Speaker
Petr Špaček
(Internet Systems Consortium (ISC))
Description
Continuing the series of talks about DNS benchmarks, we focus on zone transfers and their specifics. Zone transfers are now far more frequent than they were when the protocol was designed, and some providers have SLAs, a maximum amount of time in which to ensure a zone update is available in the DNS.
On the technical level zone transfers have some unique properties which make them very distinct from a typical DNS query:
- Primary transport protocol is TCP or TLS
- Starting a single zone transfer is multi-step process (NOTIFY, SOA query, AXFR/IXFR)
- Secondary server in clean state potentially needs to transfer
- hundreds of thousands of small zones
- or a single giant zone
- or any combination in between
- Amount of data transferred can be large - thus bandwidth and memory consumption are a concern
- Transfers can take long time - but the server must respond to other clients while transfers are in progress
- For incoming incremental transfers, the client might need to recombine previous local knowledge with the new data efficiently
- For outgoing incremental transfers, the server might need to compute differences between old and new versions of the zone
- ... and many others
As a result, meaningful zone transfer benchmark requires different tools and test methodology than traditional queries-per-second and answer latency benchmarks.
In this talk we present lessons learned from benchmarking zone transfers for the BIND project.
| Talk duration | 20 Minutes (+5 for Q&A) |
|---|
Primary author
Petr Špaček
(Internet Systems Consortium (ISC))
