Locating Mobile Agents in Distributed Computing
Environment
Reference
[1] A. Di Stefano, and C. Santoro, “Locating Mobile Agents in a Wide Distributed Environment,” IEEE T-PDS, vol 13, no. 8, Aug 2002, p. 844 – 864.
SPC - Proof of Correctness
• Temporal evolution of SPC
• Finite State Machine
• Reachability
• Completion
Query formulation and response
• q(r,m,t): Query q that returns the location of agent named m and registered at r (SAR or RAR) at time t.
• Result could be– found, successful completion of location phase– wait, while agent is migrating.
Inter region migration
Location Finding Protocol: FSM – Figure 9
No cycles. States are reachable. All paths from initial state end at the final state (S1 → S11). So agent can always be found.
Three cases of interaction & interregion migration
• Interaction before interregion migrationS1 → S2 → S5 → S11
• Interaction after interregion migrationS1 → S4 → S8 → S11
• Interaction concurrently with an interregion migration
Intraregion Migration
• FSM shows that before reaching λl the protocol finds λl.region; so the correctness proof requires confirmation that location of the correct RAR leads to the correct SAR and agent location.
Intraregion migration
Interregion – FSM – Fig 10
• Interaction before intraregion migrationS1 → S2 → S7
• Interaction after intraregion migrationS1 → S4 → S7
• Interaction concurrently with an intraregion migration
Protocol Termination Condition• Single interaction: single migration (considered)• Several interactions overlap single migrations (handled by
SAR lock)• Single interaction overlap with several migrations,
– e.g. finding phase overlaps 2 migration– RAR points to SAR, but agent migrates before SAR is queried– Chasing an agent – conditions to catch up?
Location queries
Agent migration
SAR
RAR: m in region
SARSAR (m)
SAR
RAR
SAR(m)
SAR ()SAR(m)
Query propagation
• If agent has migrated, instead of SAR/RAR sending query reply propogate query to the next location.
Figure 11: Query propogation
Figure 12: Time intervals
Termination condition derivation
Figure 12: Time intervals
(14) is hard to compute. Two simplified conditions emerge.
On iteration this yields:
•Agent α motion λ1, λ2, λ3, …. •Agent β at λ wants interaction with α
•At t location query sent to SAR λi .•At t – δ; α begins to migrate to λi+1; query sent to RAR λi.reg before t – δ. •To catchup to the agent the query to SAR λi+1 has to be before α starts a new migration.
Figure 13 Notation
Availability Comparison
• Availability = MTTF/(MTTF+MTTR)
• MTTF = 1/Λ, where Λ = average fault rate
– Compute fault rate for migration and interaction phases– Migration fault rate = Agent transfer fault rate + location updating fault rate
• Database logging (DL) approach (Figure 17)
• Path Proxies (PP) approach (Figure 18)
• SPC approach (Figure 19)– Fault rate of SAR for intraregion migration
• Fault rate of a generic site (SAR)
– Fault rate of SAR and RAR for interregion migration• Fault rate of a generic site (SAR) +
• Fault rate of a specialized site (e.g. ANS)
• Fault rate of a generic site on a LAN (RAR)
– Ratio of interregion to intraregion migration
• If source site fails then migration also fails. So ignore the source site failure. (Figure 20)
Availability Comparisons
Location interaction fault rate
• Database Logging – access the location database (Eq 21)
• Path Proxies – complete path from the birth location has to be accessed (Eq 22)
• Search by Path Chase – accesses involved updating during the migration processes; best case – all updates; worst case – only mandatory updates (Eq 23).– General solution depends on the number of regions kr
and number of locations ks in the search path. (Minimum vlues are 2 and 0 respectively)
Interaction fault rates
Discussion
• What about path length?
• Is fault rate uniform, e.g. WAN vs LAN?
• Another approach to assess availability?
Scalability• With respect to number of agents n(t), and number
of agent migrations s(i,t).– Impact the network usage UN
• Function of the number of agents• If there is no congestion (i.e. messages per link are acceptable),
then focus on US. – SPC db is distributed. We would require a distributed db for DL.
– Impacts site usage US.• Function of the number of entries in all the databases
(including proxy objects)• Improve performance (Fig 15) by
– Limiting of agents (db entries) per SAR.– FIFO is used to limit agents per SAR.
Scalability comparisons
Fig. 15 Site usage (scalability)
Summary
Top Related