Name
Abstract | ||
|---|---|---|
Network measurement is necessary to obtain an understanding of the network traffic and keep the network healthy. Flow-level measurement is widely used because it provides rich enough information while being resource efficient, in contrast to the traffic-level solutions like SNMP and packet-level solutions like tcpdump. Flow-level measurement requires to maintain a TFR for each active flow, and update the corresponding TFR when a packet arrives. In the scenario of high-speed networks, the processing time for each packet is only several or tens of nanoseconds, and there may be up to millions of concurrent active flows. Two requirements have to be met to be able to do flow-level measurement in this scenario, i.e., the memory size have to be large enough to maintain a TFR for each flow, and the access time of the memory have to be small enough to be able to update the TFR for each packet. Unfortunately, SRAM, which is fast enough, is expensive and there are limited amount available for the measurement tasks, while DRAM, which is large in capacity, is slow and cannot support the per-packet access. In this paper, we proposed a solution, PriMe, by combining SRAM with DRAM, overcoming the resource limitation by exporting TFRs to the DRAM when necessary, while enjoying the fast access of SRAM. We designed a peculiar algorithm to reduce the frequency of exporting TFRs to DRAM, as well as an algorithm to reduce the accesses of DRAM in aggregating the TFRs exported from SRAM. |
Year | Venue | DocType |
|---|---|---|
2018 | arXiv: Networking and Internet Architecture | Journal |
Volume | Citations | PageRank |
abs/1812.02861 | 0 | 0.34 |
References | Authors | |
0 | 4 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Zongyi Zhao | 1 | 0 | 1.01 |
| Xingang Shi | 2 | 166 | 22.66 |
| Xia Yin | 3 | 320 | 44.72 |
| xie | 4 | 106 | 36.98 |