Title: Extreme Networking Achieving Nonstop Network Operation Under Extreme Operating Conditions
1Extreme NetworkingAchieving Nonstop Network
Operation Under Extreme Operating Conditions
- Fred Kuhns fredk_at_cs.wustl.eduhttp//www.arl.wust
l.edu/arl
Jon Turnerjst_at_cs.wustl.eduhttp//www.arl.wustl.
edu/arl
2Motivation
- Internet subject to extreme traffic conditions.
- correlated user behavior selfish and/or
malicious users - Growing reliance on data networks.
- higher expectations for reliability and
performance - Design networks for worst-case traffic
conditions. - practice constructive paranoia
- provide carefully regulated reserved bandwidth
services - better queueing mechanisms for traffic isolation
- network mechanisms to protect web sites from DDOS
- plan for continuous upgrading of network
infrastructure - extensible routers that can adapt to new threats,
as they appear - Technology progress making extreme defenses
practical, without sacrificing performance.
3Extreme Network Services
- Lightweight Flow Setup (LFS)
- one-way unicast flow with reserved bandwidth,
soft-state - no complex signaling, wire-speed setup, easy to
deploy - Network Access Service (NAS)
- provides controlled access to LFS
- registration/authentication of hosts, users
- resource usage data collection for monitoring,
accounting - Reserved Tree Service (RTS)
- configured, semi-private network infrastructure
for information service providers - reserved bandwidth, separate queues for traffic
isolation - paced upstream forwarding with source-based
queues for isolation and DOS protection
4Can We Afford Per Flow Processing?
- If it adds value, absolutely.
- Per Flow State
- at 50/MB (fast SRAM), 200B of flow state 1
cent - at 1/MB (DRAM), 10KB of flow state 1 cent
- if used for 2000 hours (avg. of lt5 over 5
years),costs 1 mcent per hour to cover cost of
both - Per Flow Processing
- to enable average of 10 instructions/byte on
OC-192, need 12.5 GIPS - 10 i/b enough for header processing
- 100 i/b enough for DES encryption
- at 200/GIPS, a 10 Mb/s flow will cost 125
mcents/hour - by 2010, expect to do 100 inst./byte for 12.5 mc/h
5Resource Reservation in Internet?
- Bandwidth reservation can provide dramatically
better performance for some applications. - Obstacles to resource reservation in Internet.
- distaste for signaling protocols
- perceived complexity of IntServRSVP
- requires end-to-end deployment
- little motivation for service providers
- How to get resource reservation in Internet?
- keep it simple
- focus on top priorities - one-way unicast flows
- avoid complex signaling - leverage hardware
routing mechanisms - make it useful when only partially deployed
- provide motivation for ISPs to deploy it
6Lightweight Flow Setup
- Implicit, one-way, unicast flow reservation.
- to setup flow, just send packets - no advance
signaling - specify flow rate(s) in packet header (using IP
option) - flow detected and route selection triggered as
needed - route for flow pinned until flow is released or
times out - prefer routes with ample unreserved bandwidth
- Stable rate reservation.
- allocated independently by routers along path
- congested links forward packets as datagrams
- reservation request honored as bandwidth released
by other flows - Transient rate reservation.
- routers allocate bandwidth fairly among competing
flows - direct feedback of bottleneck bandwidth to senders
7IP Option for LFS
op identifies flow setup operation - release
state - reserve stable rate - reserve transient
rate - status report
- status request - ignore
allocatedrate
requestedrate
- Stable rate fraction updated by routers on path.
- may trigger usage-based accounting
- Status request flags trigger status report.
- Alloc. rate stored at last hop router for status
gen. - F.P. rates with 4 bit mantissa, 4 bit exponent.
- specify rates from 64 Kb/s to 4 Gb/s , 6
granularity
8Implementing LFS - Input Side
- If flow table entry present, use stored next hop
- If no flow table entry, lookup route create
entry - store selected next hop in flow table entry
- At access router
- check privileges and record usage in access table
- if flow setup not enabled, forward packet as
datagram
9Implementing LFS - Output Side
- If flow table entry present, use it to find
queue, otherwise create an entry allocate
queue. - If stable rate specified, update entry.
- keep list of unsatisfied reservation requests to
process as bandwidth becomes available - If transient rate, update fair share and pacing
rate.
10Example Application
- Web site specifies stable rate in outgoing
streaming media packets - Use feedback to adjust sending rate if necessary.
- Note no action required by receivers.
11Regulating LFS Usage
- Regulate LFS use to ensure availability to users.
- user-specific privileges (limit rates, reserved
flows,...) - Record usage for monitoring, accounting.
- record reservation periods, rates, bytes
delivered - User privilege and usage information stored in
host/user database. - Regulation monitoring at network access points.
- for fixed access, just use physical interface
- for roaming access to ISP or corporate network
- registration protocol executed when host connects
to network - IP tunnel for data transfers between host and
access point - all data to/from host passes through that point
12Reserved Tree Service
- Reserved tree branches out to locations where
users are. - Downstream packets forwarded on-tree, share
reserved bandwidth pipes. - last hops use datagram forwarding
- Upstream packets paced and kept in source-based
queues.
13Extreme Router Architecture
- system mgmt.
- route table cfg.
- setup for non-LFS flows
Scalableswitch fabric
Lookup routeor state forreserved flows
- Distrib. queueing
- traffic isolation
- protect res. flows
14Improving Datagram Service
- Bandwidth hogging.
- single user can take more than fair share of link
bandwidth - other users packets delayed
- Synchronization of TCP flows.
- large queues and large delays
- Deficit round-robin service.
- Discard policy
- longest queue with hysteresis
- discard front
- Provides traffic isolation.
- each queue gets fair share
- small delays for nice flows
- Aggregate queues based on source prefix.
- avoid using up queues
- limits bandwidth use from single subnet
15Super-Scalable Packet Scheduling
- Scalability of QoS packet schedulers constrained
by need to maintain sorted list of queues. - Use approximate radix sorting, with compensation
- O(1). - timing wheels with increasing granularity and
range - approximate sorting produces inter-packet timing
errors - observe errors compensate when next packet
scheduled - Fast-forward bits used to skip to empty buckets.
- Scheduler puts no limit on number of queues.
16Distributed Queueing
- Distributed queueingregulates flow of traffic
through fabric. - ensures reserved flows receive assigned
bandwidth - allocates unreservedbandwidth fairly to datagram
traffic - Periodic broadcast of bandwidth assignments.
- per flow guarantees, without per flow info.
broadcast - switch fabric repackages data so each port
receives only relevant information - update period limited to use lt5 of switch
bandwidth - adds lt100 KB to each inputs buffer space in 1K
port router
17Prototype Extreme Router
18Summary
- Growing reliance on data networks creates higher
expectations - reliability, consistent
performance. - Design for worst-case - constructive paranoia.
- Technology progress making extreme defenses
practical, without sacrificing performance. - Extensible, rapidly reconfigurable routers
essential. - reconfigurable hardware, embedded processors
- Project will develop evaluate technologies for
extreme networking . - Things that havent worked.
- PIs lumbar region
- otherwise, too early to say