Presentation is loading. Please wait.

Presentation is loading. Please wait.

The Cutting EDGE of Router Configuration D. Caldwell, A. Gilbert, J. Gottlieb, A. Greenberg, G. Hjalmtysson, and J. Rexford AT&T Labs—Research; Florham.

Similar presentations


Presentation on theme: "The Cutting EDGE of Router Configuration D. Caldwell, A. Gilbert, J. Gottlieb, A. Greenberg, G. Hjalmtysson, and J. Rexford AT&T Labs—Research; Florham."— Presentation transcript:

1 The Cutting EDGE of Router Configuration D. Caldwell, A. Gilbert, J. Gottlieb, A. Greenberg, G. Hjalmtysson, and J. Rexford AT&T Labs—Research; Florham Park, NJ Manual configuration: Type first, and ask questions later

2 Manual Configuration Dangerous –Typo in routing policy: black hole –Wrong OSPF area: no traffic on link –Missing packet filter: DoS vulnerability Expensive –Delays in deploying equipment –Hiring & training skilled engineers –Lock-in to the router vendor Disruptive –Half of network outages (Yankee Group) –BGP routing anomalies (SIGCOMM’02) –Failures of Internet services (USITS’03)

3 Why is the Situation So Bad? Networking field: emphasis on speed & features –Not manageability and simplicity –Constant change, without revisiting the design –Oodles of complex protocols and tunable parameters Router vendors: lack of abstraction –Assembly language commands –Element-level configuration –Low-level mechanisms, not intent Network administrators: melting under complexity –Learning on the job –Struggling just to keep up –The path to automation is overwhelming

4 What’s a Networking Researcher to Do? Myth: Only large service providers care. –Large enterprise networks care, big time –Other network’s problems affect the user Myth: No research lives here. –The state-of-the-art is really pitiful (please help!) –Config languages, protocol modeling, data mining, … Myth: I can’t play without special access. –Router testbeds, protocol modeling, simplicity, etc. –Get data from your campus/regional administrator Myth: Campus/regional networks are too easy. –Maybe it is harder than we think… –If easy, please scale this simplicity to larger networks!

5 Our Goal: Automated Configuration How to transition an existing network? How to get value as you move from here to there? Our approach: detailed analysis of configuration data DB interface Serial10/1/0/12:0 description CBB Customer ip address 12.7.35.1 255.255.255.252 ip access-group 666 in ! router bgp 7018 neighbor 12.7.35.2 remote-as 18585 neighbor 12.7.35.2 route-map CUST-FACE in neighbor 12.7.35.2 route-map FULL-ROUTES out neighbor 12.7.35.2 distribute-list 13 in neighbor 12.7.35.2 soft-reconfiguration-inbound ! What is your AS number? What export policy do you want? Do you want a dynamic default? What are your address blocks? Do you need to receive communities? query template RULESRULES configlet Technical Questions (TQ) interface description ip address ip access-group in ! router bgp 7018 neighbor remote-as neighbor route-map CUST-FACE in neighbor route-map out neighbor distribute-list in neighbor soft-reconfiguration-inbound [neighbor send-community] ! router

6 Lowering Barrier-to-Automation for Enterprises Large enterprise networks –Large stand-alone data networks –Retail, financial, health, business, etc. –Heterogeneity due to mergers and acquisitions –Stringent reliability and performance demands TodayEDGE Inventory database The network is the database Extracted from the configuration files Existing network Mistakes and inconsistencies Discord reports from analyzing config files Local config rules Poor (or no) documentation, policy violations Reverse-engineered by data mining on the configuration files

7 Enablement and Debugging of Growing Enterprises BOTTOM-UP APPROACH 1.Inventory database  Extract summary information  Bootstrap the database 2.Fixing config mistakes  Report errors & warnings  Allows immediate fixes 3.Codifying local policies  Reverse-engineer policies  Aids in fixing inconsistencies 4.Automated configuration  Explicitly enforce the rules  Avoid future config mistakes Web reports Discords Low level standard form (tables) Abstract network database polled queries Router configuration Automation fix Runs on many thousands of configs a day.

8 Bootstrapping the Inventory Database Extracting summaries –Routers & interfaces –Routing protocols –IP address usage –Configuration commands Network visualization –Routers and links –Attributes (capacity, routing parameters) –Layout (spring model) –Graph compression –Geographical info Dual hub-and-spoke

9 Pinpointing Configuration Mistakes Referential integrity –Multiple interfaces with same IP address –Items used but not defined, and vice versa Packet/route filters, route-maps, QoS policy,… Routing protocol (e.g., OSPF) –Local integrity checks (warnings) Network statement matches no interfaces Interface enabled in OSPF has no cost metric Interface not enabled in OSPF has a cost metric –Correct adjacencies (errors) OSPF interface is not part of a complete link Remote end of link does not participate in OSPF Interfaces on same link have different OSPF areas

10 Example: Joining Parts of OSPF Config Together hostname MyRouter ! interface POS7/0 ip address 12.123.36.74 255.255.255.252 ip ospf cost 50 ! router ospf 2 network 12.123.36.0 255.255.255.0 area 9 passive-interface Serial2/1/0/3.1 ! Remote end is in 12.123.36.72/30 Interface participates in OSPF

11 Example: Remote End in Different OSPF Area interface OSPF network OSPF passive interface link OSPF interface Extracted tables active OSPF interface OSPF link with area mismatch Intermediate tables Presentation query result Simple SQL queries

12 Using Data Mining to Infer Local Policies Pattern matching across routers –Equivalence: same configuration, same names –Synonym: same configuration, different names –Homonym: same name, different configuration Infer rules and report exceptions –“Finger daemon disabled” –“Edge interfaces have inbound packet filters” –“OSPF costs of 100 to hub1 and 200 to hub2” –“IPX interfaces have SAP encapsulation” Identifying “use cases” –Initial base configuration of each router –“Diffs” over time and across related routers –Common cases (e.g., adding link, new BGP session)

13 Conclusion Moving beyond manual configuration –Manual configuration is bad Error-prone, expensive, and disruptive –Migrating to automation is hard No inventory DB, buggy network, and poor docs –EDGE supports migration to automation Bootstrap DB, report discords, and infer policies More innovation is needed –Better router configuration languages –Protocol modeling and best common practices –Data mining on existing data networks –Autoconfigurable protocols and mechanisms


Download ppt "The Cutting EDGE of Router Configuration D. Caldwell, A. Gilbert, J. Gottlieb, A. Greenberg, G. Hjalmtysson, and J. Rexford AT&T Labs—Research; Florham."

Similar presentations


Ads by Google