LUMP: LocationtoURL mapping draftschulzrinneecritlump

1
LUMP Location-to-URL mappingdraft-schulzrinne-ec
rit-lump

• Henning Schulzrinne
• Columbia University
• Dept. of Computer Science

2
LUMP Mapping service URNs locations to URLs

• Common problem
• geo or civic location, service ? set of URLs
• e.g., Broadway/NY, 911 ? fire_at_psap.nyc.gov
• also applies to anything from AAA to pizza
  delivery
• Service providers dont trust each other (fully)
• e.g., who gets to include Jerusalem in its map
• service may depend which warlord you belong to ?
• cant wait for UN (or ICANN) to create global
  emergency services database
• Suggested approach new distributed mapping
  protocol
• LUMP location-to-URL mapping protocol
• uses SOAP, but special service URLs

3
LUMP Location-to-URL Mapping
VSP1
cluster serving VSP1
replicate root information
cluster serves VSP2
123 Broad Ave Leonia Bergen County NJ US
root nodes
NY US
NJ US
sippsap_at_leonianj.gov
search referral
Bergen County NJ US
Leonia NJ US
4
LUMP architecture
G
tree guide
G
G
G
broadcast (gossip)
T1 .us T2 .de
G
resolver
T2 (.de)
seeker 313 Westview Leonia, NJ US
T3 (.dk)
T1 (.us)
Leonia, NJ ? sippsap_at_leonianj.gov
5
Caching

• Generally, UA caches lookup results
• query Im at (X,Y), whats my PSAP?
• answer Your PSAP is sippsap_at_town.gov as long
  as you stay in polygon (X1,Y1 X2, Y2 ) this
  is valid for 12 hours
• almost no impact of node mobility on query
  frequency
• same for civic as long as you stay on Main
  Street, your town
• civic only relevant for nomadic users
• actual PSAP coverage area may be larger ? just an
  optimization
• Almost always avoids query during emergency call
• MAY re-query during call
• load distribution via DNS
• given frequency of calls for one resolver, likely
  to be no DNS caching anyway
• Further optimization query with timestamp (or
  etag) of last answer
• answer still the same, thanks for asking

6
Performance notes

• Use TCP (or TCP/TLS) for reliability
• Expect 1-2 queries/day/client
• Typical gtgt 100 queries/second/server
• almost all rows will be cached in memory
• only about 6,000 rows
• one server ? 8,640,000 queries
• probably N1 spared
• data center cost 300/month/server ?
  0.0003/user/month (1Mq/day)

7
Implementation status

• Prototype implementation
• includes referrals
• both geo and civic coordinates
• from draft WSDL (with minor fixes)
• Server
• Axis (Apache) SOAP server
• Postgres SQL geo database
• does polygon intersection
• Client
• Java servlet (web page)
• Tcl (for our SIP client)