Unified Data Platform, by Pauline Yeung of Cisco Systems

Data Engineer
Cisco Umbrella
yeungp@cisco.com
Unified Data Platform
Pauline Yeung
ClickHouse Meetup
Dec 3, 2019

© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Confidential© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Agenda
1
2
3
4
5
Problems
Use Case: Authlog
Use Case: Whois Records
Use Case: Network Tunnels
Next

© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Confidential© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
• Data Engineer at Cisco Umbrella,
Investigate team
• M. S. Computer Engineering, Santa
Clara U
• B. S. Electrical Engineering, U of
Calgary
$ whois Pauline

© 2019 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Problems

Investigate: the most powerful way to uncover threats
Console
API
SIEM, TIP
Key points
Intelligence about domains, IPs,
and malware across the internet
Live graph of DNS requests and
other contextual data
Correlated against statistical models
Discover and predict malicious
domains and IPs
Enrich security data with global intelligence
domains, IPs, ASNs, file hashes

Investigate Backend
Whois ASN
IntelDB
Umbrella
Investigate
passive
DNS
We want
• Easy, fast, and flexible platform for ad hoc
analysis of authlog, which are stored in
passive DNS.
• Increase throughput and reduce costs for
Whois database.
• Fast access to ASN and enrich security
data.
• One datastore for multiple use cases.
Share datastore with other product teams.

DNS Authoritative Log
(authlog)

Passive DNS

Domain to IP Relationships
11 JAN 2019
domain2.com
10 JAN 2019
domain1.com
12 JAN 2019
domain3.com
12.4.0.4/32

AuthLog Examples
owner name datacenter name_server
1 tabiturient.ru. tabiturient.ru. lon ns4.nic.ru.
2 thefacebook.com. certs.thefacebook.com. sea b.ns.facebook.com.
3 333az.net. nbb4yd.333az.net. yyz ns2-09.azure-dns.net.
4 dotnetwork2.co.za. d1000253-146.dotnetwork2.co.za. jnb ns3.dotnetworkdns.co.za.
name_server_ip rr ttl type timestamp
1 194.226.96.8 195.24.68.22 3600 A 2019-12-02 10:56:47
2 2a03:2880:ffff:c:face:b00c:0:35 2620:10d:c0a1:10:0:0:0:35 600 AAAA 2019-12-02 12:40:46
3 2620:1ec:8ec::9 ns1-07.azure-dns.com. 20 NS 2019-12-02 10:34:15
4 41.223.172.166 mail.d1000253-146.dotnetwork2.co.za. 3600 MX 2019-11-30 03:05:17

AuthLog Data Pipeline
authlog
producer
authlog
clickhouse
ingester
S3
archiver
resolvers
32 data centers
3 days
authlog
HBase
ingester
Investigate
UI
API Server
6 nodes, 1 replica
r4.4xlarge
16 vCPU, 122 GB, 2 TB disk
32 nodes
i3 2xlarge
8 vCPU, 61 GB
authlog
parquet
passive
DNS
120b requests/day
4b authlog/day

One Set of Questions
• What’s the increase in disk usage for passive DNS per day?
• What type of traffic contribute the most to the increase of disk usage?

AuthLog for Past 3 Days
CREATE TABLE IF NOT EXISTS authlog.alog_local (
owner String,
name String,
datacenter String,
name_server String,
name_server_ip String,
rr String,
ttl Int32,
type String,
timestamp DateTime)
ENGINE = MergeTree()
PARTITION BY toYYYYMMDD(timestamp)
ORDER BY (name, timestamp)
TTL timestamp + toIntervalDay(3)
SETTINGS index_granularity = 8192
48 golang workers write to 6 shards
ingest 1.2m rows per second
ClickHouse kafka engine does not support avro

AuthLog for Past 3 Days
CREATE TABLE IF NOT EXISTS authlog.alog(
owner String,
name String,
datacenter String,
name_server String,
name_server_ip String,
rr String,
ttl Int32,
type String,
timestamp DateTime)
ENGINE = Distributed(log_cluster, authlog, alog_local, cityHash64(name))
access all shards
4b rows per day
200 GB for 3 days

Payload for Domains in 2 Consecutive Days
CREATE TABLE IF NOT EXISTS authlog.a
ORDER BY name
AS SELECT name, type, sum(length(name) + length(rr)) AS payload
FROM authlog.alog
WHERE timestamp >= toDateTime('2019-11-28 16:00:00') and timestamp <= toDateTime('2019-
11-28 19:59:59')
GROUP BY name, type
CREATE TABLE IF NOT EXISTS authlog.b
ORDER BY name
AS SELECT name, type, sum(length(name) + length(rr)) AS payload
FROM authlog.alog
WHERE timestamp >= toDateTime('2019-11-29 16:00:00') and timestamp <= toDateTime('2019-
11-29 19:59:59')
GROUP BY name, type
took 2 minutes, 148m rows, 3.0 GB
4 hours, ¼ daily authlog

Payload for Domains Only in Day 2
CREATE TABLE IF NOT EXISTS authlog.b_only
ORDER BY (name)
AS SELECT
b.name as name,
b.type as type,
sum(b.payload) as payload
FROM a
RIGHT JOIN b ON a.name = b.name
WHERE a.name like ''
GROUP BY
b.name,
b.type
ba
users.xml
max_memory_usage = 96GB
max_bytes_before_external_group_by = 48GB
max_bytes_before_external_sort = 48GB
4 hours
Nov 28
4 hours
Nov 29

Second Level Domains with Highest Payload
SELECT
arrayStringConcat([splitByString('.', name)[-3], '.', splitByString('.', name)[-2], '.’]) AS pname,
sum(payload) / 1024 / 1024 AS payload_MB
FROM b_only
GROUP BY pname
ORDER BY payload_MB DESC
LIMIT 100
pname payload_MB
cloudfront.net. 878.2994289398193
office.com. 719.9946641921997
clienttons.com. 608.300389289856
cnr.io. 473.1693649291992
akamaihd.net. 395.0745334625244
cedexis-radar.net. 364.29007720947266
footprintdns.com. 265.04366874694824
gstatic.com. 250.41933727264404
squarespace.com. 151.24806880950928
forter.com. 151.08679962158203
wacodenver-com.mail.protection.outlook.com.
to
outlook.com.
took 6 seconds

Resource Type with Highest Payload
SELECT
type,
sum(payload) / 1024 / 1024 / 1024 AS payload_GB
FROM b_only
GROUP BY type
ORDER BY payload_GB desc
type payload_GB
A 4.34150860644877
CNAME 3.289442714303732
RRSIG 1.5495505537837744
DNSKEY 0.6393735473975539
TXT 0.5898861000314355
SELECT sum(payload) / 1024 / 1024 / 1024 as payload_GB FROM b_only
payload_GB
11.225993978790939
took 483 msec
took 131 msec

DataBricks Spark
• 1 day authlog, ~4b rows
• Process 4x authlog
• Took 9 minutes
pname payload_GB
office.com. 3.812719924375415
cloudfront.net. 3.3973056096583605
cnr.io. 2.608651074580848
clienttons.com. 2.2882667966187
cedexis-radar.net. 1.9318219376727939
type payload_GB
A 16.849326515570283
CNAME 14.351725150831044
RRSIG 3.1499833753332496
TXT 3.047112719155848
NS 1.328178352676332
payload_GB
41.024286944419146
took 5 seconds
took 1.3
seconds
took 490 msec

Whois Record Data

WHOIS Record Data
§ Who registered the domain
§ Contact information used
§ When/where registered
§ Expiration date
§ Historical data
§ Correlations with other
malicious domains
See relationships between
attackers’ infrastructure

$ whois facebook.com
:
Domain Name: FACEBOOK.COM
Registry Domain ID: 2320948_DOMAIN_COM-VRSN
Registrar WHOIS Server: whois.registrarsafe.com
Registrar URL: https://www.registrarsafe.com
Updated Date: 2019-10-17T18:52:06Z
Creation Date: 1997-03-29T05:00:00Z
Registrar Registration Expiration Date: 2028-03-30T04:00:00Z
Registrar: RegistrarSafe, LLC
Registrar IANA ID: 3237
Registrar Abuse Contact Email: abusecomplaints@registrarsafe.com
Registrar Abuse Contact Phone: +1.6503087004
Domain Status: clientDeleteProhibited https://www.icann.org/epp#clientDeleteProhibited
Domain Status: clientTransferProhibited https://www.icann.org/epp#clientTransferProhibited
Domain Status: serverDeleteProhibited https://www.icann.org/epp#serverDeleteProhibited
Domain Status: serverTransferProhibited https://www.icann.org/epp#serverTransferProhibited
Domain Status: clientUpdateProhibited https://www.icann.org/epp#clientUpdateProhibited
Domain Status: serverUpdateProhibited https://www.icann.org/epp#serverUpdateProhibited
API request: domainName
API response: WhoisRecord_rawText

Registry Registrant ID:
Registrant Name: Domain Admin
Registrant Organization: Facebook, Inc.
Registrant Street: 1601 Willow Rd
Registrant City: Menlo Park
Registrant State/Province: CA
Registrant Postal Code: 94025
Registrant Country: US
Registrant Phone: +1.6505434800
Registrant Phone Ext:
Registrant Fax: +1.6505434800
Registrant Fax Ext:
Registrant Email: domain@fb.com
Registry Admin ID:
Admin Name: Domain Admin
Admin Organization: Facebook, Inc.
Admin Street: 1601 Willow Rd
Admin City: Menlo Park
Admin State/Province: CA
Admin Postal Code: 94025
Admin Country: US
Admin Phone: +1.6505434800
Admin Phone Ext:
Admin Fax: +1.6505434800
Admin Fax Ext:
Admin Email: domain@fb.com
Tech Name: Domain Admin
Tech Organization: Facebook, Inc.
Tech Street: 1601 Willow Rd
Tech City: Menlo Park
Tech State/Province: CA
Tech Postal Code: 94025
Tech Country: US
Tech Phone: +1.6505434800
Tech Phone Ext:
Tech Fax: +1.6505434800
Tech Fax Ext:
Tech Email: domain@fb.com
Name Server: A.NS.FACEBOOK.COM
Name Server: B.NS.FACEBOOK.COM
DNSSEC: unsigned
:
API request: contactEmail
API response: list of domainName
API request: list of nameServer
API response: list of domainName

Questions
• Continue to maintain 29 nodes cluster, running Ubuntu Trusty, ElasticSearch 1.6?
• Migrate to AWS ElasticSearch 7.1?
• Migrate to AWS Aurora Postgres?
• Migrate to ClickHouse?
don’t need full text search
does not support shard for scaling
update is slow
insert efficient for bulk insert only
no secondary index
no

Whois Data Pipeline
whois
ingester
whois
indexer
ClickHouse
ElasticSearch
Investigate
UI
API Server
29 nodes, 2 replicas
2 indices
12 TB
6 nodes, 2 replicas
3 tables, 1 materialized view
2 TB
download
download
1 index
2 tables, 1 MV

Domain Table
CREATE TABLE IF NOT EXISTS whois.domain_local(
domainName String,
contactEmail String,
RegistryData_rawText String,
WhoisRecord_rawText String)
ENGINE = ReplicatedReplacingMergeTree('/clickhouse/tables/{shard}/whois.domain_local',
'{replica}’)
PRIMARY KEY (domainName)
ORDER BY (domainName)
CREATE TABLE IF NOT EXISTS whois.domain(
domainName String,
RegistryData_rawText String,
WhoisRecord_rawText String)
ENGINE = Distributed(whois_cluster, whois, domain_local, cityHash64(domainName))
48 golang writers write to 6 shards
cityhash.Hash64([]byte(domainName)) % numHosts
for merging,
ClickHouse selects the last inserted row,
or if version column exists,
selects the row with the max value in the version column

Email Table
CREATE MATERIALIZED VIEW IF NOT EXISTS whois.email_mv_local(
domainName String)
ENGINE = AggregatingMergeTree
ORDER BY (contactEmail, domainName)
POPULATE
AS SELECT contactEmail, domainName
FROM db.domain_local
WHERE contactEmail != ''
GROUP BY contactEmail, domainName
CREATE TABLE IF NOT EXISTS whois.email_mv(
domainName String)
ENGINE = Distributed(whois_cluster, db, email_mv_local, cityHash64(contactEmail))
domain table is 150 GB
email table is 3 GB

NS Table
CREATE TABLE IF NOT EXISTS whois.ns_local(
nameServer String,
domainName String)
ENGINE = ReplicatedReplacingMergeTree('/clickhouse/tables/{shard}/whois.ns_local', '{replica}')
PRIMARY KEY (nameServer, domainName)
ORDER BY (nameServer, domainName)
CREATE TABLE IF NOT EXISTS whois.ns(
nameServer String,
domainName String)
ENGINE = Distributed(whois_cluster, whois, ns_local, cityHash64(nameServer))
name server table is 4 GB

Whois Queries
SELECT WhoisRecord_rawText FROM domain_local FINAL WHERE domainName = 'facebook.com’
SELECT WhoisRecord_rawText FROM domain FINAL WHERE domainName = 'facebook.com’
SELECT domainName FROM email_mv WHERE contactEmail = 'domain@fb.com’
domainName
buyfbfansnow.com
facebook-hardware.com
instagram-engineering.net
pokerface-book.com
what3app.com
SELECT * FROM ns WHERE nameServer LIKE ‘%.facebook.com’
nameServer domainName
ns1.facebook.com djgabeholm.com
ns2.facebook.com shellpriv.com
ns3.facebook.com arabfashioncompany.com
a.ns.facebook.com zuckerberg.com
b.ns.facebook.com zuckerberg.net
took 9 msec
postgres 7 msec
took 21 msec
data selected fully merged, slower
took 16 msec, 2549 rows
took 36 msec, 7075 rows
Will add TLD column, e.g. facebook.com
ORDER By TLD, nameserver, domainName
expect query < 10msec

Network Tunnels

Network Tunnels
• Network tunnels deliver the
branch office traffic to the
Cisco’s cloud edge where
Umbrella runs a number of
security functions.
• Firewall, web security, DNS
security.
Provisioned per
organization

Network Tunnels
DNS
security
web security
S3
Network
Tunnels UI
API Server
Tunnel
Visibility
Sensors
states
events
ClickHouse
downloadnetwork tunnels
3 nodes, 1 replica
firewall
script
CSV

Event Table
CREATE TABLE IF NOT EXISTS tunnel_viz.event_local(
OrgID UInt32,
TunnelID UInt32,
EventTime DateTime,
EventID String,
EventType String,
PeerID String,
PeerIP IPv4,
PeerPort UInt16,
Code LowCardinality(String),
Reason String)
PARTITION BY toYYYYMM(EventTime)
PRIMARY KEY (OrgID, TunnelID, EventTime)
ORDER BY (OrgID, TunnelID, EventTime)
TTL EventTime + toIntervalDay(120)
dictionary encoding, 10 unique codes
event table, 3.3m rows, 260 MB

Event Table
CREATE TABLE IF NOT EXISTS tunnel_viz.event(
OrgID UInt32,
TunnelID UInt32,
EventTime DateTime,
EventID String,
EventType String,
PeerID String,
PeerIP IPv4,
PeerPort UInt16,
Code LowCardinality(String),
Reason String)
ENGINE = Distributed(cdfw_cluster, tunnel_viz, event_local, murmurHash3_32(OrgID))"
fairly even distribution for integer

Event Queries
SELECT Code, count() as c FROM event GROUP BY Code ORDER BY c DESC
┌─Code─────────────────┬──────c─┐
│ PROPOSAL_MISMATCH_CHILD │ 2164575 │
│ PEER_AUTH_FAILED │ 871937 │
│ RETRANSMIT_SEND │ 243559 │
│ RETRANSMIT_SEND_TIMEOUT │ 42829 │
│ UNIQUE_REPLACE │ 26540 │
│ PARSE_ERROR_BODY │ 2717 │
│ CERT_REVOKED │ 1004 │
│ LOCAL_AUTH_FAILED │ 186 │
│ TS_MISMATCH │ 9 │
│ VIP_FAILURE │ 1 │
└─────────────────────┴────────┘
took 23 msec, 10 unique codes

What we learned
• Cheap
• Fast
• Flexible
• Good compression
• Cluster isolation for multiple
stores
• Ad hoc analysis for authlog
using 200 GB storage
• Lower cost and acceptable
performance for whois
database.
• Share hardware for different
type of datastores.

ClickHouse Wish List
• Support avro in kafka engine.
• Balance cluster and copy data after failed or added node.

Questions?

Backup – Other Use Cases

Threat Library

Threat Library Data Pipeline
Web UI
API Server
S3
DNS query log
threat/attack
DNS query log
blocked domains
run job
blocked domains
Kubernetes
Cluster
threat/attack feed 1..n blocked domains
threat/attack
ClickHouse
Airflow
1 pod, 256 GB disk

Blocked Domains
CREATE TABLE IF NOT EXISTS attribution.blocked(
datetime DateTime,
domain String,
threat LowCardinality(String),
attack LowCardinality(String),
count UInt32)
ENGINE = ReplacingMergeTree()
PARTITION BY toYYYYMMDD(datetime)
PRIMARY KEY (datetime, domain, threat, attack)
ORDER BY (datetime, domain, threat, attack)
TTL datetime + toIntervalDay(30)
dictionary encoding, 37 threats, 118 attacks

ASN

Autonomous Systems
• IP to ASN
ASN Attribution

Domain à IP à ASN relationships
AS 701AS 3462 AS 12271
1.168.6.17
domain1.com
100.2.65.157 104.162.93.136

ASN Data Pipeline
data
importer
ClickHouse
Aurora
Postgres
Web UI
API Server
1 write 1 read
download
download
CSV
multiple
tables

Postgres
WITH b as (
SELECT asn, cidr FROM delta_bgp_routes
WHERE (cidr >>= CAST('104.244.42.193' AS ip4r))
AND (period && DATERANGE(CURRENT_DATE - integer '2', CURRENT_DATE,'[]')))
SELECT a.asn, b.cidr, a.description, a.creation_date AS creationDate, a.ir
FROM delta_autonomous_systems AS a, b
WHERE (a.asn = b.asn)
AND (period && DATERANGE(CURRENT_DATE - integer '2', CURRENT_DATE,'[]’));
asn | cidr | description | creationdate | ir
-------+-----------------+----------------------------------+--------------+----
13414 | 104.244.42.0/24 | TWITTER - Twitter Inc., US 86400 | 2010-07-09 | 3
Took 52 ms

/etc/clickhouse-server/asn_dictionary.xml<yandex>
<dictionary>
<name>asn_dict</name>
<layout>
<ip_trie />
</layout>
<structure>
<key>
<attribute>
<name>prefix</name>
<type>String</type>
</attribute>
</key>
<attribute>
<name>asn</name>
<type>UInt32</type>
<null_value />
</attribute>
<attribute>
<name>country</name>
<type>String</type>
<null_value>??</null_value>
</attribute>
<attribute>
<name>created_at</name>
<type>DateTime</type>
<null_value />
</attribute>
<attribute>
<name>registry</name>
<type>UInt32</type>
<null_value />
</attribute>
<attribute>
<name>description</name>
<type>String</type>
<null_value />
</attribute>
<attribute>
<name>datastr</name>
<type>String</type>
<null_value />
</attribute>
</structure>
<source>
<file>
<path>/opt/dictionaries/asnprefixes.csv</path>
<format>CSVWithNames</format>
</file>
</source>
<lifetime>300</lifetime>
</dictionary>
</yandex>

ASN Dictionary
• 1.5 minutes in Spark job to download and generate CSV
SELECT dictGetString('asn_dict', 'datastr', tuple(IPv4StringToNum('143.202.186.23’)))
143.202.186.0/24 264076 BR 1445817600 4 BREM TECHNOLOGY LTDA - ME, BR
SELECT dictGetString('asn_dict', 'datastr', tuple(IPv6StringToNum('2800:5f0:800::1’)))
40.0.0.0/19 4249 US 0789782400 3 LILLY-AS - Eli Lilly and Company, US
took 2 ms
took 2 ms

Unified Data Platform, by Pauline Yeung of Cisco Systems

Unified Data Platform, by Pauline Yeung of Cisco Systems

Recomendados

Recomendados

Mais conteúdo relacionado

Mais procurados

Mais procurados (20)

Semelhante a Unified Data Platform, by Pauline Yeung of Cisco Systems

Semelhante a Unified Data Platform, by Pauline Yeung of Cisco Systems (20)

Mais de Altinity Ltd

Mais de Altinity Ltd (20)

Último

Último (20)

Unified Data Platform, by Pauline Yeung of Cisco Systems