(ATS4-PLAT01) Core Architecture Changes in AEP 9.0 and their Impact on Administrators
0 likes•422 views
The document outlines significant changes in the core architecture of AEP 9.0, including the shift to 64-bit only services and updates to job management and service handling on both Windows and Linux platforms. It details improvements in configuration management, deployment options, and reverse proxy configurations aimed at enhancing performance and scalability. Additionally, it provides insights into load balancing, shared data management, and operational modifications relevant to administrators managing the software.
(ATS4-PLAT01) Core Architecture Changes in AEP 9.0 and their Impact on Administrators
- 1. (ATS4-PLAT01) Core Architecture Changes in AEP 9.0 and Their Impact on Administrators Steven Bush R&D, AEP Core Infrastructure [email protected]
- 2. Platforms • Except where noted, the information within this presentation applies to both Windows and Linux installations. • Starting in 9.0, AEP server installations are now 64-bit only
- 3. Services • 8.5 and earlier – 32 and 64 bit – Apache • 9.0 – 64 bit only – Apache • Manages running jobs • Other services including access to protocol database and files – Tomcat • Back end service provider – Security filter – Internal configuration and system database – Periodic cleanup tasks – Other scheduled tasks – AEPManager (Windows) • Responsible for managing the other services
- 4. Tomcat Cleanup tasks • Remove stale job directories – Expired and invalid jobs • Queue processing – When max running jobs is reached • Remove stale lck files – <InstallDir>/temp/lck • Each running job creates a file with basic information • Running jobs reports, when to queue based on lck file count • If the job process crashes, the lck file remains • Clean up /var/tmp files (Linux) – Used for OS synchronization files (mutexes, conditions, pipes)
- 5. Managing the services (Windows) • AEPManager service – Starts and shuts down Tomcat and Apache – Monitors status of each service to restart when necessary – Right click “My Computer”, choose “Manage”, Navigate to Services. Start/Stop AEPManager. • Modified ApacheMonitor.exe – Based on Apache Foundation apachemonitor.exe – Shows all installed Apache services – Combines all services within same AEP install directory – 8.5 and earlier: use apachemonitor.exe that ships in apache folder
- 6. Managing the services (Windows) • Services Administration Panel • Processes and Memory – Task Manager – Process Explorer • Improved Task Manager • http://technet.microsoft.com/en-US/sysinternals
- 7. Managing the services (Linux) • startserver, stopserver scripts still used – Starts and stops both Apache and Tomcat – /etc/rc.d used for autostart services • Installed by root installation script (scirootinstall) • ps and top • “free –m” to check memory (-m = Megabytes) – read “-/+ buffers/cache” line for real results
- 8. Launch Performance Example which reads 500 molecules and returns a bit of JSON text. Elapsed Time/Throughput vs Number of Simultaneous Clients
- 9. Exporting/Importing Configurations • Export configuration from administration portal – Clone server configurations – Set up new nodes for load balancing configuration
- 10. Maintenance Mode • Disable launching jobs without shutting down the server – Professional UI can still be used to build protocols – Admin portal is still accessible
- 11. New Deployment Options 8.5 9.0 Single Server Yes Yes Windows 32/64 bit 64 bit Linux 32/64 bit 64 bit Private Cluster Linux Linux Public Cluster Linux Linux Grid Engine Linux Linux SSL Only/HTTP Only No Yes Port Forwarded No Yes Load Balanced Web Farm No Yes
- 12. New Globals/URL creation • Should no longer build URLs from server name and port – Will not work with reverse proxies and single port setups – Had problems on clusters and grids • Use @ServerRoot and @ServerSSLRoot instead – Reflects configuration of server • SSL/HTTP only • Reverse proxy/port forwarded • Load balanced • Clusters and grids • Old globals will continue to work for basic configurations
- 13. New Globals/URL Creation 8.5 9.0 HTTP URLs http://$(ServerName):$(ServerPort)/myFile $(ServerRoot)/myfile HTTPS URLs https://$(ServerName):$(ServerSSLPort)/myFile $(ServerSSLRoot)/myfile Internal HTTP URLs Same as client @/pilot_settings/ROOT Internal HTTPS URLs Same as client @/pilot_settings/SSLROOT Internal Server Name Same as client @/pilot_settings/InternalServerName Shared Data Directory @ScitegicRoot @SharedPublicDir
- 14. Protocol Links • http://<server>:<port>/auth/launchjob – Faster launch times – Improved features • streaming results • Integrated authentication – runjob.pl links map to launchjob with _compat=1 flag
- 15. SSL and HTTP Only Modes • Setting found in Admin Portal – Server ConfigurationServer Port Usage • Changes all outgoing URLs to point to correct endpoint – In SSL only mode, @ServerRoot becomes https://.... – In HTTP only mode, @ServerSSLRoot becomes http://.... • Does not turn off second port by default – To disable listening on the unused port, contact support
- 16. Reverse Proxy
- 17. Reverse Proxy • Deploy behind a port forwarding server – Deployment into DMZs – Multiple subnet deployments • Some clients outside firewall, others inside • “SSL termination” option – Connection to reverse proxy is via SSL – From proxy to AEP server is HTTP only
- 18. Reverse Proxy: Configuration • Configure in Admin portal – SetupReverse Proxy and Load Balancing – Set the default server name for your reverse proxy (e.g. firewall.mydomain.net) • Add aliases that might be used as well (e.g. 10.1.2.3,firewall,othername.mydomain.net) – Set the port numbers that are available for your reverse proxy • Leave HTTP or SSL blank if not used
- 20. Load Balanced
- 21. Load Balanced vs. Cluster vs. Grid • Load Balanced (Windows or Linux) – Large numbers of simultaneous users – Individual web requests are distributed – Parallel subprotocols are not distributed – Ideal for short running web based applications that require simultaneous user scalability or failover – Cannot be used for ad-hoc use • Cluster (Linux) – Large numbers of long running jobs – Running jobs are distributed – Parallel subprotocols are distributed – Ideal for long running jobs with large parallelization needs – Useful for ad-hoc use • Grid (Linux) – Similar to cluster – Load is controlled by third party grid software (eg. OGE, LSF, PBS) – Slowest launch times – Ideal for large numbers of very long running jobs and taking advantage of existing grid systems – Possibly slow for ad-hoc use
- 22. Load Balancer Requirements • Install behind any off the shelf load balancer – The AEP server itself does not require affinity, but some applications may choose to impose this limitation – One requirement for the load balancer: Must preserve the HTTP “Host” header.
- 23. Load Balanced: Benefits • Throughput – Huge improvement under high usage • Performance – No difference under low usage versus single servers – Moderate improvement under high usage • High Availability – If a node goes down, other nodes continue functioning
- 24. Load Balanced: Performance Example which reads 500 molecules and returns a bit of JSON text tested against 1-3 nodes. Elapsed Time/Throughput vs Number of Simultaneous Clients
- 25. Load Balanced: Restrictions • Protocol Database is read only – Deploy protocols and components using packages – Can not build models – Intended for applications that rarely change • Parallel subprotocols do not distribute across nodes – Set “Server” to the name of the load balancer instead of “localhost” • Admin Portal connected through the load balancer can only show running jobs – Connect to individual nodes for full portal access – Use “Export/Import” to update each node
- 26. Load Balanced: Shared Data • “Folder Locations” in Admin portal • Jobs directory • Users directory • Shared Public directory – Use @SharedPublicDir instead of @ScitegicRoot • On a single server system, both are equal
- 27. Load Balanced: Configuration • SetupReverse Proxy and Load Balancing – Follow same setup as for reverse proxy – Check box for “Load Balanced” • SetupFolder Locations – Jobs directory – Users directory – Shared Public directory • Use @SharedPublicDir instead of @ScitegicRoot
- 28. The information on the roadmap and future software development efforts are intended to outline general product direction and should not be relied on in making a purchasing decision.