dc.contributor.author |
Bardinelli, Joshua |
|
dc.contributor.author |
Zhang, Yifan |
|
dc.contributor.author |
Su, Jianchang |
|
dc.contributor.author |
Huang, Linpu |
|
dc.contributor.author |
Parilla, Aidan |
|
dc.contributor.author |
Jarvi, Rachel |
|
dc.contributor.author |
Kulkarni, Sameer G. |
|
dc.contributor.author |
Zhang, Wei |
|
dc.contributor.other |
ACM Workshop on eBPF and Kernel Extensions (SIGCOMM 2024) |
|
dc.coverage.spatial |
Australia |
|
dc.date.accessioned |
2024-08-14T13:17:23Z |
|
dc.date.available |
2024-08-14T13:17:23Z |
|
dc.date.issued |
2024-08-04 |
|
dc.identifier.citation |
Bardinelli, Joshua; Zhang, Yifan; Su, Jianchang; Huang, Linpu; Parilla, Aidan; Jarvi, Rachel; Kulkarni, Sameer G. and Zhang, Wei, "hyDNS: acceleration of DNS through kernel space resolution", in the ACM Workshop on eBPF and Kernel Extensions (SIGCOMM 2024), Sydney, AU, Aug. 04-08, 2024. |
|
dc.identifier.uri |
https://doi.org/10.1145/3672197.3673439 |
|
dc.identifier.uri |
https://repository.iitgn.ac.in/handle/123456789/10342 |
|
dc.description.abstract |
The Domain Name System (DNS) is a core component of Internet infrastructure, mapping domain names to IP addresses. The recursive resolver plays a critical role in this process, requiring high performance due to multiple request-response exchanges. However, its performance is hindered by costly message copying, user-kernel space transitions, and kernel stack traversal. Kernel bypass techniques can mitigate these issues but often result in resource waste or deployment challenges.
To overcome these limitations, We present hyDNS, a hybrid recursive resolver that combines eBPF offloading in the kernel with a user-space resolver. The DNS kernel cache allows most requests to be served before reaching the kernel network stack. To manage limited DMA memory, excess requests are passed to user space once a threshold is reached, enabling the system to handle high query loads. hyDNS uses programmable NICs to create a scalable kernel cache, implementing a lockless per-core eBPF hash map. Filters on the NIC direct requests to each core. Preliminary results show significant performance improvements with eBPF offloading, achieving up to 4.4× the throughput and a 65% reduction in latency compared to user space implementations. |
|
dc.description.statementofresponsibility |
by Joshua Bardinelli, Yifan Zhang, Jianchang Su, Linpu Huang, Aidan Parilla, Rachel Jarvi, Sameer G. Kulkarni and Wei Zhang |
|
dc.language.iso |
en_US |
|
dc.publisher |
Association for Computing Machinery (ACM) |
|
dc.subject |
Domain Name System (DNS) |
|
dc.subject |
eBPF (extended Berkeley Packet Filter) |
|
dc.subject |
XDP (eXpress Data Path) |
|
dc.subject |
In-Kernel Cache |
|
dc.title |
hyDNS: acceleration of DNS through kernel space resolution |
|
dc.type |
Conference Paper |
|