Possible Solutions

To solve the low lookup performance issue of the KAD network, we discuss three possible schemes in this section and show their advantages and weaknesses respec-tively.

3.5.1 Self-Maintenance Scheme

One possible scheme is to assign the task of maintaining publishing tables back to the local peers (published peers) themselves. The published peers can verify the online status of entries in their publishing tables within a short time period (e.g.

10 minutes). They can also directly synchronize the information of peers from their well-maintained routing tables. With this modication, stale records in the publishing tables can be signicantly reduced, and retrieved information by other peers can be more useful. Other advantages of this self-maintenance scheme may include: it does not change any fundamental structure of the current system; it unies the maintenance pattern of the routing table and the publishing tables; it may also be easily implemented. However, the self-maintenance scheme improves

Chapter 3 Section 3.5

the availability of the publishing tables just based on reducing the number of use-less records. Because of the selsh behavior of peers and poor incentive policy, this scheme cannot fundamentally improve the lookup performance of the KAD network. The total amount of useful information available to users was not in-creased. Additionally, since it needs to renew publishing tables frequently, it may also introduce far more table-maintaining trac into the system.

3.5.2 Chunk-Based Publishing/Retrieving Scheme

We can use a chunk-based publishing/retrieving scheme to replace the current

le-based publishing/retrieving scheme. In aMule/eMule, the chunk-based down-loading and updown-loading scheme has already been employed to speed up the content sharing process. Under the chunk-based scheme, a le is usually separated into several chunks. Each chunk can be downloaded from dierent peers and uploaded to others simultaneously. However, instead of publishing the information of an ob-tained chunk immediately, a peer in aMule/eMule publishes the information of an entire le into the KAD network. That is, under the current publishing/retrieving scheme, the le will not be published into the KAD network until it has been fully obtained.

In a typical P2P le-sharing environment, a peer usually runs both the upload-ing and the downloadupload-ing processes concurrently, and this peer will not leave the system before it completes its download tasks. Thus if the information of peers, who possesses some chunks of the whole le, can also be published, the availabil-ity of the publishing table will be signicantly improved. The obvious advantage of this modication is that it can increase the amount of useful information for users retrieving in the future. However, this scheme requires a redesign of the publishing/retrieving scheme, which may not be easily implemented. Moreover, it will also complicate the maintenance of the publishing tables and introduce more trac into the system.

3.5.3 Strict Fairness Scheme

The third solution is to adapt some eective incentive policy to mitigate the self-ish behavior of peers. Currently, aMule/eMule has already deployed a local credit system to reward the sharing behavior. However, this policy is neither ecient nor fair [45]. The KAD network has not employed any fairness strategy to pro-mote the publishing and retrieving process exclusively. Since the selsh behavior of peers will lead to the low lookup performance of the KAD network, some strict fairness modication must be introduced into the KAD network. One possible strict fairness scheme may base on how to fairly allocate sources location infor-mation among requestors. After receiving requests for sources location, the KAD network will return a dierent number of sources according to the credit that the requester has, while the credit value depends on peers' previous sharing behavior in the KAD network. The detailed policy is as follows.

• For the requester who stays longer and shares enough content location infor-mation in the KAD network, i.e., it will be awarded a high value of credit, the requestor would be provided all known sources information for the KAD network. Through this rule, publishers may be promoted to stay in the system longer for obtaining more credit.

• For the requester whose credit value is lower than a threshold or without credit, the number of returned sources information would be proportional to the requester's credit. This rule aims to punish publishers who will leave the system soon after publishing their downloaded content into the KAD network.

With this scheme, if a publisher leaves the system soon, the publisher's benet will be decreased when it tries to retrieve information in the future and vice versa. In addition, since adopting the fairness-based modication will keep the publishing/retrieving scheme of the KAD network unchanged, it may make the implementation relatively easy. However, even though the modication based on

Chapter 3 Section 3.6

Table 3.6: Comparison of dierent modication schemes Methods improve Pa improve the

amount of useful

current scheme No No No No No

self-maintenance Yes No No No Yes

chunk-based Yes Yes No Yes Yes

fairness policy Yes Yes Yes Yes Yes

the fairness design has a great advantage by fundamentally reducing the selsh behavior of peers and accordingly improve the lookup performance of the KAD network, the introduction of a strict fairness policy will still increase the complexity of the KAD network.

We have summarized the comparison of these modications in Table 3.6. It looks like no modication is perfect. We believe that a combination of these possible modications will be the most valid scheme to deal with the lookup per-formance issue of the KAD network.