What characterics of passive measurement tools (e.g., routers and monitors) are most important for backbone engineers?
How critical is the trade-off between hardware performance and convenience of having stats in existing equipment?
How critical is the operating system (UNIX vs. DOS) on a stand-alone monitor
How great is the need for monitors - DS3, OC3, OC12, OC48?
What data is most meaningful to collect with respect to:
diagnosing network problems
capacity planning and trends analysis
analysis of the effects of changes in topology and trends in traffic
Characterization:
How is traffic (e.g., specific applications and protocols) changing over time?
Is traffic growth superlinear in number of attached users? E.g., Ithiela de Sola Pool's theory that 'the more people attach, the more valuable the network is to each of them, since there's more folks you'd want to talk to so the PER-user imposition on the network increases... '
If superlinear growth is expected then are pricing model adjusting at a pace to permit ISPs to install capacity sufficient to meet increased demand?
What are the implications of Internet growth for customer differentiation?
How is growth expected to effect networks' performance?
How are trends in Net usage ( e.g., audio, multimedia, multicast, push technologies) effecting evolving infrastructures, e.g., the need for specialized servers, routers, etc.?
Is the pace of innovation (both in equipment and applications) sustainable and what is the effect of obsolete hardware/software in the infrastructure on Net performance?
How much traffic growth is added users (customer growth) or additional traffic PER user?
What would be required to have routers provide full AS path information in real-time?
How much of typical traffic is local vs. wide area, e.g., as depicted through analysis of AS path lengths and hop lengths?
How much locality/favoritism characterizes traffic in a given trace? and how would one best draw or compare these data?
What does the balance of traffic look like?
trade balance, e.g., US import/export
webhosting ISP / transit ISP / user ISP, potentially for eventual settlement purposes
Are packet characteristics (sizes, traffic groupings) changing and (if so) are equipment vendors addressing those changes in next generation designs? What about equipment in the existing infrastructure?
Flows:
How elastic are flows? i.e., how much traffic is responsive to network congestion signals (e.g., TCP out of slow start).
How does the number of flows change as a function of the flow granularity?
When is the SYN/FIN method needed for analysis, versus the flow model? Howmuch difference is there for a given metric? How does this measure relate to specific technologies, e.g., RED, WFQ, traffic shaping, RSVP/state?
Hardware:
How should/can the router modulate its own buffer tuning mechanisms corresponding to packet size distributions of (expected) workloads?
What are the most meaningful traffic measurements affecting router design and router performance? E.g., patterns (timing & sequencing) of packet arrivalsto routers; intra-flow packet arrival distributions (particularly for per-flow differentiated services or weighted queueing/service mechanisms).
How can passive metrics best be used to gage the efficiency of network's use of bandwidth for a given payload? How do layer 3 protocols and applications respond to evidence of network congestion?
UDP and TCP Specific Issues:
How well is TCP/IP accommodating the growth in the Internet? What are ISPs' concerns (if any) about TCP as it relates to evolving traffic patterns and hardware?
How are changes in the performance of DNS traffic affecting ISPs? Is DNS carrying enough more semantics now that most packets are above 512 bytes and hus requiring the establisment of a TCP connection? What affect might protocol changes (e.g. DNS authentication, DNS routing) have on the existing infrastructure?
Will the trend be toward less UDP DNS and more TCP DNS, which will incur more load on the network and servers? Would this make the infrastructure more critical to do hierarchical delegations of subdomains?
How are the 'loads' on DNS root name servers changing? and do we need to monitor/rethink the architecture for increased (hierarchical) scaling/robustness/survivability?
