection.

The cache location problem

This paper studies the problem of where to place network caches. Emphasis is given to caches that are transparent to the clients since they are easier to manage and they require no cooperation from the clients. Our goal is to minimize the overall flow or the average delay by placing a given number of caches in the network. We formulate these location problems both for general caches and for transparent en-route caches (TERCs), and identify that, in general, they are intractable. We give optimal algorithms for line and ring networks, and present closed form formulae for some special cases. We also present a computationally efficient dynamic programming algorithm for the single server case. This last case is of particular practical interest. It models a network that wishes to minimize the average access delay for a single web server. We experimentally study the effects of our algorithm using real web server data. We observe that a small number of TERCs are sufficient to reduce the network traffic significantly. Furthermore, there is a surprising consistency over time in the relative amount of web traffic from the server along a path, lending a stability to our TERC location solution. Our techniques can be used by network providers to reduce traffic load in their network.

Understanding TCP fairness over wireless LAN

As local area wireless networks based on the IEEE 802.11 standard see increasing public deployment, it is important to ensure that access to the network by different users remains fair. While fairness issues in 802.11 networks have been studied before, this paper is the first to focus on TCP fairness in 802.11 networks in the presence of both mobile senders and receivers. In this paper, we evaluate extensively through analysis, simulation, and experimentation the interaction between the 802.11 MAC protocol and TCP. We identify four different regions of TCP unfairness that depend on the buffer availability at the base station, with some regions exhibiting significant unfairness of over 10 in terms of throughput ratio between upstream and downstream TCP flows. We also propose a simple solution that can be implemented at the base station above the MAC layer that ensures that different TCP flows share the 802.11 bandwidth equitably irrespective of the buffer availability at the base station.

On the placement of Internet instrumentation

The IDMaps project aims to provide a distance map of the Internet from which relative distances between hosts on the Internet can be gauged. Many distributed systems and applications can benefit from such a distance map service, for example, a common method to improve user-perceived performance of the Internet is to place data and server mirrors closer to clients. When a client tries to access a mirrored server, which mirror should it access? With IDMaps, the closest mirror can be determined based on distance estimates between the client and the mirrors. In this paper we investigate both graph theoretic methods and ad hoc heuristics for instrumenting the Internet to obtain distance maps. We evaluate the efficacy of the resulting distance maps by comparing the determinations of the closest replica using known topologies against those obtained using the distance maps.

Constrained mirror placement on the Internet

Internet service providers and infrastructural companies often employ mirrors of popular content to decrease client download time and server load. Due to the immense scale of the Internet and decentralized administration of the networks, companies have a limited number of sites (relative to the size of the Internet) where they can place mirrors. Mirrors of popular content are usually replicated on every site to maximize reachability to clients. We study the performance improvements as the number of mirrors increases under different placement algorithms subject to the constraint that mirrors can be placed only at certain locations. Although there are extensive theoretical studies on center placement and, analytical and empirical studies on Web cache placement, we are not aware of any published literature on mirror placement especially in the case of constrained mirror placement. Our results show that increasing the number of mirror sites under the constraint is effective in reducing client download time and reducing server load only for a surprisingly small range of values regardless of the mirror placement algorithm.

A simple efficient approximation scheme for the restricted shortest path problem

In this short paper we give a very simple fully polynomial approximation scheme for the restricted shortest path problem. The complexity of this @e-approximation scheme is O(|E|n(loglogn+1/@e)), which improves Hassins original result (Math. Oper. Res. 17 (1) (1992) 36) by a factor of n. Furthermore, this complexity bound is valid for any graph, regardless of the cost values. This generalizes Hassins results which apply only to acyclic graphs. Our algorithm is based on Hassins original result with two improvements. First we modify Hassins result and achieve time complexity of O(|E|n(loglog(UB/LB)+1/@e)), where UB and LB are upper and lower bounds for the problem. This modified version can be applied to general graphs with any cost values. Then we combine it with our second contribution, which shows how to find an upper and a lower bound such that UB/LB=

A Stable Network-Aware VM Placement for Cloud Systems

Virtual Machine (VM) placement has to carefully consider the aggregated resource consumption of co-located VMs in order to obey service level agreements at lower possible cost. In this paper, we focus on satisfying the traffic demands of the VMs in addition to CPU and memory requirements. This is a much more complex problem both due to its quadratic nature (being the communication between a pair of VMs) and since it involves many factors beyond the physical host, like the network topologies and the routing scheme. Moreover, traffic patterns may vary over time and predicting the resulting effect on the actual available bandwidth between hosts within the data center is extremely difficult. We address this problem by trying to allocate a placement that not only satisfies the predicted communication demand but is also resilient to demand time-variations. This gives rise to a new optimization problem that we call the Min Cut Ratio-aware VM Placement (MCRVMP). The general MCRVMP problem is NP-Hard, hence, we introduce several heuristics to solve it in reasonable time. We present extensive experimental results, associated with both placement computation and run-time performance under time-varying traffic demands, to show that our heuristics provide good results (compared to the optimal solution) for medium size data centers.

An efficient approximation for the generalized assignment problem

We present a simple family of algorithms for solving the Generalized Assignment Problem (GAP). Our technique is based on a novel combinatorial translation of any algorithm for the knapsack problem into an approximation algorithm for GAP. If the approximation ratio of the knapsack algorithm is α and its running time is O(f(N)), our algorithm guarantees a (1 + α)-approximation ratio, and it runs in O(M ċ f(N) + M ċ N), where N is the number of items and M is the number of bins. Not only does our technique comprise a general interesting framework for the GAP problem; it also matches the best combinatorial approximation for this problem, with a much simpler algorithm and a better running time.

Approximating total flow time on parallel machines

Approximating Total Flow Time on Parallel Machines Stefano Leonardi* Danny Raz t We consider the problem of optimizing the total flow time of a stream of jobs that are released over time in a multiprocessor setting. This problem is NP-hard even when we allow preemption, and have only two machines. Although the total (or average) flow time is widely accepted as a good measurement of the overall quality of service, no approximation algorithms were known for this basic scheduling problem. This paper contains two main results. We first prove that when preemption is allowed, Shortest Remaining Processing Time (SRPT) is an O(log(min{ m, P})) approximation algorithm for the total flow time, w%ere n is the number of jobs, m is the number of machines, and P is the ratio between the maximum and the minimum processing time of a job. We also provide an tl(log ~) and an Cl(log P) lower bounds on the (worst case) competitive ratio of any randomized algorithm for the on-line problem in which jobs are known at their release times. Thus, we show that up to a constant factor SRPT is an optimal on-line algorithm. Our second main result addresses the non-preemptive case. We present a general technique that allows to transform any preemptive solution into a non-preemptive solution at the expense of an 0(R) factor in the approximation ratio of the total flow time. Combining this technique with our previous result yields an 0( ~ log R) approxil_c mation algorithm for this case. We also show an fl(rr 3 ) lower bound on the approximability of this problem (assuming P # NP). ●Dipartimento di Informatica e Sisternistica, University di Roma “La Sapienza”. This work waa partially done while the author was a post-dot at the International Computer Science Institute (ICSI), Berkeley. This work is partly supported by EU ESPRIT Long Term Research Project ALCOM-IT under contract n 20244, and by Italian Ministry of Scientific Research Project 40% “Algoritrni, Modelli di Calcolo e StruttureInformative.” Email: leonflklis.uniromal .it tInternational Computer Science Institute (ICSI), Berkeley. Research supported in part by National Science Foundation operating grants CCR-9304722 and NCR-9416101. Email: draz@icsi.berkeley.edu Penuission 10 m~ke digilallhal-d copies t~t’all or pall {~1’lllisltl:llcriill Ii)r personal LJr classroom (Ist IS gmnlcd 14ilhoul IJc pr(Ij Idml 111:11IIIC copies arc I101 mndc w distrihl!lcd Ibr p’otilo!’Wnmcrt?int :KiiWINJgL>. Ihc LX~p~righlnoliw. Ihe line ol’llm ptlblicolioll :ind ils dnlc :Ippc:iI. ;IIIII Imticc is given Iha( copyrigh[ is by pcrmissioll (Jftllc :\~hl. lI1c ‘1’0copy olllcr,!isd. 10 rcpldllish. [0 post on servers or 10 rcdisl]-ihitlt 10 11s[s. rc[l~lircsspccilic pmmisstotl :md/or I>c .qT(X “ 97 El Paso. “1’..s;,sI lS,\ L@yrigl][ 1997 ;\~\l 0-X079 I -XXX-(I 97 ‘[15 ,.S3.5(1

Efficient reactive monitoring

Networks are monitored in order to ensure that the system operates within desirable parameters. The increasing complexity of networks and services provided by them increases this need for monitoring. Monitoring consists of measuring properties of the network, and of inferring an aggregate predicate from these measurements. Conducting such monitoring introduces traffic overhead that may reduce the overall effective throughput. This paper studies ways to minimize the monitoring communication overhead in IP networks. We develop and analyze several monitoring algorithms that achieve significant reduction in the management overhead while maintaining the functionality. The main idea is to combine global polling with local event driven reporting. The amount of traffic saving depends on the statistical characterization of the monitored data. We indicate the specific statistical factors that affect the saving and show how to choose the right algorithm for the type, of monitored data. In particular, our results show that for Internet traffic our algorithms can save more than 90% of the monitoring traffic.