Internet Scale Overlay Hosting

[under construction]

Network overlays have become a popular tool for implementing Internet applications. While content-delivery networks provide the most prominent example of the commercial application of overlays, systems researchers have developed a variety of experimental overlay applications, demonstrating that the overlay approach can be an effective method for deploying a broad range of innovative systems. Rising traffic volumes in overlay networks make the performance of overlay platforms an issue of growing importance. Currently, overlay platforms are constructed using general purpose servers, often organized into a cluster with a load-balancing switch acting as a front end. This project explores more integrated and scalable architectures suitable for supporting large-scale applications with thousands to many millions of end users. In addition, we are studying various network level issues relating to the control and management of large-scale overlay hosting services.

Overvew

An Overlay Hosting Service (OHS) is a shared infrastructure that supports multiple overlay networks. There are two major physical components to an OHS, the Overlay Hosting Platforms (OHP) and the links joining OHPs to one another. These links are expected to have provisioned bandwidth, allowing the OHS to deliver provisioned capacity to the overlay networks that operate within it. Access to the OHS infrastructure uses the Internet for the final hop. This traffic can be carried over UDP tunnels, or virtual links, depending on the deployment context (e.g. MPLS or VLAN).

Overlay Hosting Service

The OHPs contain flexible processing resources that can be allocated to different overlays. The resources allocated to an overlay node can range from a small fraction of a general-purpose server, up to hundreds of high performance multi-core processor subsystems. To enable a wide range of overlay network architectures and service models, it's important for OHSs to be as flexible as possible. Overlay network providers should be free to define their own protocols, packet formats and service models, within the framework provided by the OHS.

The use of provisioned links between OHPs allows OHS providers to deliver provisioned overlay links to the overlay providers, making it possible for overlay networks to provide consistent performance to their users. For access over UDP tunnels, true QoS may be difficult to achieve, but even here there is the potential for significantly better performance than can be achieved over end-to-end connections in the public Internet.

To clarify the role of various elements of an OHS, it's helpful to introduce some terminology. The infrastructure and core services provided by the OHS are referred to as the substrate. We also use substrate to refer to the core services provided by the OHPs. We use overlay to refer to each overlay network hosted by an OHS, and we use the term overlay node to refer to the individual nodes within the overlay network. For brevity, we sometimes use platform in place of overlay hosting platform, and node in place of overlay node.

Scalable Overlay Hosting Platforms

The purpose of an Overlay Hosting Platform (OHP) is to provide resources for use by individual overlay. We seek to provide the greatest possible flexibility to overlays, while maintaining appropriate separation among different overlays. This of course includes separation of their use of memory and mass storage, but also their use of network bandwidth and processing capacity.

We start by articulating some high level objectives for an effective OHP design.

• Scalable performance. If overlay networks are to deliver Internet scale with router-like performance, OHPs must be capable of scaling up to terabit IO capacities and the ability to execute hundreds of basic operations per byte of data forwarded. This requires the use of scalable interconnection networks and flexible processing subsystems capable of very high throughputs. At the same time, we expect that a significant fraction of overlay services may be quite small, so the system should be able to support a mixture of large and small overlays, with the largest being several of orders of magnitude larger than the smallest.

• Stability and reliability. An OBP providing Internet-scale service delivery must be highly stable and reliable. In particular, failures in one overlay network should not be able to affect the operation of others on the same platform. Moreover, the core components should exhibit a high level of intrinsic reliability, and the system should be capable of rapidly replacing failed subsystems, by switching in equivalent spares.

• Ease of use. To enable small organizations to take advantage of overlay hosting services, it's important that the time and effort required to create a new overlay be relatively modest.

• Technology diversity and adaptability. Overlay hosting services represent a new class of technology platform, and it is likely that the design of OHPs will evolve over time, as experience develops. It's important that an OHP system architecture be open to the incorporation of new technology components as they become available.

• Flexible allocation of link bandwidth. Link bandwidth is a key resource in any network or distributed system. OHPs should support flexible allocation of bandwidth to different slices, including both reserved and shared bandwidth models.

• Isolation of slices. The OHP must allow different slices to co-exist without interference. Ideally, each slice should have the illusion that it is operating within a dedicated environment. This means that resources like memory and mass storage must be free from modification by other slices and that each slice have the ability to reserve dedicated processing capacity and link bandwidth.

• Minimal constraints on slices. The OHP should place as few constraints as possible on the slices it hosts. In particular, it should not place any constraints on data formats or limit the ways in which slices provide various capabilities, or constrain the way in which they use their assigned processing resources.

Architectural Options

This section discusses two high level system architecture options for scalable overlay hosting platforms and the issues arising from these options.

Virtualized line card architecture

Consideration of a conventional router or switch leads naturally to an architecture in which line cards are replaced by virtualized line cards that include a substrate portion and generic processing resources that can be assigned to different virtual line cards. The substrate supports configuration of the generic processing resources so that different meta line cards can co-exist without interference. On receiving data from the physical link, the substrate first determines which virtual line card it should be sent to and delivers it. Virtual line cards pass data back to the substrate, in order to forward it through the shared switch fabric, on input, or to the outgoing link, on output.

Virtualized Line Card Architecture

One issue with this architecture concerns how to provide generic processing resources at a line card, in a way that allows the resources to be shared by different overlays. Conventional line cards are often implemented using Network Processors (NP), programmable devices that include high performance IO and multiple processor cores to enable high throughput processing. It seems natural to take such a device and divide its internal processing resources among multiple overlays. For example, an NP with 16 processor cores could be used by up to 16 different overlays, by simply assigning processor cores. Unfortunately, current NPs are not designed to be shared. All processing cores have unprotected access to the same physical memory, making it difficult to ensure that different overlays don’t interfere with one another.

Also, each processor core has a fairly small program store. This is not a serious constraint in conventional applications, since processing can be pipelined across the different cores, allowing each to store only the program it needs for its part of the processing. However, a core implementing all the processing for one overlay must store the programs to implement all the processing steps for that overlay. The underlying issue raised by this discussion is that efficient implementation of an architecture based on virtualized line cards, requires components that support fine-grained virtualization and conventional NPs do not.

The virtualized line card approach is also problematic in other respects. Because it associates processing resources with physical links, it lacks the flexibility to support overlays with a wide range of processing needs. Some overlays may require more processing per unit IO bandwidth than NPs provide, and this is difficult to provide with a virtualized line card approach. The virtualized line card approach also does not easily accommodate alternate implementation approaches for overlay nodes (such as configurable logic).

Processing pool architecture

The processing pool architecture separates the processing resources used by overlays from the physical link interfaces. This allows a more flexible allocation of processing resources and reduces the need for fine-grained virtualization. This architecture, illustrated in Figure 3, provides a pool of Processing Engines (PE), that are accessed through the switch fabric. The line cards that terminate the physical links forward packets to PEs through the switch fabric, but do no processing that is specific to a particular overlay. There may be different types of PEs, including some implemented using network processors, others implemented using conventional microprocessors and still others implemented using FPGAs. The NP and FPGA based PEs are most appropriate for high throughput packet processing, the conventional processor for control functions that require more complex software or for overlays with a high ratio of processing to IO. An overlay node may be implemented using a single PE or multiple PEs. In the case of a single PE, data will pass through the physical switch fabric twice, once on input, once on output. In a node that uses multiple PEs to obtain higher performance, packets may have to pass through the switch fabric a third time.

The primary drawback of the processing pool architecture is that it requires multiple passes through the switch fabric, increasing delay and increasing the switch capacity needed to support a given total IO bandwidth. The increase in delay is not a serious concern in wide area network contexts, since switching delays are typically 10 <math>\mu</math>s or less. The increase in capacity does add to system cost, but since a well-designed switch fabric represents a relatively small part of the cost of a conventional router (typically 10-20%), we can double, or even triple the capacity without a proportionally large increase in the overall system cost. Also, since OHPs can be expected to have a higher ratio of processing to IO than conventional network routers, the impact of the higher bandwidth use is significantly reduced.

The great advantage of the processing pool architecture is that it greatly reduces the need for fine-grained virtualization within NP and FPGA-based subsystems, for which such virtualization is difficult. Because the processing pool architecture brings together the traffic for each individual overlay node, there is much less need for PEs to be shared among multiple nodes. The one exception to this is nodes with such limited processing needs that they cannot justify the use of even one complete PE. Such nodes can still be accommodated by implementing them on a general purpose processor, running a conventional operating system that supports a virtual machine environment. Later, we discuss another approach that allows such nodes to share an NP-based PE for fast path forwarding, while relying on a virtual machine running within a general purpose processor to handle exception cases.

Another advantage of the processing pool architecture is that it simplifies sharing of the switch fabric. The switch traffic must maintain traffic isolation among the different overlay nodes. One way to ensure this is to constrain the traffic flows entering the switch fabric so as to eliminate the possibility of internal congestion. This is difficult to do in all cases. In particular, nodes consisting of multiple PEs should be allowed to use their “share” of the switch fabric capacity in a flexible fashion, without having to constrain the pair-wise traffic flows among the PEs. However allowing this flexibility makes it possible for several PEs in a given metarouter to forward traffic to another PE at a rate that exceeds the bandwidth of the interface between the switch fabric and the destination PE.

There is a straightforward solution to this problem in the processing pool architecture. To simplify the discussion, we separate the handling of traffic between line cards and PEs from the traffic among PEs in a common metarouter. In the first case, we can treat the traffic as a set of point-to-point streams that are rate-limited when they enter the fabric. Rate-limiting these flows follows naturally from the fact that they are logical extensions of traffic flows on the external links. Because the external link flows must be rate limited to provide traffic isolation on the external links, the internal flows within the switch fabric can be configured to eliminate the possibility of congestion.

For PE-to-PE traffic, we cannot simply limit the traffic entering the switch, since it’s important to let PEs communicate freely with other PEs in the same node, without constraint. However, because entire PEs are allocated to nodes in the processing pool architecture, it’s possible to obtain good traffic isolation in a straightforward way, for this case as well. In general, we need two properties from the switch fabric. First, it must support constrained routing, so that traffic from one to node cannot be sent to PEs belonging to another node. Second, we need to ensure that congestion within one node does not affect traffic within another node. The emergence of Ethernet as a backplane switching technology provides the first property. Such switches support VLAN-based routing that can be used to separate the traffic from different nodes. The second property is satisfied by any switching fabric that is nonblocking at the port level. While some switch fabrics fail to a fully achieve the objective of nonblocking performance, this is the standard figure of merit for switching fabrics a and most come reasonably close to achieving it.

Abstraction vs. Transparency

There are two very different approaches that one can take to providing a flexible overlay hosting capability. The raw resource approach seeks to make resources available, without imposing any specific usage model on those resources. Such systems can provide a variety of different resources, so long as they can be used safely within a fairly generic system framework. This makes it easy to incorporate new types of processing resources in to the overall system architecture as they become available.

The abstract programming interface approach attempts to shield overlay network developers from the characteristics of the underlying hardware components, by providing a system level abstraction through which users can implement new functionality. This has some obvious appeal, since it can allow developers to work at a higher level, and to readily port their overlay network functionality to take advantage of new hardware subsystems, without a major new development effort.

Both approaches have their merits and limitations. The raw resource approach gives developers complete control over their allocated resources, allowing them to take make most effective use of the underlying resources and maximize the system performance. The abstract interface approach has the potential to greatly reduce development effort, but may make it difficult to achieve performance objectives.

Implementation Options

T

specifics for GENI and SPP

Control of Overlay Hosting Services

General control architecture, including design of a control overlay network.

Internet Scale Overlay Applications

Network games work.

Scalable audio.

Mapping Overlays onto an OHS Infrastructure

Jing's work.

Issues for Multi-domain Overlay Hosting

Control issues and multi-domain resource mapping.

References

[BA06]

Bavier, A., N. Feamster, M. Huang, L. Peterson, J. Rexford. “In VINI Veritas: Realistic and Controlled Network Experimentation,” Proc. of ACM SIGCOMM, 2006.

[BH06]

Bharambe, A., J. Pang, S. Seshan. “Colyseus: A Distributed Archi-tecture for Online Multiplayer Games,” In Proc. Symposium on Networked Systems Design and Implementation (NSDI), 3/06.

[CH02]

Choi, S., J. Dehart, R. Keller, F. Kuhns, J. Lockwood, P. Pappu, J. Parwatikar, W. D. Richard, E. Spitznagel, D. Taylor, J. Turner and K. Wong. “Design of a High Performance Dynamically Extensible Router.” In Proceedings of the DARPA Active Networks Conference and Exposition, 5/02.

[CH03]

Chun, B., D. Culler, T. Roscoe, A. Bavier, L. Peterson, M. Wawr-zoniak, and M. Bowman. “PlanetLab: An Overlay Testbed for Broad-Coverage Services,” ACM Computer Communications Review, vol. 33, no. 3, 7/03.

[CI06]

Cisco Carrier Routing System. At www.cisco.com/en/ US/products/ps5763/, 2006

[DI02]

Dilley, J., B. Maggs, J. Parikh, H. Prokop, R. Sitaraman, and B. Weihl. “Globally Distributed Content Delivery,” IEEE Internet Computing, September/October 2002, pp. 50-58.

[FO07]

Force 10 Networks. “S2410 Data Center Switch,” http:// www.force10networks.com/products/s2410.asp, 2007.

[FR04]

Freedman, M., E. Freudenthal and D. Mazières. “Democratizing Content Publication with Coral,” In Proc. 1st USENIX/ACM Sym-posium on Networked Systems Design and Implementation, 3/04.

[GE06]

Global Environment for Network Innovations. http://www.geni.net/, 2006.

[HI98]

Mike Hicks_ Pankaj Kakkar_ Jonathan T_ Moore_ Carl A_ Gunter_ and Scott Nettles. “PLAN, A packet language for active networks,” In Proceedings of the Third ACM SIGPLAN International Conference on Functional Programming Languages, 1998.

[IXP]

Intel IXP 2xxx Product Line of Network Processors. http://www .intel.com/design/network/products/npfamily/ixp2xxx.htm.

[KA02]

Karlin, Scott and Larry Peterson. “VERA: An Extensible Router Architecture,” In Computer Networks, 2002.

[KO00]

Kohler, Eddie, Robert Morris, Benjie Chen, John Jannotti and M. Frans Kaashoek. “The Click modular router,” ACM Transactions on Computer Systems, 8/2000.

[KO04]

Kontothanassis, L. R. Sitaraman, J. Wein, D. Hong, R. Kleinberg, B. Mancuso, D. Shaw and D. Stodolsky. “A Transport Layer for Live Streaming in a Content Delivery Network,” Proc. of the IEEE, Special Issue on Evolution of Internet Technologies, 9/04.

[PA03]

Pappu, P., J. Parwatikar, J. Turner and K. Wong. “Distributed Queueing in Scalable High Performance Routers.” Proceeding of IEEE Infocom, 4/03.

[PE02]

Peterson, L., T. Anderson, D. Culler and T. Roscoe. “A Blueprint for Introducing Disruptive Technology into the Internet,” Proceed-ings of ACM HotNets-I Workshop, 10/02.

[RA05]

Radisys Corporation. “Promentum™ ATCA-7010 Data Sheet,” product brief, available at http://www. radisys.com/files/ATCA-7010_07-1283-01_0505_datasheet.pdf.

[RH05]

Rhea, S., B. Godfrey, B. Karp, J. Kubiatowicz, S. Ratnasamy, S. Shenker, I. Stoica and H. Yu. “OpenDHT: A Public DHT Service and Its Uses,” Proceedings of ACM SIGCOMM, 9/2005.

[SP01]

Spalink, T., S. Karlin, L. Peterson and Y. Gottlieb. “Building a Robust Software-Based Router Using Network Processors,” In ACM Symposium on Operating System Principles (SOSP), 2001.

[ST01]

Stoica, I., R. Morris, D. Karger, F. Kaashoek and H. Balakrishnan. “Chord: A scalable peer-to-peer lookup service for internet applica-tions.” In Proceedings of ACM SIGCOMM, 2001.

[ST02]

Stoica, I., D. Adkins, S. Zhuang, S. Shenker, S. Surana, “Internet Indirection Infrastructure,” Proc. of ACM SIGCOMM, 8/02.

[TU06]

Turner, J. “A Proposed Architecture for the GENI Backbone Plat-form,” In Proceedings of ACM- IEEE Symposium on Architectures for Networking and Communications Systems (ANCS), 12/2006.

[VS06]

Linux vServer. http://linux-vserver.org