Remote Storage Replication
Our work on storage focuses on a remote replication service for disaster recovery and business continuance. We have examined the performance of a commercial replication sysetm and its feasibility for remote replication. The network latency due to synchronous replication is difficult to tolerate in scenarios where businesses
are required by regulation to separate their secondary sites from the primary by hundreds of miles. We propose a semantic-aware remote replication system to meet the contrasting needs of both system efficiency and safe remote replication with tight recovery-point and recovery-time objectives. A position paper on this subject is set to appear in Oct. 2006 at the 2nd International Workshop on Storage Security and Survivability (storagess_paper).
I have been working on congestion control and avoidance for a long time, with particular emphasis on explicit feedback of congestion information from the network. I have been involved in the congestion control efforts of most packet networking technologies, especially TCP/IP, Frame Relay, ATM, and IEEE 802.17 (Resilient Packet Rings.) Sally Floyd and I have proposed the use of Explicit Congestion Notification (ECN) for TCP/IP networks. A web page (including a pointer to RFC 3168, which is a Proposed Standard) is: ECN (Explicit Congestion Notification) in TCP/IP .
Our work on the Distributed Open Signaling Architecture (DOSA) architecture was one of the first to integrate Quality of Service with Call Signaling for IP Telephony. A paper on DOSA is: (DOSA-Infocom 2000 Paper). Some of these ideas were patented, and in October 2003, we received the AT&T Strategic Patent Award, in recognition of a patent that significantly contributes to AT&T's business (see AT&T Strategic Patent Award ). Our architectural work on IP Telephony, "Telephony Over Packet networkS (TOPS)". TOPS allows users to move between terminals or to use mobile terminals while being reachable by the same name. TOPS users can have multiple terminals and control how calls are routed to them. TOPS allows for terminals with a range of capabilities such as support for video, whiteboard and other media with a variety of coding formats. TOPS retains the necessary information on terminal capabilities to determine the appropriate type of communication to be established with the remote terminal. The architecture assumes that the underlying network supports the establishment of end-to-end connectivity between terminals, with an appropriate quality of service. A paper on this appeared in IEEE JSAC (TOPS-paper).
At AT&T, in the past I have worked on Multicast, Signaling (Lightweight signaling for ATM), IP over ATM. IP Telephony and Networked-multimedia support. Before coming to AT&T, I worked on congestion management and network I/O while at Digital Equipment Corporation for 11 years.
1. Congestion Avoidance and Control
a. Connectionless Networks:
I have worked on congestion control for connectionless packet oriented networks with window flow control such as DECnet, OSI and TCP/IP for a long time. The most significant work on this is the so-called DECbit work, which uses a single bit feedback for congestion avoidance. Its mechanisms for increase/decrease (Additive Increase Multiplicative Decrease) and a binary feedback for explicit congestion notification have been widely adopted in many different network architectures including TCP/IP, OSI, ATM and Frame Relay. A paper on this was published in ACM Sigcomm 1988 and the full paper was published in the ACM Transactions on Computer Systems in 1990 (DECbit-TOCS). Numerous other papers were published on this topic. An early version of a paper on the high-level principles behind DECbit is DEC-TR-506.
One follow-on work to achieve Max-Min fairness with binary feedback used "Selective Feedback" (DEC-TR-510).
b. ATM Networks
Between 1995 and 1997, I worked on issues related to ATM. I worked on algorithms for achieving max-min fairness for the ABR service in ATM. Issues of scale were of concern. Scalability is aided by efficient techniques to compute the max-min fair allocation, (described in HPN_95, brdcom96,) as well as minimizing the amount of state information to be maintained (using discrete rates ). We have published several papers on a rate-allocation algorithm for the ABR service. I also looked at the interaction of TCP's window-flow controlled mechanisms with a rate-controlled environment.
We have also analyzed in detail the effect of bi-directional traffic on window flow control protocols such as TCP. The effect of ack-compression is a degradation in throughput and unfairness. We quantify it. We have also looked at its effect when the channels are asymmetric - when the bandwidth in one direction is substantially smaller than the other.
A formal specification of the source/destination behavior that is specified in the ATM Forum's Traffic Management specification is another activity I have been involved in efsm ).
Other work related to ATM congestion management is in understanding
the issues of time-scale, (described in infocom_96 ) and exploring ways of
integrating different mechanisms for ATM congestion management ( rate_credit integration ).
2. Support for Compressed Video over Networks
I also worked on supporting compressed video (source-adaptable
semi-real-time) traffic on ATM networks, using the Explicit Rate
algorithms of the type recommended for the ABR service explicit_rate_video ), and on
multipoint-to-multipoint communication over ATM seam ). A more detailed paper on
support of compressed video over rate based control schemes appeard in
Infocom 1997, and a journal version in IEEE/ACM Transactions on
Networking in 1999 (explicit-rate-video-TON).
I also worked on smoothing of compressed video. Supporting compressed video efficiently on networks is a challenge because of its burstiness. Although a large number of applications using compressed video allow adaptive rates, it is also important to preserve quality as much as possible. We propose a smoothing and rate
adaptation algorithm for compressed video, called SAVE, that is used in conjunction with explicit rate based control in the network. SAVE smooths the demand from the source to the network, thus helping achieve good multiplexing gains. SAVE maintains the quality of the video and ensures that the delay at the source buffer does not exceed a bound. We show that SAVE is effective by demonstrating its performance across 28 different traces (entertainment and teleconferencing videos)
that use different compression algorithms. Two journal papers related to SAVE are: IEEE/ACM Transactions on Networking, 1998 (SAVE-TON-1998) and IEEE Transactions on Multimedia, 1999 (SAVE-Multimedia).
We proposed techniques to carry real-time data efficiently over ATM,
using an encapsulation that we call Real-Time AAL5. This was discussed
at the ATM Forum, and is described in Forum_SAA_0139.
3. Network I/O
I have worked on issues related to high performance Network I/O. In particular, at Digitial Equipment Corporation, I worked on architecture and scheduling for network adapters and the protocol implementation in end-systems and host operating systems. A paper on Performance Considerations for Designing Network Interfaces appeared in IEEE JSAC in 1993 (Designing-NICS-JSAC-1993). Another paper on scheduling considerations in the host operating system for network I/O, particularly to mitigate "receive livelocks" appeared in ACM Transactions on Computer Systems in 1997 (kk-mogul-1997).
Work of Interest.
The first A few of other areas I have worked on, while at Digital Equipment Corporation were in Operating System support for ATM, particularly IP over ATM, Operating System support for a Video-on-Demand Server ( VOD_paper ), and on the issues of network overload in Interrupt-Driven kernels (Networking Implementation ). An interesting effect of the CSMA/CD protocol used on Ethernet is called the "Ethernet Capture Effect". We came up with an interesting solution for the problem which is in widespread use (Capture Effect.) My past work at DEC included work on DECbit, on network I/O (adapters and protocol implementation in end-systems) and file system performance.