1. WebSphere XD Compute Grid High Performance Architectures
Snehal S. Antani, WebSphere XD Technical Lead SOA Technology Practice IBM Software Services

2. Overview Key Components of WebSphere XD Compute Grid
Job Scheduler [formerly named the Long Running Scheduler (LRS)]
Parallel Job Manager (PJM)
Grid Endpoints [formerly named Long Running Execution Environment (LREE)]
High Performance Architectures and Considerations

3. XD Compute Grid Components
Job Scheduler (JS)
The job entry point to XD Compute grid
Job life-cycle management (Submit, Stop, Cancel, etc) and monitoring
Dispatches workload to either the PJM or LREE
Hosts the Job Management Console (JMC)
Parallel Job Manager (PJM)-
Breaks large batch jobs into smaller partitions for parallel execution
Provides job life-cycle management (Submit, Stop, Cancel, Restart) for the single logical job and each of its partitions
Is *not* a required component in compute grid
Grid Endpoints (GEE)
Executes the actual business logic of the batch job

4. XD Compute Grid Components
Load Balancer
EJB
Web Service
JMS
Command Line
Job Console
GEE
WAS
User JS
WAS
GEE
WAS
PJM
WAS

5. Key Influencers for High Performance Compute Grids
Proximity to the Data
Bring the business logic to the data: co-locate on the same platform
Bring the data to the business logic: in-memory databases, caching
Affinity Routing
Partitioned data with intelligent routing of work
Divide and Conquer
Highly parallel execution of workloads across the grid
On-Demand Scalability

6. Proximity to the Data- Co-location of business logic with data
Frame
Job Scheduler
WAS z/OS
WAS z/OS
Controller
Controller
GEE
GEE
GEE
GEE
GEE
GEE
Servant
Servant
Servant
Servant
Servant
Servant
DB2 on z/OS

7. Proximity to the Data- Bring data to the business logic with caching
Frame
CPU
CPU
Job Scheduler
LPAR
GEE
GEE
GEE
Data Grid near-cache
Data Grid near-cache
Data Grid near-cache
CPU
CPU
CPU
CPU
CPU
CPU
LPAR
LPAR
LPAR
CPU
CPU
DG Server
DG Server
LPAR
LPAR
Database

8. Affinity Routing- Partitioned data with intelligent routing of work
Frame
Job Scheduler
Records A-M
Records N-Z
WAS z/OS
WAS z/OS
Controller
Controller
A-D
E-I
J-M
N-Q
R-T
W-Z
GEE
GEE
GEE
GEE
GEE
GEE
Servant
Servant
Servant
Servant
Servant
Servant
DB2 Data Sharing Partition
DB2 Data Sharing Partition
Records A-M
Records N-Z

9. Affinity Routing- Partitioned data with intelligent routing of work
Frame
CPU
CPU
Job Scheduler
GEE
GEE
GEE
Data Grid near-cache
Data Grid near-cache
Data Grid near-cache
CPU
CPU
CPU
CPU
CPU
CPU
Records A-I
Records J-R
Records S-Z
CPU
CPU
DG Server
DG Server
Records A-M
Records N-Z
Database

10. Divide and Conquer- Highly Parallel Grid Jobs
Large Grid Job
Frame
CPU
CPU
PJM
GEE
GEE
GEE
Data Grid near-cache
Data Grid near-cache
Data Grid near-cache
CPU
CPU
CPU
CPU
CPU
CPU
Records A-I
Records J-R
Records S-Z
CPU
CPU
DG Server
DG Server
Records A-M
Records N-Z
Database

11. On-Demand Scalability- With WebSphere z/OS
Frame
Job Scheduler
WAS z/OS
WAS z/OS
Controller
Controller
zWLM
zWLM
GEE
GEE
GEE
GEE
GEE
GEE
Servant
Servant
Servant
Servant
Servant
Servant
DB2 on z/OS

12. On-Demand Scalability- With XD Operations Optimization
Frame
CPU
CPU
CPU
CPU
Job Scheduler
On-Demand Router
LPAR
LPAR
GEE
GEE
GEE
Data Grid near-cache
Data Grid near-cache
Data Grid near-cache
CPU
CPU
CPU
CPU
CPU
CPU
LPAR
LPAR
LPAR
CPU
CPU
DG Server
DG Server
LPAR
LPAR
Database

13. Backup

14. OG Server Miss (Access DB)
Near-Cache Hit
Probability of cache hit
Time (ms) to retrieve data from cache
Data Access
Probability that data is in cache server
OG Server Hit
Time (ms) to retrieve data from cache server
Probability of cache miss
Near-Cache Miss
Time (ms) to retrieve data from other storage
Probability that data must be retrieved from database
OG Server Miss (Access DB)
Time (ms) to retrieve data from database
Data Access time (ms) = (Probability of near- cache hit) * (Time to retrieve data from near-cache) + (Probability of near-cache miss) * (time to retrieve data from other storage); Time to retrieve data from other storage (ms) = (Probability that data is in cache server) * (Time to retrieve data from cache server) + (Probability that data must be retrieved from database) * (time to retrieve data from database);

15. OG Server Miss (Access DB)
Near-Cache Hit P1 S1
Data Access
OG Server Hit P3 S3 P2
Near-Cache Miss S2 P4
Data Access = (Near-Cache Hit) + (Near-Cache Miss)
Near-Cache Hit = (P1)(S1)
Near-Cache Miss = (P2) * [ (P3)(S3) + (P4)(S4) ]
Improve data access time by:
Increase P1:
increase cache size (increase heap, etc)
establish request affinity
Decrease S1:
Dynamically add more CPU
Decrease S2
Increase P3
Increase size of cache server
Establish query/data affinity
Decrease S3
Decrease S4
Reduce network latency
OG Server Miss (Access DB) S4

16. OG Server Miss (Access DB)
Example calculation
Near-Cache Hit
30%
1 ms
Data Access
OG Server Hit
47.2 ms
70%
10 ms
70%
Near-Cache Miss
67 ms
30%
OG Server Miss (Access DB)
Data Access = (Near-Cache Hit) + (Near-Cache Miss)
Near-Cache Hit = (P1)(S1)
Near-Cache Miss = (P2) * [ (P3)(S3) + (P4)(S4) ] Near-cache miss = (.7)(10) + (.3)(200)
= = 67 ms
Data Access = (.3)(1) + (.7)(67)
= = 47.2 ms
200 ms

17. OG Server Miss (Access DB)
Example calculation- effects of increasing size of near-cache
Near-Cache Hit
60%
1 ms
Data Access
OG Server Hit
27.4 ms
70%
10 ms
40%
Near-Cache Miss
67 ms
30%
OG Server Miss (Access DB)
Data Access = (Near-Cache Hit) + (Near-Cache Miss)
Near-Cache Hit = (P1)(S1)
Near-Cache Miss = (P2) * [ (P3)(S3) + (P4)(S4) ] Near-cache miss = (.7)(10) + (.3)(200)
= = 67 ms
Data Access = (.6)(1) + (.4)(67)
= = 27.4 ms
(47.2 – 27.4) / 47.2 = 42% improvement in data access time
200 ms

18. OG Server Miss (Access DB)
Example calculation- effects of adding more CPU and decreasing network latency to the DB
Near-Cache Hit
30%
1 ms
Data Access
OG Server Hit
26.2 ms
70%
10 ms
70%
Near-Cache Miss
37 ms
30%
Data Access = (Near-Cache Hit) + (Near-Cache Miss)
Near-Cache Hit = (P1)(S1)
Near-Cache Miss = (P2) * [ (P3)(S3) + (P4)(S4) ] Near-cache miss = (.7)(10) + (.3)(100)
= = 37 ms
Data Access = (.3)(1) + (.7)(37)
= = 26.2 ms
(47.2 – 25.9) / 47.2 = 45% improvement in data access time
OG Server Miss (Access DB)
100 ms

19. WebSphere XD Compute Grid Infrastructure Topology Considerations