Graphistry founder Leo Meyerovich walks through visual graph analytics using Graphistry GPU acceleration & graph-app-kit for the March session of Graph Gurus. The focus is on GPU visual graph analytics to understand your data’s relationships.. and without the coding.
Topics include:
* Why visual graph analysis is becoming more popular
* No-code: Using Graphistry for analyzing social media manipulation by some of the biggest cryptocurrency communities and nation-state sponsors
* Low-code: The new graph-app-kit project (Graphistry, Streamlit, Jupyter, Nvidia RAPIDS.ai), including a live TigerGraph demo, ecosystem tour, and interactive discussion
* GPU acceleration: Processing event logs at 100GB/s+ for graph edge processing, using just Python/UI via Graphistry+RAPIDS for GPU computing
1. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y
Low-coding & no-coding GPU-accelerated visual graph analytics
Graph Gurus, 2021
Leo Meyerovich, CEO
@LMeyerov
2. G R A P H I S T R Y info@graphistry.com
1. Graph era of visual analytics: Graphistry
2. No-code: Data -> relationship insights in 10s
3. Scale with GPUs: 412GB/s on a single GPU node
4. Low-code apps with graph-app-kit
Today: 100X graph tech
3
3. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
Tech
Security, fraud, user analytics, genomics, …
Graph, viz, GPUs, automation
Users
100X VISUAL INVESTIGATIONS
4. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
5
(launch graph-app-kit)
GRAPHISTRY
5. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
6
Graph era of visual analytics
knowledge graph
neural search
event & log analysis
correlation
high-cardinality
& high-dimensionality
process mining
people analytics
embedding spaces
graph neural nets
Identity graph
dimensionality reduction
workflow automation
graph API
virtual graph
hypergraphs
GPUs
informatics
6. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
7
Tables getting longer: Events
Time
More tables: Every co is a tech co
Tables getting wider: Metadata, AI scores, …
Name, Email, IP, ..
IT analyst:
Digital assets & users: Dependencies, activities, ..
Fintech analyst:
Companies, people: Dependencies, trades, …
Product analyst
Users, offerings: Journeys, preferences, …
Security analyst:
User, asset, incident: Timelines and scopes, …
Fraud analyst:
Account, payment: Timeline, pattern, outlier, …
What is my graph of …
Data becoming graph-y
journey
correlation
7. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
8
IT analyst:
Digital assets & users: Dependencies, activities, ..
Fintech analyst:
Companies, people: Dependencies, trades, …
Product analyst
Users, offerings: Journeys, preferences, …
Security analyst:
User, asset, incident: Timelines and scopes, …
Fraud analyst:
Account, payment: Timeline, pattern, outlier, …
What is my graph of …
Tabular views hide
the relationship insights
??
8. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
9
Visual graph tools
help us work with the
relationships in our data
Understand Explore Collaborate Automate
Visual graph tools
increase data project ROI
& eliminate project risks
Deliver faster Build better Multiply reach
9. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
Data Scientist
Notebooks
Dev API for
Embedding
Analyst
Tool Suite
3. Automate
Templatize, link, & embed
1. Connect DBs & APIs
as a unified virtual graph
2. Visual analytics
• 100X via GPUs
• Correlate w/ graph
• Time, histograms, …
Graphistry: Visually answer relationship questions across your data sources
Graphistry Hub (Cloud) Cloud Marketplace On-Prem
10. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
11
Demo: Graph investigation of
social media manipulation
in the XRP cryptocurrency community
geoff@socialforensics.com
11. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
12
geoff@socialforensics.com
”[OK…] as long as it’s really cool”
12. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
13
“I had to wait 2 minutes in
[Tool X] for every little thing,
and now I can see everything,
so I use Graphistry instead”
New: No-code csv graph
13. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
14
Follow for XRP Especially in Japan
… that
are fake
And if inauthentic
XRP
COMMUNITY
SEGMENTATION
… Likely of interest to SBI Holdings,
a Japanese megacorp that then went big on XRP …
14. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
15
Big with Arabic
speakers
Who are
inauthentic
15. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
16
1 year later…
16. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
17
70% of the XRP Arabic accounts were removed due to Twitter’s breach & sweep
17. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
18
… Ripple CTO David Schwartz
still has not changed his icon to a robot ...
18. G R A P H I S T R Y info@graphistry.com
Graph is powerful
Market & user analysts
✓ More ROI: Find influencers, communities, …
✓ Less waste: Don’t spend on fraudsters, spammers, wrong communities…
Fraud & abuse analysts
✓ Detect & investigate fake accounts & activities
✓ Detect behaviors like abuse, misinformation, …
Graph tech: No code & GPUs
✓ See bigger picture, “I was waiting 2 minutes in <TOOL X> and…”
✓ Enable domain expert to not get bogged down
19
geoff@socialforensics.com
19. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
20
data scientist, data engineer, dev, …
scientist, analyst, investigator, …
sales, marketing, trading, ...
Graph is useful for coders + non-coders
20. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
21
Embedding APIs
No-code
Low-code create dashboards + automations
supercharge existing tools
Built to multiply the reach & ROI of graph projects
21. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
22
Graph low-coding on GPUs
Demo: Windows logs
Graphistry is building the first
RAPIDS-native visual analytics platform
22. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
23
your
app
…
graphistry.hypergraph(
cudf.read_csv(‘events.csv’),
entity_types=[‘user’, ‘user_ip’, ‘asset_id’]
)['graph'].plot()
Give an inch, get a mile
Interactive
Filters
Time bar
Sharing
Search
…
23. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
24
your
app
…
graphistry.hypergraph(
cudf.read_csv(‘events.csv’),
entity_types=[‘user’, ‘user_ip’, ‘asset_id’]
)['graph'].plot()
# cudf: optional GPU Python lib
# … that devours heavier files
Give an inch, get a mile
Interactive
Filters
Time bar
Sharing
Search
…
24. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
25
Launch with everything
ready for 1+ GPUs
25. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
26
BENCHMARK
Windows event logs
280M rows
320 GB raw text
44 GB mostly raw parquet
Time column: 1.5GB
26. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
27
import cudf
gdf = cudf.read_parquet(‘my_logs.parquet’)
print(“latest time:”, gdf[‘_time’].max())
### Time: 0.6s +/ 0.1s
### Rate: 1.9GB/s
### Rate: 369M rows/s
### PCI is 8GB/s…
Graph low-coding on a GPU: Hello – max time
### already on GPU
gdf[‘_time’].max()
### Time: 3.6ms +/- 720us
### Rate: 412GB/s
### Rate: 78B rows/s
280M timestamps (1.5GB)
27. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
28
Graph low-coding on a GPU:
Compute login graph
# Extract table: logins[[ ‘EventCode’, ‘src_NetworkAddress’, ‘dst_ComputerName’]]
gdf['EventCode’] = gdf['_raw'].str.extract(r'EventCode=(d+)').astype('int32’)
logins_gdf = gdf[ gdf['EventCode'] == 4624 ]
logins_gdf['src_NetworkAddress'] = logins['_raw'].str.extract(r'Source Network Address:t(.*)n’)
logins_gdf['dst_ComputerName’] = logins['_raw'].str.extract(r'ComputerName=(.*)n’)
### Rate: 25 GB/s
### Rate: 25M rows/s
28. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
29
Graph low-coding
on 8 GPUs
dgdf = dask_cudf.read_parquet(‘logs.parquet’)
is_login = “EventCode=4624”
dgdf[‘_raw’].str.match(is_login).count().compute()
### Time: 37s
### Rate: 1GB/s
### Rate: 8M rows/s
### ~100X below our target
29. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
30
Graph low-coding on 8 GPUs:
Compute login graph
… Sign up at hub.graphistry.com to find out!
SSD 1
SSD 2
…
SSD 20
GPU 1 (40GB)
GPU 2 (40GB)
…
GPU 8 (40GB)
120 GB/s
1 STORAGE NODE 1 GPU NODE
300 GB/s
4 NUMA nodes
(the hard part)
30. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
31
31. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
32
Introducing graph-app-kit
32. G R A P H I S T R Y info@graphistry.com
G R A P H I S T R Y info@graphistry.com
33
Most people are busy so
graph-app-kit helps you
make Easy Buttons
33. G R A P H I S T R Y info@graphistry.com
34
…
quicklaunch
connect
low-code
share
G R A P H I S T R
Y
34. G R A P H I S T R Y info@graphistry.com
35
privately
script
point-and-click graph apps
(public + private)
…
35. G R A P H I S T R Y info@graphistry.com
36
Minimal app
• Name
• Some text
Auto-loads into app picker
Connector Templates
• CSV
• TigerGraph
• Neptune
• …
36. G R A P H I S T R Y info@graphistry.com
37
Pipeline template
Most apps are copy/edit of: Form controls -> run queries & filter -> viz
37. G R A P H I S T R Y info@graphistry.com
1. Graph era of visual analytics: Graphistry
2. No-code: Data -> relationship insights in 10s
3. Scale with GPUs: 412GB/s on a single GPU node
4. Low-code apps with graph-app-kit
100X graph platform
38
graphistry.com/get-started
github.com/graphistry/pygraphistry
github.com/graphistry/graph-app-kit
G R A P H I S T R Y
Subscribe, explore, & contribute:
38. G R A P H I S T R Y info@graphistry.com
39
graphistry.com/get-started
github.com/graphistry/pygraphistry
github.com/graphistry/graph-app-kit
G R A P H I S T R Y
Subscribe, try, & contribute: