.:MattTopper.com:.

As I’m sure all of you have seen Larry finally announced our new Exalogic box tonight.  An amazing piece of hardware and software married together.  I’ll probably be posting more about this over the next few weeks but here are my notes from the keynote below:

360 Cores
30 Servers
40 GBps Infiniband Backend
Direct connection to Exadata
Integrated Storage Appliance to house application software and files, patch the VM by downloading one file and it patches all the software on the device by placing it on the storage appliance
2 Guest OSes: Solaris and Linux
Coherence for memory synchronization, illusion of one unified memory system
Software optimized for the hardware
No single points of failure on the box at all
Duplexed fault tolerant storage
Can patch the hardware and software consistently across all their customers because its all engineered the same with standard configurations
Internet Apps: 12x improvement, over 1 million http requests per second, facebooks traffic in 2 full racks
Messaging Apps: 4.5x improvement, 1.8 million messages per second, all china rail ticketing in 1 rack
2.8 TB of DRAM
960 GB of solid state disk for persistence
4 TB Read Cache
40 TB SAS Disk Storage
72 GB Write Cache
1.2 microsecond latency
10 GB Ethernet to the Database
10x Network Latency reduction
Eliminated buffer copies
64K packets vs. standard 4K
3x throughput vs. 10 GbE
Parallelized Message Queues in WebLogic for Exalogic, multiplexed over infiniband
Dynamic load balancing to Exadata Database RAC nodes with transaction affinity to appropriate RAC nodes to maximize locality
SQLNet over Inifiniband (SDP) to maximize JDBC performance
Instant failover in case of node failure due to Infinibands underlying protocol
Cache Coherence
Instance state replication across nodes
Near instant access to data from other compute nodes
Lossless infiniband network enables instantaneous state failure detection and failover
Solid State Disk eliminates Java VM heap limitations
Reduces GC pause times and increases cache capacity
1/4 of a rack to 8 full racks in a single cloud
Runs all applications, not just WebLogic
Built in application isolation and security with weblogic domains
Built in network isolation and security with infiniband partitions and virtual lanes
Build for elastic capacity on demand
Maintain a balanced system as your size and number of racks grow
Standardized and easy to manage
Exadata machines phone home when there are problems, unified management and monitoring interfaces
Single patch for exalogic systems
all customers run the same configuration
all software components can be patched together
all patches are built, packaged, and tested together
enterprise manager automates patch and upgrade procedures
IBMs best vs Exalogic
Exalogic => $1,075,000, 40% more CPUs over IBM, horizontal scale out infrastructure, inifiniband fabric, fully fault tolerant
Power 795 => $4,440,000, Old SMP vertical scale-up system, no fault tolerance
Elastic and Virtual Public and Private Clouds in a Box
Best Performance and Cost Performance in one

Exalogic VM and OS
Both Linux and Solaris as the guest OSes
Elastic Capacity on Demand to add and remove VMs
Fault Isolation to the virtual network level
2-4% CPU overhead for OracleVM virtualization
Instantaneous migration across VMs
Single Route I/O (SR-IOV) Virtualization
dedicated I/O bypassing hypervisor
low latency
50% better infiniband utilization
Dual Personality to Next Version of Linux
OEL -> 5000 customers
Compatible with RHEL
Never been a compatibility bug between RHEL and OEL
RH does not test releases with Oracle products
RedHat is slow to mainline community enhancements, their current kernel is nearly 4 years old
Announcing: The Unbreakable Enterprise Kernel for Linux
Fast, modern, reliable, and optimized for Oracle
Used by Exa-machines for extreme performance
Allows oracle to innovate without sacrificing compatibility
Oracle vs RedHat
5x on flash cache reads
2.4x on solid state disk access
3x Infiniband RDS messages (IOPS)
1.8x transactions per minute on 8 socket database OLTP
Bigger servers up to 4096 CPUs and 2TB of memory
Up to 4 PB of disks
Advanced NUMA support
APCI 4.0 support for power and cooling savings
Data integrity to make sure corrupt data cannot be written
Database data integrity enabled ASM drivers now
Hardware Fault Management built in
Diagnostic Tools with less overhead and performance for tracing
Upgrade by recompiling the kernel, no reinstall needed, OEL and RHEL customers both

Some more updates:
All of the middleware machines are disk less and boot from the storage server, this also centralizes all the logs
The storage server is a Sun Storage 7000 at its core
Most organizations will get over 240 managed servers per full rack
You obviously don’t have to utilize an Exadata rack with Exalogic, but its certainly going to be better with the Infiniband backends
Each managed server should be assigned it own VIP for failover and load balancing as they move across the nodes that maps to a single application VIP
Any OEL5 or Solaris 11 application should be able to run on the machine out of the box, doesn’t get the secret sauce of java performance, but will likely see gains due to the matched hardware.
Most customers will also need to license the ExaLogic Elastic Cloud software and WebLogic Suite for each compute node they purchase (per processor). Can reuse if they already have them.
BI Publisher is shipped to report on top of the audit logs generated from the server, it is recommended that the logs are loaded into an Oracle Database for easier reporting.
ExaLogic machines will be available in the ETC’s throughout the country on the rotation program.
No word yet if you can combine a 1/2 rack of ExaLogic and a 1/2 rack of ExaData into a single physical rack when ordering.