Abstract

A popular technique for reducing the bandwidth load on Web servers is to serve the content from proxies. Typically these hosts are trusted by the clients and server not to modify the data that they proxy. SSL splitting is a new technique for guaranteeing the integrity of data served from proxies without requiring changes to Web clients. Instead of relaying an insecure HTTP connection, an SSL splitting proxy simulates a normal Secure Sockets Layer (SSL) connection with the client by merging authentication records from the server with data records from a cache. This technique reduces the bandwidth load on the server, while allowing an unmodified Web browser to verify that the data served from proxies is endorsed by the originating server.

SSL splitting is implemented as a patch to the industry-standard OpenSSL library, with which the server is linked. In experiments replaying two-hour access.log traces taken from LCS Web sites over an ADSL link, SSL splitting reduces bandwidth consumption of the server by between 25% and 90% depending on the warmth of the cache and the redundancy of the trace. Uncached requests forwarded through the proxy exhibit latencies within approximately 5% of those of an unmodified SSL server.

Keywords: SSL splitting; Barnraising; CDN; DHT; Untrusted proxy

Article Outline

1. Introduction

2. Goals

3. Design of SSL splitting

3.1. SSL overview

3.2. Interposing a proxy

3.3. Proxy-server protocol extensions

3.4. Dropping the encryption layer

3.5. Server-proxy signaling

4. Implementation

4.1. Server: modified OpenSSL

4.2. Proxy

4.3. Message formats

5. Cooperative Web caching using SSL splitting

5.1. Joining and leaving the proxy set

5.2. Redirection

5.3. Distributing the Web cache

5.4. Deploying barnraising

6. Evaluation

6.1. General experimental setup

6.2. Bandwidth savings for a single file

6.3. Bandwidth savings for trace-driven loads

6.3.1. Web trace files

6.3.2. Measurements

6.3.3. Simulations

6.4. Latency

7. Discussion

7.1. Transparent proxying

7.2. The implications of key exposure

7.3. Alternative proxy design

7.4. Feedback loop

8. Related work

8.1. Verifying integrity

8.2. HTTPS proxies

8.3. Content-distribution systems

9. Summary

Acknowledgements

References

Vitae