TY - JOUR
T1 - Time to live of identifier-to-locator mappings
T2 - With-reset or no-reset
AU - Li, Xiaoqian
AU - Zukerman, Moshe
AU - Zhou, Huachun
AU - Luo, Hongbin
AU - Qiu, Feng
AU - Zhang, Hongke
PY - 2014/10/1
Y1 - 2014/10/1
N2 - In networks with identifier/locator separation, border routers often cache recently used identifier-to-locator mappings of remote hosts and set a time to live (TTL) for each mapping. We consider two options to update the TTL of a mapping, designated as no-reset (NR) and with-reset (WR) modes. In the NR mode, once a mapping is cached, its TTL is set to a default value called cache timeout, and the mapping is removed when the TTL expires. By contrast, in the WR mode, whenever the mapping is queried, its TTL is reset to the cache timeout. Although it is not difficult to intuitively observe that the WR mode performs better than the NR mode in terms of cache hit rate, it is still of interest to evaluate what is the benefit gained by using theWR mode rather than the NR mode and what is the cost incurred to achieve this gain. Another equivalent question is what is the additional cost incurred in terms of resources for the NR mode to perform similarly to the WR mode. We model the cache hit rate and compare the performance of the WR and NR modes for flat and aggregatable identifiers by analysis and simulation, and estimate how much more cache size the NR mode requires to achieve similar cache hit rate and communication interruptions to the WR mode. The comparison results provide rational to use the WR mode, rather than the NR mode, for identifier-to-locator mapping caches.
AB - In networks with identifier/locator separation, border routers often cache recently used identifier-to-locator mappings of remote hosts and set a time to live (TTL) for each mapping. We consider two options to update the TTL of a mapping, designated as no-reset (NR) and with-reset (WR) modes. In the NR mode, once a mapping is cached, its TTL is set to a default value called cache timeout, and the mapping is removed when the TTL expires. By contrast, in the WR mode, whenever the mapping is queried, its TTL is reset to the cache timeout. Although it is not difficult to intuitively observe that the WR mode performs better than the NR mode in terms of cache hit rate, it is still of interest to evaluate what is the benefit gained by using theWR mode rather than the NR mode and what is the cost incurred to achieve this gain. Another equivalent question is what is the additional cost incurred in terms of resources for the NR mode to perform similarly to the WR mode. We model the cache hit rate and compare the performance of the WR and NR modes for flat and aggregatable identifiers by analysis and simulation, and estimate how much more cache size the NR mode requires to achieve similar cache hit rate and communication interruptions to the WR mode. The comparison results provide rational to use the WR mode, rather than the NR mode, for identifier-to-locator mapping caches.
KW - Cache miss
KW - Communication interruptions
KW - Hit rate
KW - Identifier-to-locator mapping
KW - Model
KW - TTL reset
UR - http://www.scopus.com/inward/record.url?scp=84911985140&partnerID=8YFLogxK
UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-84911985140&origin=recordpage
U2 - 10.1002/dac.2478
DO - 10.1002/dac.2478
M3 - RGC 21 - Publication in refereed journal
SN - 1074-5351
VL - 27
SP - 2324
EP - 2342
JO - International Journal of Communication Systems
JF - International Journal of Communication Systems
IS - 10
ER -