Defeating Online Gaming’s Biggest Enemy: Latency

Scott Nelson

ColumbusStateUniversity

ABSTRACT

In this paper we look at several ways to combat online games biggest enemy; latency. We will explore a concept of Latency EQalization (LEQ) using programmable routers. We will then look other technologies using programmable routers and the success they are having. Finally, we will look at combining a programmable network and the use of network accelerators to further reduce latency times.

1. INTRODUCTION

Online gaming has become one of the fastest growing sectors on the Internet. With that growth comes increasing demands for network resources. This increased pressure leads to increased latency: the length of time it takes for a packet of data to get from one designated point to another.[1] Latency can be caused by many factors such as the time needed to send and receive packets, time for propagation from one link to another, and time waiting in a router queue due to congestion.[2] While the effect of latency on the player can be minimal for a omnipresent game, such as World of Warcraft; it can be significant on a first-person shooter game such as FEAR. The delay in action/response can result in poorly perceived game play due to the delayed responses.

In the paper Latency Equalization: A Programmable Routing Service Primitive [3], the authors propose using programmable network nodes to create Latency EQualization(LEQ). The methodology is to place programmable nodes (routers)in a network, then have the nodes plot pathways through the network and send packets of data over a predetermined pathway so that each client(player) will have approximately the same latency time as the player with the longest time.

Implementing this architecture would be relatively easy since it can co-exist with current networks as there is no need to change the current underlying routing protocols. The LEQ protocols would overlay the existing ones. Also, only a few programmable nodes are required to achieve LEQ.

According to the authors, there is a benefit to this method. Programmable routing is at least 22% (with one node) and up to 90% (with only five nodes) more effective in reaching LEQ than a conventional shortest path routing algorithm.[3]

There are a few obstacles to implementing this method however. As of today, there have not been any serious studies of LEQ from a network perspective. Moreover, there are no proposed routing architectures that would help implement LEQ.[3]

This paper will look at what may be capable if the programmable router with LEQ is combined with the use of WAN accelerators in regards to decreasing latency.

Section 2 of this paper will examine related work regarding to latency reduction in networks using programmable routers. Section 3 will be a detailed look at the combination of LEQ, and the related works, with WAN network acceleratorsto decrease latency times in networks with programmable routers. The last section will be a conclusion to the paper.

2. RELATED WORK

The concept of programmable routers has started to take hold. Many companies are now experimenting with virtual networks that allow multiple logic networks to operate on the same physical infrastructure.[4] To do so, the idea is to have the data plane, the part of the router that decides what to do with incoming packets, implemented in the hardware rather than the software by multiplexing the hardware resources. According to the paper Building a Fast, Virtualized Data Plane with Programmable Hardware[4], hardware based forwarding promises fast packet forwarding rates, thus reducing latency.

Tests have also begun on a new technology called OpenFlow.[5] OpenFlow would allow uniform interfaces between different devices that could be then be programmed with a “flow table” by combining similar packets into a flow(a unit of control), then sending that flow to the appropriate port.[6] In a 2008 demonstration, OpenFlow engineersran a first person shooter game between Connecticut and Japan where algorithms were used to move server codeto an optimal location, thereby suiting the needs of the clients by keeping network latency to a minimum.[5]

3. PROBLEM/SOLUTION

The problem of latency affects not only online games, but streaming audio/video, VoIP…just about any peer to peer online service. Whether the longer the delay in a game action or the longer the delay in a meeting conversation or the longer it takes for a music file to download, all result in user becoming dissatisfied with the application’s performance.

The use of programmable hardware is a very good start to shortening latency. Imagine a combination of the three previously mentioned technologies. A virtual network that could determineduring a gaming session that most players are on the east coast. The network then moves the source code from the west coast to a server in the Midwest to reduce latency for the majority of players. Then the network performs LEQ so all gamers have the same approximate latency time.

In this paper, I now add wide area network (WAN) accelerators to this network. A WAN accelerator would do four different things. 1) The accelerator would send an ack back to the sending host as soon as the packet is received ensuring the sender keeps entering packets into the stream. 2) The accelerator would convert any TCP streams into UDP streams to cross the network, then change it back on the receiving end. Since most live streaming events already use UDP, this would not be a significant change. 3) The accelerator would recognize data patterns and then cache the repeating data for faster local access. 4) The accelerator would be able to compress certain types of data requiring fewer packets having to be sent.[7]

If a gaming company or service were building a new backbone network for their software, they could place source servers on each coast and one in the Midwest, being able to take advantage of OpenFlow. The network would be able to move the main source code sever between the three regional servers depending upon the network traffic. An example would be a major online tournament at the ComiCon convention on the west coast. The gaming company could move the source server to the west coast to facilitate faster responses for the tournament. A month later, the company could move the source server to the east cost for a tournament at the DragonCon convention. The company would also have the ability to move the source server across the country in response to hourly/daily gaming trends.

Then the gaming company wouldlayout a network using programmable routers, being able to implement a virtual network and/or take advantage of LEQ. The layout would place edge routers in key geographical locations, such as southeast, northeast, etc… The network would allow for the server to rout certain clients along determined pathsso as to attempt to make the latency delays as equitable as possible for all clients.

Finally the company would install WAN accelerators at each source server and each edge router. This would have the potential to reduce latency to a minimum for all clients.

This paper did not address “last mile” latency, the delay caused by home network devices such as dial-up/cable modems and wireless home networking devices as these can vary greatly depending upon the type of hardware and connections/links used.

4. CONCLUSION

The goal of the paper is to try to present a combination of new, LEQ and OpenFlow, and existing, programmable routers and WAN accelerators,technologies to help reduce latency in live online sessions in an effort to achieve a better user experience. There have been innovative approaches in the past few years to reducing latency. While most live streaming networks are already using at least one of these technologies, if all of them could be combined into one network the possibilities could be incredible.

5. REFERENCES

[1]

[2] M.Claypool et. al. Latency and Player Actions in Online Games. Communications of the ACM, November 2006/Vol.49, No. 11.

[3] M. Yu et. al. Latency Equalization: A Programmable Routing Service Primitive. PRESTO ’08, August 22, 2008.

[4] Muhammad Bilal Anwer et.al. Building a Fast, Virtualized Data Plane with Programmable Hardware. ACM SIGCOMM Computer Communication Review, Volume 40, Number 1, January 2110.

[5] New Technology Makes Internet Switches, Routers Open and Programmable. Business Wire, 2008-10-28.

[6] Adam Greenhalgh et. al. Flow Processing and the Rise of Commodity Network Hardware. ACM SIGCOMM Computer Communication Review, Volume 49, Number 2, April, 2009.

[7]