innovation driving opportunity for the web infrastructure
play

Innovation Driving Opportunity for the Web Infrastructure John - PowerPoint PPT Presentation

Innovation Driving Opportunity for the Web Infrastructure John Fowler Executive Vice President, Systems Sun Microsystems, Inc. Communications Arrives Smart Mobs Co-Creation Wikis Blogosphere Social The Tags Networks Long Tail Next


  1. Innovation Driving Opportunity for the Web Infrastructure John Fowler Executive Vice President, Systems Sun Microsystems, Inc.

  2. Communications Arrives Smart Mobs Co-Creation Wikis Blogosphere Social The Tags Networks Long Tail

  3. Next Generation Software

  4. • $/Performance OPEN • $/Gigabyte NOW Storage NOW Compute COMING This is not your • Many cores Networking Daddy's • Many threads Infrastructure • Open platforms • Huge bandwidth • Open platforms � Everything connected � Open standards everywhere � Horizontal scaling by definition

  5. Web 2.0 Architecture External INTERNET Clients Resources Horizontal Scaling, Reverse network intensive (load balancing, Compute Nodes Proxies filtering) Web/App Cache Servers Servers Distributed Object Stores Database Storage NFS Nodes Nodes STORAGE COMPUTE (unstructured data) NETWORK

  6. Web 20kit Structure Geocoder Driver Web/App Server SYSTEMS Database Storage UNDER Local Mogile FS TEST STORAGE COMPUTE NETWORK

  7. Power, Heat, Space Scale Understanding the Infrastructure It’s Driving You CRAZY! Performance

  8. Compute More threads, greater efficiency, lower energy consumption, less heat Memory latency Compute Memory latency compute time Higher frequencies = more heat Growth of thread-rich applications

  9. Cores and Threads Are on the MOVE 16 CORE 8 CORE

  10. Application Performance and Memory Capacity Throughput 4GB 16GB 64GB 256GB 1TB Memory Capacity

  11. Open Storage Servers + Storage + • Built on OpenSolaris OpenSolaris • Performance of ZFS + SSDs • Cost efficiency of volume Hardware • Scale Easily > Millions of files > Gigabytes/sec • Management simplicity > Analytics with dTrace > Diagnostics with FMA

  12. Why Applications Don’t Perform Waiting for DATA • Today’s Multi-Core, Server Multi-Socket application server designs are increasingly held back by slow storage High • When requesting data, the Performance Multi-Core, server spends most of its Multi Socket CPUs DRAM time waiting for storage 260 Times • Application performance Slower remains sluggish regardless of the server CPU horsepower 15K RPM Disk Drives • The traditional remedy of adding more expensive DRAM may no longer suffice as data Storage sets double every 2 years

  13. Integrating Solaris (TM) ZFS & Co- NEWS Developing with Industry Leaders TODAY! COMING SOON! D E n t e r p r i s e S S E n t e r p r i s e H D D � 180 Write IOPS O P S 0 - 8 , 0 0 0 W r i t e I u p t o 5 , 0 0 � � 320 Read IOPS R e a d I O P S 3 0 , 0 0 0 - 4 0 , 0 0 0 u p t o � � 146GB 3 2 G B � $ p e r I O P S : 2 . 4 3 0 8 $ p e r I O P S : 0 . 1

  14. Reliability Comparisons � � Bad Block Management � � Wear leveling � � ECC Correction ! � No Moving Parts Power Consumption 13 Watts 2 Watts Thermal Operating Range 5-55 Celsius 0-70 Celsius 1.2 Million 2 Million MTBF

  15. ZFS* Hybrid Storage Pool Model Automated Data Management High Performance Read & Write Cache Pool � ZFS combines main memory and SSDs for read caching � Adaptive Replacement Cache (ARC) Read/ Write ZFS Intent Log pool Cache � ZFS is a transactional file system SSDs ZIL � Intent Log stores small transactions without SSDs doing the transaction � ZFS can use a separate intent log device (slog) HDD � Typically NVRAM or dedicated SDD pool used Pool High Capacity Pool � The ZFS pool manager automatically stripes across any number of high capacity HDDs � ZFS IO-scheduler bundles disk IO to optimize Shipping Now in arm movement and sector allocation *Other names and brands may be claimed as the property of others.

  16. New Server Memory Hierarchy L1 Cache L2 Cache Pico Sec. L3 Cache Nano Sec. RAM Micro Sec. FLASH Milli Sec. Disk

  17. The Hybrid Storage Pool Gaining Performance and Capacity Balance Capacity Performance Quad Channel FB-DIMM Xeon 7350 2.93Mhz 6x Nx 1TB HDD 24x SSD 32 DIMM Slot s SAS (16 cores) 3G 1066MT FSB DDR2 667 DDR2 667 1066MT FSB PCI-e 8x 1066MT FSB DDR2 667 DDR2 667 1066MT FSB 840,000 IOPS 845,000 IOPS 840,000 IOPS Sun estimation of performance with assumption of 18,000 IOPs per Ghz per Core for System. 200 IOPS per HDD and 35,000 IOPS per SSD 1 1

  18. Hybrid Storage Pool in Action Up to 3x the Performance and Reducing Power Consumption by 5x 4 Xeon 7350 Processors (16 cores) 32GB FB DDR2 ECC DRAM SunFire X4450 Server Hybrid Storage Pool Traditional Storage Pool (1) 32GB SSD ZIL Device (1) 80GB SSD Cache Device (5) 400GB 4200RPM SATA Drives (7) 146GB 10,000RPM SAS Drives Before 2010, Flash will be in Every Server 1 1

  19. ZFS/ Solaris Sun Studio Our Communities

  20. Web 2.0 Architecture External INTERNET Clients Resources Horizontal Scaling, Reverse network intensive (load balancing, Compute Nodes Proxies filtering) Web/App Cache Servers Servers Distributed Object Stores Database Storage NFS Nodes Nodes STORAGE COMPUTE (unstructured data) NETWORK

  21. • Visit the Sun booth today Let's Break the > See our demos Rules > Attend our presentation on the Together Web20kit at 3:45PM, Salon A-D • Learn more about working with us > Try and Buy > Start Up Essentials • Join the Sun Developer Network > developers.sun.com/services • Test drive and download our technologies Visit Us sun.com/web

  22. Thank You John Fowler john.fowler@sun.com Executive Vice President, Systems Sun Microsystems, Inc.

Recommend


More recommend