1. Applications : Network Monitoring Theodore Johnson AT&T Labs – Research johnsont@research.att.com Contributors: Chuck Cranor Vladislav Shkapenyuk Oliver Spatscheck

2. What is the challenge in network monitoring? Very high data rates. – Optical links : gigabit/sec and higher (to OC192), Millions of packets/sec. Complex queries. – Extract dynamic substreams (flows, TCP sessions, etc.). Many sequence numbers, types of sequencing – Simulate network protocols (IP defragmentation, etc.). Multiple data sources. – SNMP, Netflow, BGP, packet sniffers, router tables, etc. – Many layered protocols: multimedia, VPN, etc. – Dozens to hundreds of heterogeneous deployments. Every bad thing happens (most interesting part). Overcome a prejudice that database technology is too slow and rigid for network monitoring. Years-long lead time for core deployments.

3. Gigascope Gigascope is a lightweight and flexible stream database specialized for network monitoring. Current deployments: –Research’s link to Internet (100 Mbit/sec). –Large on-line game site (Gigabit Ethernet). Ongoing deployment: –AT&T backbone internet routers (OC48). Developed in close cooperation with network analysts.

4. PROTOCOL GAMEPROTOCOL (UDP) { ullong gp_header gp_header (snap_len 134); bool gp_is_ack_request gp_is_ack_request (snap_len 134); bool gp_is_ack_response gp_is_ack_response (snap_len 134); uint gp_ack_id gp_ack_id (snap_len 134); uint gp_sequence_number gp_sequence_number (snap_len 134); } select timestamp, sourceIP, destIP, source_port, dest_port, len, total_length, gp_header from GAMEPROTOCOL where sample_hash[50, sourceIP, destIP] and protocol=17 and offset=0 select tb, lowIP, highIP, protocol, SUM(len), COUNT(*) from IPV4 where sample_rand[75] group by protocol, time/60 as tb, UMIN(sourceIP, destIP) as lowIP, UMAX(sourceIP, destIP) as highIP

5. Speed isn’t important, packet loss is

6. select tb*60, MIN(timestamp), MAX(timestamp), destIP, dest_port, hostheader, count(*) from O.TCP where ipversion=4 and offset=0 and protocol=6 and str_match_start[TCP_data,'GET'] group by time/60 as tb, str_extract_regex(TCP_data,'[Hh][Oo][Ss][Tt]:[0-9A-Za-z\\.: ]*') as hostheader, destIP, dest_port select tb, localip, peerid, asid, count(*), sum(len) from I.DataProtocol where ipversion=4 and (dest_port=25 or source_port=25) group by time/60 as tb, destIP as localip, getlpmid(sourceIP, 'peerid.tbl') as peerid, getlpmid(sourceIP, 'asnid.tbl') as asid

7. Second App: Intrusion Detection Current technology (e.g. Snort): –Low-speed only, cumbersome specification language. –Can’t trigger alarms/recording from aggregates (e.g. recent Ping attack on root DNS servers). Needed: fast trigger processing. –Post alarm to network manager (not challenging). –Run query Q on the next 500 packets (challenging). –Run query Q on the previous 500 to next 500 packets (very challenging). Potential solutions: –Push trigger processing to lowest query processing levels. –Snap backwards on input buffers to run in-the-past queries.

8. Summary Very high data rates. –Fast and lightweight architecture. –Early data reduction is critical. Complex queries. –Language constructs to capture substreams (e.g., netflow, TCP session, gaming session). –Extensive support for user-defined functions (getlpmid). –Support user-defined operators (e.g., IP defragmentation) Currently working on a view mechanism Multiple data sources –The current naming scheme is quite limited. Collaboration across disciplines –Database, Networking, Systems, Software, Algorithms Accept that you can’t do everything. –But you can do most of the tedious work.