Abstract
A key challenge in designing load balancing strategies is to achieve low delay in large-scale systems while only using minimal communication overhead. Motivated by these issues, we introduce a novel scheme in which the dispatcher becomes aware of idle servers without any explicit communication from either side, using absence of messages at predefined time instants. The proposed scheme achieves provably vanishing queueing delays while using strictly less than one message per job on average.
Highlights
Well-designed load balancing strategies provide an effective mechanism for improving delay performance and achieving efficient resource utilization in parallel-processing systems such as cloud networks and data centers
Besides these typical performance criteria, communication overhead and implementation complexity have emerged as crucial concerns due to the immense numbers of servers in large-scale deployments. Such scalability challenges have fueled an urgent interest in load balancing algorithms that yield excellent delay performance while only requiring low implementation overhead
The JOQ algorithm greatly reduces the number of messages exchanged in order to make the probability of queueing and the expected queueing delay vanish, compared to existing algorithms in the literature
Summary
Well-designed load balancing strategies provide an effective mechanism for improving delay performance and achieving efficient resource utilization in parallel-processing systems such as cloud networks and data centers. This is inevitable, since results in the seminal paper [5] show that it is fundamentally impossible to achieve a vanishing queueing delay with a finite communication overhead per job, unless memory is available at the dispatcher to store state information The latter is exactly what is done in the so-called Join-theIdle-Queue (JIQ) scheme [2,7], where servers advertise their availability by sending a ‘token’ to the dispatcher whenever they become idle, generating at most one message per job. In the present paper we debunk the abovementioned notion and introduce a scheme that allows for vanishing queueing delays and that uses strictly less than one message per job on average This scheme exploits the crucial insight that an idle state need not be explicitly signaled by using a message, but can be implicitly inferred by the dispatcher when not receiving a message from a server at a pre-arranged time instant.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
