International Conference on Measurement and Modeling of
Computer Systems

Part of Federated Computing Research Conference (FCRC) 2011

San Jose, California
JUNE 7-11, 2011


Detailed Schedule

Tuesday, June 7th - Day 1
8:15am - 9:35am
  Tutorial 1 (1.5h): "Building Accurate Workload Models using Markovian Arrival Processes", Giuliano Casale (Imperial C. London)
Handouts [pdf]
10:00am - 11:20am
  Tutorial 2 (1.5h): "Non-Asymptotic Analysis of Mobile Ad-Hoc Networks", Florin Ciucu (TU Berlin, Deutsche Telekom Lab)
1:30pm - 5:00pm
  Tutorial 3 (3h): "Online Ad Serving: Theory and Practice", Aranyak Mehta, Vahab S. Mirrokni (Google)
Handouts [pdf]
Wednesday, June 8th - Day 2
8:15am - 11:20am
  Tutorial 4 (3h): "Cloud Data Center Networks", Sudipta Sengupta (Microsoft Research)
2:00pm - 3:30pm
  Tutorial 5 (1.5h): "Cloud Computing: Recent trends, challenges and open problems", Andres Lagar-Cavilla, Kaustubh Joshi, (AT&T)
Handouts [ppt]
4:00pm - 5:30pm
  Tutorial 6 (1.5h): "Storage systems in a virtualized world", Ajay Gulati, Irfan Ahmad, (VMware)

Special Discounts for ACM Sigmetrics Tutorial Program available

Thanks to recent donations and sponsorships, ACM Sigmetrics is able to offer additional student support. The additional funds will allow a significant number of students to register for the Sigmetrics Tutorials program at a 60% discounted rate, reducing the registration fee for one tutorial day from $150 to only $60.

The Sigmetrics tutorial program spans two exciting days with tutorials covering a broad range of timely topics ranging from online ad serving, cloud computing challenges, datacenter networks and storage systems to stochastic network calculus and workload modeling. Tutorials are presented by a number of excellent speakers from industry as well as academia. Most tutorials are self-contained, assuming little to no prior background and providing an excellent opportunity to learn about a new area and to identify new open resarch problems. For details on the tutorial program and the speakers please look at Tutorial Details below.

If you are interested in this opportunity, please send a brief application e-mail with your bio and one paragraph explaining your motivation for attending the tutorial(s) to the Tutorial and Workshops chair (bianca at cs dot toronto dot edu). Also, if you know students who might be interested please forward this announcement to them. Note that this opportunity is also open to students who have already registered for the conference or the tutorials. The deadline for applications is this Thursday, June 2, although we will accept late applications if funds are still available.

Tutorial Details

Tutorial 1: Building Accurate Workload Models using Markovian Arrival Processes,
Giuliano Casale (Imperial C. London)
Handouts [pdf]

Performance evaluation of computer systems and networks is often based on modeling techniques such as Markov chains and queueing systems. However, the problem of parameterizing such models in an accurate way from empirical data is still open, despite its fundamental role for obtaining good predictions. Furthermore, workload fitting techniques are important for creating realistic synthetic traces to be used in simulation studies. Features such as heavy-tails or long-range dependence are examples of complications that arise when fitting a workload model.

The tutorial will introduce and reviews recent results on Markovian workload models, a class of tractable workload descriptions which enable the fitting of measured time series (e.g., packet inter-arrival times, job processing times, ... ) into a form that is compositional with Markov chains and queueing systems. Emphasis will be put on understanding the definition, limits of applicability, and practical fitting of such workload models. The tutorial will mostly focus on recent results for phase-type distributions and for the Markovian Arrival Process (MAP), that is one of the most general Markov-modulated process for time-series modeling.

The tutorial is self-contained and does not assume strong theoretical background.

Speaker Biography:
Giuliano Casale received the M.Eng. and Ph.D. degrees in computer engineering from Politecnico di Milano, Italy, in 2002 and 2006 respectively. He joined in 2010 the Department of Computing at Imperial College London where he holds an Imperial College Junior Research Fellowship. His research interests include performance modeling, workload characterization, and resource management. He is co-author of the Java Modelling Tools performance evaluation suite ( Prior to joining Imperial College, he was a full-time researcher at SAP Research UK in 2009 and a postdoctoral research associate at the College of William and Mary, Virginia, in 2007 and 2008. In Fall 2004 he was a visiting scholar at UCLA. He serves as program co-chair for QEST 2012 and for ACM SIGMETRICS/Performance 2012. He is a member of the ACM, the IEEE, and the IEEE Computer Society.

Tutorial 2: Non-Asymptotic Analysis of Mobile Ad-Hoc Networks
Florin Ciucu (TU Berlin, Deutsche Telekom Lab)

The class of Gupta-Kumar results, which predict the throughput capacity in wireless networks, is restricted to asymptotic results. This tutorial presents a methodology to address a corresponding non-asymptotic analysis based on the framework of the stochastic network calculus, in a rigorous mathematical manner. In particular, we derive explicit closed-form results on the distribution of the end-to-end capacity and delay, for a fixed source-destination pair, in a network with broad assumptions on its topology and degree of spatial correlations. The results are non-asymptotic in that they hold for finite time scales and network sizes, as well as for bursty arrivals.

The generality of the results enables the investigation of several interesting problems, such as 1) the effects of time scales or randomness in topology on the network capacity, or 2) the time scales at which nodes should choose between single-hop vs. multi-hop transmissions.

Speaker Biography:
Florin Ciucu was educated at the Faculty of Mathematics, University of Bucharest (B.Sc. in Informatics, 1998), George Mason University (M.Sc. in Computer Science, 2001), and University of Virginia (Ph.D. in Computer Science, 2007). Between 2007 and 2008 he was a Postdoctoral Fellow in the Electrical and Computer Engineering Department at the University of Toronto. Currently he is a Senior Research Scientist at Deutsche Telekom Laboratories (T-Labs) and TU Berlin. His research interests are in the stochastic analysis of communication networks, resource allocation, and randomized algorithms. Florin is a recipient of the ACM Sigmetrics 2005 Best Student Paper Award.

Tutorial 3: Online Ad Serving: Theory and Practice

Aranyak Mehta, Vahab S. Mirrokni (Google)
Handouts [pdf]

As an important part of any ad system, online ad serving is a rich source of interesting algorithmic, learning and economic problems. In this tutorial, we survey recent algorithmic and economic results known in the context of online ad serving in various ad systems, including sponsored search and contextual ads, graphical (or display) ads, and other hybrid and multi-layer ad exchange markets. In particular, we discuss various objective functions in designing such systems, and present (online) stochastic optimization techniques aiming to solve such multi-objective problems, and finally report experimental results, evaluating the performance of these techniques on real-world data sets. Throughout the tutorial, we will touch on online matching problems, primal-dual technique, optimal stochastic control, and power-of-multiple-choices in online ad serving.

Speaker Biographies:
Aranyak Mehta is a Research Scientist at Google Research in Mountain View, CA. He received a B.Tech from IIT Bombay, a Ph.D. from Georgia Tech, and held a visiting scientist position at IBM Research Almaden, before joining Google. His research interests lie in algorithmic game theory, Internet economics and combinatorial algorithms. At Google, he works on algorithmic and auction theoretic problems related to online advertising.

Vahab Mirrokni is a Research Scientist at Google Research in New York. He received his B.Sc. from Sharif University in 2001 and his Ph.D from MIT in 2005. He joined Google after two years at Microsoft Research, Redmond and one year at and MIT. Vahab's main research interests are algorithmic game theory, combinatorial optimization, and the Internet Economics. He is the co-winner of the SODA 2005 and EC 2008 best student paper and best paper awards. At Google, he works on algorithmic and economic problems in search and online advertisement.

Tutorial 4: Cloud Data Center Networks: Technologies, Trends, and Challenges
Sudipta Sengupta (Microsoft Research), (sudipta at microsoft dot com)

Why the Topic is Timely: Large scale data centers are enabling the new era of Internet cloud computing. The computing platform in such data centers consists of low-cost commodity servers that, in large numbers and with software support, match the performance and reliability of expensive enterprise-class servers of yesterday, at a fraction of the cost. The network interconnect within the data center, however, has not seen the same scale of commoditization or dropping price points. Today's data centers use expensive enterprise-class networking equipment and associated best-practices that were not designed for the requirements of Internet-scale data center services -- they severely limit server-to-server network capacity, create fragmented pools of servers that do not allow any service to run on any server, and have poor reliability and utilization. The commoditization and redesign of data center networks to meet cloud computing requirements is the next frontier of innovation in the data center.

Innovations in Data Center Networking: Recent research in data center networks addresses many of these aspects involving both scale and commoditization. By creating large flat Layer 2 networks, data centers can provide the view of a flat unfragmented pool of servers to hosted services. By using traffic engineering methods (based on both oblivious and adaptive routing techniques) on specialized network topologies, the data center network can handle arbitrary and rapidly changing communication patterns between servers. By making data centers modular for incremental growth, the up-front investment in infrastructure can be reduced, thus increasing their economic feasibility. This is an exciting time to work in the data center networking area, as the industry is on the cusp of big changes, driven by the need to run Internet-scale services, enabled by the availability of low-cost commodity switches/routers, and fostered by creative and novel architectural innovations.

What the Tutorial will cover: We will begin with an introduction to data centers for Internet/cloud services. We will survey several next-generation data center network designs that meet the criteria of allowing any service to run on any server in a flat un-fragmented pool of servers and providing bandwidth guarantees for arbitrary communication patterns among servers (limited only by server line card rates). These span efforts from academia and industry research labs, including VL2, Portland, SEATTLE, Hedera, and BCube, and ongoing standardization activities like IEEE Data Center Ethernet (DCE) and IEEE TRILL. We will also cover other emerging aspects of data center networking like energy proportionality for greener data center networks.

Speaker Biography:
Dr. Sudipta Sengupta is currently at Microsoft Research, where he is working on data center systems and networking, peer-to-peer applications, wireless access, non-volatile memory for cloud/server applications, and data deduplication. Previously, he spent five years at Bell Laboratories, Lucent Technologies, where he advanced the state-of-the-art in Internet routing, optical switching, network security, wireless networks, and network coding.

Dr. Sengupta has taught advanced courses/tutorials on networking at many academic/research and industry conferences (please see list below). He received a Ph.D. and an M.S. in Electrical Engg. & Computer Science from Massachusetts Institute of Technology (MIT), USA, and a B.Tech. in Computer Science & Engg. from Indian Institute of Technology (IIT), Kanpur, India. He was awarded the President of India Gold Medal at IIT-Kanpur for graduating at the top of his class across all disciplines. He has published 65+ research papers in some of the top conferences, journals, and technical magazines. He has authored 40+ patents (granted or pending) in the area of computer networking.

Dr. Sengupta won the IEEE Communications Society William R. Bennett Prize for 2011 and the IEEE Communications Society Leonard G. Abraham Prize for 2008 for his work on oblivious routing of Internet traffic. At Bell Labs, he received the President's Teamwork Achievement Award for technology transfer of research into Lucent products. His work on peer-to-peer based distribution of real-time layered video received the IEEE ICME 2009 Best Paper Award. At Microsoft, he received the Gold Star Award which recognizes excellence in leadership and contributions for Microsoft's long term success. Dr. Sengupta is a Senior Member of IEEE.

Tutorial 5: Cloud Computing: Recent trends, challenges and open problems

Andres Lagar-Cavilla, Kaustubh Joshi (AT&T)
Handouts [ppt]

In this tutorial, we will provide a bird's eye view of burgeoning research opportunities in the energetic field of cloud computing. We will explore the challenges users of clouds face, in terms of leveraging the scalability potential of on-demand infrastructure while ensuring reliability, availability, data consistency, durability, and a modicum of quality of service for their applications. We will also explore the challenges facing cloud providers in terms of exploiting the opportunities of statistical multiplexing in shared infrastructure to optimize cost, as well as tackling security and fault-tolerance. Our goal is for the audience to leave the room with a panorama of the contribution space for researchers in cloud computing.

Speaker Biographies:
Andres Lagar-Cavilla is a software systems researcher who does experimental work on virtualization, operating systems, security, cluster computing, and mobile computing. He has a B.A.Sc. from Argentina, and an M.Sc. and Ph.D. in Computer Science from University of Toronto, and can be found online at

Kaustubh Joshi is a researcher at AT&T Shannon Labs who does works on the boundary of analytics and systems, and is interested in the provisioning, diagnosis, dependability, and performance of large Internet-scale systems. He currently works on virtualization, cloud infrastructures, and IP telecommunications systems. Kaustubh holds a B.Eng. in Computer Engineering from Pune University, India, and a M.S. and Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign.

Tutorial 6: Storage systems in a virtualized world

Ajay Gulati, Irfan Ahmad (VMware)

Storage management in virtualized environments is considered as one of the biggest cost factors. According to some estimates, majority of the cost and performance problems are related to storage devices. In this tutorial, we will discuss the following key topics:

1. Storage technologies and architectures: Here we will discuss various technologies and protocols used in a virtual storage environment. Some of these include clustered file systems, NFS access, clustered storage architectures. Recent advances in industry standards will be discussed including offloaded block copy and block zeroing acceleration, SCSI atomic test-and-set for cluster locking operations.

2. Anatomy of an IO: We will discuss path taken by IO re- quests from a virtual machine guest operating system through virtual device emulation, para-virtualized devices and interrupt coalescing for virtual scsi devices and host hypervisor stacks. We will show tradeoffs for passthrough of devices directly into the guest for performance reasons.

3. Performance monitoring and trouble-shooting: Here we will discuss various workload characterization techniques such as vscsiStats, tracing tools, online histograms, and performance troubleshooting methods using various IO stats available via esxtop, vscsiStats etc.

4. Resource Management: We will discuss various IO resource management solutions to provide better performance isolation among virtual disks and load balancing across storage devices (including PARDA, mClock and dmClock). We will also describe very recent work on online modeling storage devices in the wild and its application to diverse goals like admission control, capacity planning, load balancing and congestion control.

5. Future challenges and research directions: We will discuss the upcoming technological trends: SSD devices in particular and some of the future research directions in terms of cloud scale storage, multi-tiered storage, etc.

The tutorial is intended for storage researchers, students, practitioners, administrators and enthusiasts. The level ranges from beginner to expert depending on topic. We will present high level architectural overviews for most topics but also go into details for some.

Speaker Biographies:
Ajay Gulati is a senior researcher at VMware and a member of distributed resource management team. Prior to joining VMware, Ajay got his Phd from Rice University in 2007, where his dissertation was on storage performance virtualization and providing QoS in shared storage systems. He has published and presented his research at many conferences such as Sigmetrics, Usenix FAST, OSDI, PODC and SPAA. He has also given talks on various storage related topics at VMworld, which is an industry conference on virtualization. At VMware, his work has lead to new storage management features such as Storage I/O control and Storage DRS.

Irfan Ahmad is a Staff Engineer at VMware in the ker- nel and distributed resource management team. Most recently, he has been working on automatic IO load balancing on VMware's Storage DRS project. Prior to that, he led the development team for the Storage I/O Control feature and developed VMware's virtual HBA interrupt coalescing algorithm. His research interests include distributed IO scheduling, working set estimation, performance modeling and decentralized algorithms. Irfan has published in the area of storage workload characterization and modeling, interrupt coalescing, deduplication, I/O scheduling and load balancing at various conferences including FAST, USENIX ATC, IISWC. He makes regular appearances at VMworld as well as invited talks at university campuses. Irfan has been at VMware for 8 years prior to which he worked at a small microprocessor company called Transmeta on their code morphing processor.