Measuring control plane latency in SDN-enabled switches
read more
Citations
Software-defined networking (SDN): a survey
Detour: Dynamic Task Offloading in Software-Defined Fog for IoT Applications
Incremental Flow Scheduling and Routing in Time-Sensitive Software-Defined Networks
Survey of Consistent Software-Defined Network Updates
STAR: Preventing flow-table overflow in software-defined networks
References
OpenFlow: enabling innovation in campus networks
B4: experience with a globally-deployed software defined wan
Onix: a distributed control platform for large-scale production networks
DevoFlow: scaling flow management for high-performance networks
Achieving high utilization with software-driven WAN
Related Papers (5)
Frequently Asked Questions (14)
Q2. What is the primary factor contributing to inbound latency?
Their conversations with the switch vendor suggest that the limited bus bandwidth between the ASIC and switch CPU is the primary factor contributing to inbound latency.
Q3. What causes the underlying problems of SDN?
The authors find that the underlying causes are linked to software inefficiencies, as well as pathological interactions between switch hardware properties (shared resources and how forwarding rules are organized) and the control operation workload (the order of operations, and concurrent switch activities).
Q4. What is the reason for the delay in BCM-1.3?
Conversations with Broadcom indicated that TCAM modification should ideally be fast and independent of table size, so the underlying cause appears to be less optimized switch software in BCM-1.0.
Q5. How does the research examine latencies in production switches?
(2) The authors find that outbound latency, i.e., the latency involved in the switch installing/modifying/deleting forwarding rules provided by control applications, is also high—3ms and 30ms per rule for insertion and modification, respectively, in Broadcom.
Q6. What is the reason why the authors studied 3 commercial switches?
In [8], the authors also studied 3 commercial switches (HP Procurve, Fulcrum, Quanta) and found that delay distributions were distinct, mainly due to variable control delays.
Q7. How many rules are inserted in the TCAM?
the authors observe that the burst of B rules is divided into several groups, and each group is reordered and inserted in the TCAM in order of increasing priority.
Q8. What is the reason for the latency of rule deletions?
The authors conclude that the per-rule modification latency on BCM-1.0 and IBM is impacted purely by table occupancy, not by rule priority structure.
Q9. What is the effect of the step-function effect?
The authors see two effects: (1) the latencies alternate between two modes at any given time, and (2) there is a step-function effect after every 300 or so rules.
Q10. What is the way to schedule a set of rule updates?
Dionysus [11] optimally schedules a set of rule updates while maintaining desirable consistency properties (e.g., no loops and no blackholes).
Q11. How do the authors determine whether the latencies will ever reach what hardware can support?
given that software will continue to bridge control and data planes in SDN switches, the authors remain skeptical whether latencies will ever reach what hardware can natively support.
Q12. How is the insertion latency in the case?
Even in the best case (Intel), per-rule insertion latency of 1ms is higher than what native TCAM hardware can support (100M updates/s [1]).
Q13. Which open flow protocol supports rules defined over any header fields?
For Intel, BCM-1.0, and IBM, the authors install/modify/delete rules in the single table supported by OpenFlow 1.0; for BCM1.3, the authors use the highest numbered table, which supports rules defined over any L2, L3, or L4 header fields.
Q14. How many rules are inserted into a table with low priority?
In contrast, in Figure 6b, newly inserted high priority rules will displace low priority rules in the table, so when S = 400 the completion time is about 3x higher than S = 100.