Single and Multiple Buffer Processing

Breaking the Barrier of 2 for the Competitiveness of Longest Queue Drop

“…Although LQD is the best known online algorithm for buffer management in shared-memory switches, determining its true competitiveness remains an elusive problem and has been described as a significant open problem in buffer management [20,32]. After the initial analysis which showed that LQD is 2-competitive and not better than √ 2-competitive [1,21] progress on the upper bound has been limited to special cases (e.g.…”

“…We refer the reader to the survey by Goldwasser [20] for an overview of online algorithms for buffer management problems. Additionally, the survey of Nikolenko and Kogan [32] incorporates some more recent work to several of the problems described in [20]. In the following, we discuss some of the results related to online buffer management for switches.…”

Deadline-Aware Energy Management in Data Centers

“…The methods aiming to reduce data center power consumption could be classified into four approaches [1]: powerproportionality, which attempt to guarantee that servers consuming power in proportion to their utilization [2]- [4]; energyefficient server design, which attempt to determine the proper server architecture for a given workload [5]- [7]; dynamic server provisioning, which attempts to determine the times when the servers should be kept on or off [8], [9]; consolidation and virtualization, which attempts to reduce the power consumption by resource sharing [10]- [12]. We refer the reader to [13]- [17] for further literature related to our work.…”

BASEL (Buffer mAnagement SpEcification Language)

“…// considered priorities for admission and // processing fifo(p1,p2) = (p1.arrival < p2.arrival) srpt(p1,p2) = (p1.processing < p2.processing) rsrpt(p1,p2) = (p1.processing > p2.processing) // default congestion condition for all // considered policies defCongestion() = lambda q, (q.currSize >= q.size) // initializing a generic buffering architecture q1=Queue(B); out=Port(q1); q1.proPrio(p1,p2)=fifo(p1,p2); q1.congestion=defCongestion(q1); [20,28]. Each row represents a management algorithm for a single queue; e.g., the first row shows a simple greedy algorithm that admits every incoming packet if possible (see congestion()), and processes them in fifo() order; it is O(k)-competitive for maximum processing requirement k. In BASEL, this algorithm looks as follows: q1.admPrio=fifo; q1.procPrio=fifo;…”

“…// LQF: HOL packet from Longest-Queue-First lqf(q1,q2) = (q1.currSize > q2.currSize); // SQF: HOL packet from Shortest-Queue-First sqf(q1,q2) = (q1.currSize < q2.currSize); // MAXQF: HOL packet from queue that // admits max processing maxqf(q1,q2)= (q1.weightSched > q2.weightSched); // MINQF: HOL packet from queue that admits // min processing minqf(q1,q2)= (q1.weightSched < q2.weightSched); // CRR: Round-Robin with per cycle resolution crr(q1,q2) = (q1.weightSched < q2.weightSched); crrPostSchedAct() = lambda port, (port.getCurrQueue().weightSched += k); // PRR: Round-Robin with per packet resolution prr(q1,q2) = (q1.weightSched < q2.weightSched); prrPostSchedAct() = lambda port, (let q = port.getCurrQueue() in if (q.getHOL().processing == 0) q.weightSched += k * k)); Table 2 summarizes various online scheduling policies as shown in [21,28]. Observe that buffer occupancy is not a good characteristic for throughput maximization: lqf() and sqf() have bad competitive ratios, while a simple greedy scheduling policy Min-Queue-First (MQF) that processes the HOL packet from the non-empty queue with minimal required processing (minqf()) is 2-competitive.…”