Minimizing user-perceived latency is crucial for Content Distribution Networks (CDNs) hosting interactive services. Latency may increase for many reasons, such as interdomain routing changes and the CDN's own load-balancing policies. CDNs need greater visibility into the causes of latency increases, so they can adapt by directing traffic to different servers or paths. In this paper, we propose a tool for CDNs to diagnose large latency increases, based on passive measurements of performance, traffic, and routing. Separating the many causes from the effects is challenging. We propose a decision tree for classifying latency changes, and determine how to distinguish traffic shifts from increases in latency for existing servers, routers, and paths. Another challenge is that network operators group related clients to reduce measurement and control overhead, but the clients in a region may use multiple servers and paths during a measurement interval. We propose metrics that quantify the latency contributions across sets of servers and routers. Based on the design, we implement the LatLong tool for diagnosing large latency increases for CDN. We use LatLong to analyze a month of data from Google's CDN, and find that nearly 1% of the daily latency changes increase delay by more than 100 msec. Note that the latency increase of 100 msec is significant, since these are daily averages over groups of clients, and we only focus on latency-sensitive traffic for our study. More than 40% of these increases coincide with interdomain routing changes, and more than one-third involve a shift in traffic to different servers. This is the first work to diagnose latency problems in a large, operational CDN from purely passive measurements. Through case studies of individual events, we identify research challenges for managing wide-area latency for CDNs.

Replicating content across a geographically distributed set of servers and redirecting clients to the closest server in terms of latency has emerged as a common paradigm for improving client performance. In this paper, we analyze latencies measured from servers in Google's content distribution network (CDN) to clients all across the Internet to study the effectiveness of latency-based server selection. Our main result is that redirecting every client to the server with least latency does not suffice to optimize client latencies. First, even though most clients are served by a geographically nearby CDN node, a sizeable fraction of experience latencies several tens of milliseconds higher than other in the same region. Second, we find that queueing delays often override the benefits of a client interacting with a nearby server.

We improve the efficiency of static timing analysis when false paths are considered. The efficiency of timing analysis is critical for the performance driven optimization program because timing analysis is invoked heavily in the inner loop. However, when false paths are dealt with in timing analysis, a large number of tags needs to be created and propagated, thus deteriorating efficiency. In this paper, weminimize the number of the tags through a biclique-covering approach, which iteratively removes a tag if the false path information in the tag is covered by the union of other tags.With the produced tags, we remove the false path timing and guarantee to cover the nonfalse path timing. Since the minimum biclique covering of the general bipartite graph is NP complete [Indag. Math., vol. 39, p. 211, 1977], [Discrete Math., vol. 149, no. 1–3, p. 159, 1996], we use a minimal degree ordering approach to perform the biclique-covering minimization. The experimental results show significant reduction on the number of tags.

We improve the efficiency of static timing analysis when false paths are considered. The efficiency of timing analysis is critical for the performance driven optimization program because timing analysis is invoked heavily in the inner loop. However, when false paths are dealt in timing analysis, a large number of tags need to be created and propagated, and thus deteriorated the efficiency. In this paper, we minimize the number of the tags through a biclique covering approach, which iteratively removes a tag if the false path information in the tag is covered by the union of other tags. The produced tags remove the false path timing and guarantee to cover the true path timings. Since the minimum biclique covering of the general bipartite graph is NP complete, we use a minimal degree ordering approach to perform the biclique covering minimization. The experimental results show significant reduction on the number of tags.

This paper proposes a new paradigm for the construction of reversible two point transforms or planar rotations. We show that the transform coefficients for integer to integer mappings through an integer rotation matrix are redundant in modular arithmetic. This redundancy can be exploited by quantizing transform coefficients in a critical manner, producing reversible and unit determinant transforms. For a subset of such critically quantized transforms, the quantization process can be performed by rounded integer division along each dimension. Such transforms are formed by a subset of Pythagorean triads, and can be used to implement reversible image rotations from a countable set of rotation angles. The subjective quality of the rotation so obtained compares favorably with the three shear or lifting algorithm.

This paper describes the construction of computationally efficient transforms which can significantly reduce the complexity of a video decoder without loss in compression efficiency. In particular, these inverse transforms can be implemented using purely 16-bit arithmetic with similar rate-distortion performance to a 32-bit or floating point transform. To allow for variable block size coding, four 2D transforms are considered: one each for 8&8, 8&4, 4&8 and 4&4 blocks respectively. The design criteria for a pure 16 bit arithmetic implementation of the inverse transform are introduced, and used to derive the only useful set of transforms under these conditions. These transforms have been used in the WMV9/VC-1 codec to achieve significant reduction in computational complexity.

Many applications can benefit from knowledge of latency between hosts. Recently, there have been proposals such as global network positioning (GNP) (T. S. E. Ng et al., 2002) and Internet coordinate system (ICS) (H. Lim et al., 2003) that attempt to provide this for hosts on the Internet. The schemes work by using a small set of nodes called "landmarks" to compute the coordinates of other hosts. We investigate the effect of placement of landmarks on the performance of these proposals. We show that with random placement, the estimation error of these schemes varies widely. We propose heuristics for placing landmarks that show good performance. We also investigate the behaviour of these schemes with measurement errors and show that GNP is less affected by measurement errors than ICS.

DNS is a critical component of the operation of Internet applications. However, DNS performance in the wide-area is not well understood. A number of studies present DNS performance measurements [1], [2], [3], [4], but the measurements are out of date, are not collected at client locations (e.g., they are taken at root servers), or are collected at very few client locations.In this paper we present the largest known study of wide-area DNS performance at clients. We use data obtained under a variety of network environments such as location in the Internet topology, connection technology and client ISP. We identify DNS system performance measures and investigate the degree to which they vary from site to site. We report on measures that are relatively consistent throughout the system such as the fraction of names whose lookups succeed, and those that vary widely, such as overall response times and response times from root and gTLD servers. We also discuss the impact of some of these measures on DNS performance for non-cached domain names, confirming some current notions about DNS operation while challenging others.

Summary form only given. The increased information content of hyperspectral imagery over multispectral data has attracted significant interest from the defense and remote sensing communities. We develop a mechanism for compressing hyperspectral imagery with no loss of information. The challenge of hyperspectral image compression lies in the non-isotropy and non-stationarity that is displayed across the spectral channels. Short-range dependence is exhibited over the spatial axes due to the finite extent of objects/texture on the imaged area, while long-range dependence is shown by the spectral axis due to the spectral response of the imaged pixel and transmission medium. A secondary, though critical, challenge is one of speed. In order to be of practical interest, a good solution must be able to scale up to speeds of the order of 20 MByte/s. We use an integerizable eigendecomposition along the spectral channel to optimally extract spectral redundancies. Subsequently, we apply wavelet-based encoding to transmit the residuals of eigendecomposition. We use contextual arithmetic encoding implemented with several innovations that guarantee speed and performance. Our implementation attains operating speeds of 550 kBytes of raw imagery per second, and achieves a compression ratio of around 2.7:1 on typical AVIRIS data. This demonstrates the utility and applicability of our algorithm towards realizing a deployable hyperspectral image compression system.

In this paper, we address 3D image stabilization using a framework for the estimation of scene structure from a monocular motion field. We show that our algorithm rapidly and accurately determines the focus of expansion (FOE) in an optical flow field. This involves computing the least squares error of a large system of equations without actually solving the equations, to generate an error surface that describes the goodness of fit as a function of the hypothesized FOE. Consequently, we recover the rotational motion which we use to perform 3D image stabilization.

Researchers in image processing and computer vision fields have agonized over the last twenty-five years, to come up with robust methods for the structure from motion (SFM) problem. Two dominant approaches, based on flow and feature correspondences have been pursued. Despite tremendous efforts, only limited success in special cases can be claimed. We feel that with the availability of newer, general, robust methods for computing optical flow, fast methods for estimating the focus of expansion (FOE), inexpensive cameras, and inertial information, robust, real-time solutions to this problem may be possible in the near future. Some of our recent work in these areas is presented.

Image stabilization and image mosaicking are fundamental image sequence operations that compensate for camera motion. Stabilization is a precursor to background deletion, region of interest extraction, motion discontinuity estimation and video compression. The effectiveness of an end-to-end algorithm is closely tied to the stabilization accuracy of the input sequence, which in turn depends upon (i) the accuracy of the estimated motion in the scene, (ii) the choice and compliance of the global motion model and (iii) the camera calibration. We deal with (i) and demonstrate the merits of a robust technique for computing optical flow, using overlapped basis functions proposed by Srinivasan and Chellappa (see CAR-TR-845, Univ. of Maryland, 1996). In this technique, we regularize the ill-conditioned gradient constraint equation by modeling optical flow as a linear combination of an overlapped set of basis functions. The solution, which can be shown to be optimal, is obtained by a numerically stable sparse matrix inversion, giving a reliable flow field estimate over a large fraction of the frame. We employ an iterated least squares method for consolidating the local model parameters into a global model, for which we choose a six parameter affine transformation. We argue that our flow field model offers a higher accuracy and robustness than conventional optical flow techniques, and is well suited for image stabilization and mosaicking.

Classical wavelet filter banks are optimized for perfect or near-perfect reconstruction of a signal when the noise-free output of the analysis filter bank is recombined by the synthesis filter bank. Quantization gives rise to noise in the transform domain, affecting the reconstruction. This paper deals with the analysis of transform domain noise in subband systems, and the design of synthesis filters that minimize the output mean square error.