Grant: Discovery Grants Program - Individual
University of Toronto
$42,000.00
Adapting Computer Networks to Modern Application Requirements
EF 2024-25
Résumé
In our increasingly digital world, computer networks form the lifeblood of sectors ranging from business and education to entertainment and healthcare. But keeping pace with rising bandwidth and performance demands is proving difficult. These challenges are heightened by the end of Moore's law which restricts our ability to increase capacity simply by adding more hardware.
The common approach to meet these escalating demands is special-purpose design. This approach involves closely integrating network and application functionalities for joint optimization. For instance, training of Large Language Models (e.g., ChatGPT, Llama, or Bard) requires extremely high performance networks tightly coupled with application requirements through extensive research and development efforts across all layers of the network stack and end-hosts.
Special-purpose design is not without its challenges though: 1) it is very costly (monetary investments and R 2) it is fraught with risk - what if the design doesn't meet the application's needs? 3) it is often very often slow, leading to delayed development cycles.
In light of these challenges, we ponder: what if the network could auto-adjust to meet application requirements? In other words, can we create a dynamic network, intelligent enough to understand the demands of its applications and adapt accordingly?
We believe the key to answering this question is to create new information exchange channels between applications and the network. Current efforts in this direction mainly focus on creating APIs between applications and networks where applications explicitly share their requirements with the network. However, this approach relies on application developers to understand network requirements and limitations, which can be impractical due to the complexity of developing large-scale distributed applications.
Alternatively, we propose systematic and automatic integration of networks and applications. The key is to use correlations of network events (e.g., packet loss resulting from microbursts) with events generated on the end-host (application and transport layer events such as start of a new flow, or a sudden change in rate). Our preliminary works in this area demonstrate the feasibility of creating such signals in an automated manner and their potential to address complex issues.
Our proposal paves the way for a wide range of network solutions. Signals from end-hosts that show significant change in traffic can help with adaptive routing and load balancing, dynamic flow and packet scheduling, etc. based on future events as opposed to current state of the network used today. Challenging problems like incast, and microburst mitigation can become much simpler. We can even envision switches aggregating the information they receive from end-hosts and provide much richer congestion signals to the senders helping with congestion control.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-05752
Name nom
Ganjali, Yashar
Department département
Computer Science
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
42000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
803
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer communications
Area of application fr
Téléinformatique
Research subject code
2718
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Communication and information theory
Research subject fr
Théorie des télécommunications et de l'information
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Adapting Computer Networks to Modern Application Requirements
Keywords
computer networking; network systems; transport protocols; high performance; performance optimization; data center networks
Application summary
In our increasingly digital world, computer networks form the lifeblood of sectors ranging from business and education to entertainment and healthcare. But keeping pace with rising bandwidth and performance demands is proving difficult. These challenges are heightened by the end of Moore's law which restricts our ability to increase capacity simply by adding more hardware.
The common approach to meet these escalating demands is special-purpose design. This approach involves closely integrating network and application functionalities for joint optimization. For instance, training of Large Language Models (e.g., ChatGPT, Llama, or Bard) requires extremely high performance networks tightly coupled with application requirements through extensive research and development efforts across all layers of the network stack and end-hosts.
Special-purpose design is not without its challenges though: 1) it is very costly (monetary investments and R 2) it is fraught with risk - what if the design doesn't meet the application's needs? 3) it is often very often slow, leading to delayed development cycles.
In light of these challenges, we ponder: what if the network could auto-adjust to meet application requirements? In other words, can we create a dynamic network, intelligent enough to understand the demands of its applications and adapt accordingly?
We believe the key to answering this question is to create new information exchange channels between applications and the network. Current efforts in this direction mainly focus on creating APIs between applications and networks where applications explicitly share their requirements with the network. However, this approach relies on application developers to understand network requirements and limitations, which can be impractical due to the complexity of developing large-scale distributed applications.
Alternatively, we propose systematic and automatic integration of networks and applications. The key is to use correlations of network events (e.g., packet loss resulting from microbursts) with events generated on the end-host (application and transport layer events such as start of a new flow, or a sudden change in rate). Our preliminary works in this area demonstrate the feasibility of creating such signals in an automated manner and their potential to address complex issues.
Our proposal paves the way for a wide range of network solutions. Signals from end-hosts that show significant change in traffic can help with adaptive routing and load balancing, dynamic flow and packet scheduling, etc. based on future events as opposed to current state of the network used today. Challenging problems like incast, and microburst mitigation can become much simpler. We can even envision switches aggregating the information they receive from end-hosts and provide much richer congestion signals to the senders helping with congestion control.