Red Hat Summit 2009 Bryan Che

Red Hat Enterprise MRG Update

Bryan Che

Product Manager, Red Hat

September 2, 2009

About Red Hat Enterprise MRG

Integrated platform for high

performance distributed

computing

High speed, interoperable, open standard AMQP Messaging

Deterministic, low-latency Realtime

kernel

High performance & throughput computing Grid scheduler for

distributed workloads and Utility/Cloud computing

MRG 1.0 launched at Red Hat Summit 2008 in Boston

M

RG

M

essaging

Enterprise Messaging System that

Implements AMQP (Advanced Message Queuing Protocol), the first open

messaging standard

Spans all use cases in one

implementation to consolidate architectural silos (fast messaging, reliable messaging, large file transfer, publish/subscribe, eventing, etc)

Uses Linux-specific optimizations to achieve breakthrough performance

on Red Hat Enterprise Linux and MRG Realtime

Runs on non-Linux platforms without the full performance and quality of service benefits that Red Hat Enterprise Linux provides

Features introduced in MRG 1.1

latency messaging

Active-Standby/Active-Active Broker Clustering

Security: SASL auth, SSL encryption, role-based access control

Native .NET Messaging Client, including WCF

Performance enhancements

Queue Semantics: Ring Queue, Last Value Queue, TTL

Federation dynamic routes JMSXUserId support

M

RG

M

essaging Infiniband Throughput:

Over 2 Million Reliable Messages/second

8 Bytes 16 Bytes 32 bytes 64 Bytes 128 Bytes 256 Bytes 512 Bytes 1024 Bytes 0 500000 1000000 1500000 2000000 2500000 0 200 400 600 800 1000 1200 1400 1600 2,012,741 1,953,957 1,975,520 1,870,218 1,758,274 1,781,642 1,671,314 1,545,205 15.35 29.84 60.29 99.26 214.63 434.97 816.07 1508.99

HP Blades 4 Node AMQP inf iniband

Messages MB/Sec Packet Sizes M e ssa g e R a te 8 systems: 4 load drivers, 4 compute nodes HP ProLiant BL460c G6 Dual Intel(R) Xeon(R) CPU X5570 @ 2.93GHz 24GB RAM Infiniband 4X QDR IB Dual-port Mezzanine HCAs(1 port connected) BLc 4X QDR IB Switch

M

RG

M

essaging Infiniband RDMA Latency:

Under 65 Microseconds Reliably Acknowledged

1 3 5 7 9111315171921232527293133353739414345474951535557596163656769717375777981838587899193959799101 0.0000 0.0100 0.0200 0.0300 0.0400 0.0500 0.0600 0.0700 0.0800 0.0900 0.1000

HP Nehalem BL460c Infiniband Latencytest with RDMA

8 Bytes 16 Bytes 32 Bytes 64 Bytes 128 Bytes 256 Bytes 512 Bytes 10 24 Bytes La te nc y (m s)

M

RG

M

essaging on KVM Virtualized Performance:

Over 1 million Messages/Second Throughput

16 32 64 128 256 512 1024 2048 4096 0 200000 400000 600000 800000 1000000 1200000 0 100 200 300 400 500 600 700 800 900 104 6081 1023869 902689 ₈₈₀₉₆₅ 804 04 5 74 1297 5554 65 36914 5 210634

RHEL 5.4 KVM AMQP 2-Guest

Dell Poweredge R710 Intel Nehalem, 2 10Gbit VT-d

Msg/sec Throughput MB/sec Msg Size (bytes) M es sa ge s / S ec

M

RG

M

essaging on KVM Virtualized Performance:

<200 Microsecond Latency

Ave Lat (milsec) 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 0.000 0.050 0.100 0.150 0.200 0.250

RHEL5.4 KVM AMQP Messaging Perf

Dell Poweredge R710 Intel Nehalem, 2 10Gbit VT-d

Bare Metal KVM vtd KVM novtd (Samples of 10k messages) La te nc y in M ili se co nd s

M

R

G

R

ealtime

Enables applications and

transactions to run predictably, with

guaranteed response times

Provides microsecond accuracy

Provides competitive advantage &

meets SLA's

Travel web site: missed booking Program trading: missed trades Command & Control: life & death

Upgrades RHEL5 to Realtime OS

Provides replacement kernel for RHEL 5; x86/x86_64

Preserves RHEL Application Compatibility

Red Hat Leads Upstream Linux

Realtime Development

M

R

G

R

ealtime Tools

MRG includes a new tuning tool, TUNA

MRG includes a new MRG Realtime

Latency Tracer

Runtime trace capture of longest latency codepaths – both kernel and application. Peak detector

Selectable triggers for threshold tracing

Detailed kernel profiles based on latency triggers

Existing standard RHEL5 based

performance monitoring tools remain

relevant

Hardware Matters

Hardware can have a big effect on realtime performance

Hardware drivers may need to be updated to handle threaded interrupts

Many system BIOS's include System Management Interrupts (SMIs)

Cause non-deterministic latency beneath the operation system by taking CPU cycles for things like power management, administration

SMI latencies cannot be resolved by realtime linux—they require the hardware OEM to remove SMIs or make them configurable

Red Hat has worked with OEMs to certify systems for MRG Realtime

MR

G

G

rid

Provides both HPC/HTC and

Utility/Cloud Computing

Brings advantages of scale-out and flexible deployment to any application or workload Delivers better asset utilization, allowing

applications to take advantage of all available computing resources

Handles “Holiday Rushes”

Executes across multiple platforms and in virtual machines

Provides seamless and flexible

computing across:

Local grids Remote grids

Remote clouds (Amazon EC2)

Cycle-harvesting from desktop PCs

Fully supported in MRG 1.1

Web-Based Management Console

Unified management across all of MRG for job, system, license management, and workload management/monitoring Low Latency Scheduling

Sub-second job scheduling via AMQP Messaging

Virtualization Support

Cloud Integration with Amazon Ec2 Concurrency Limits

limits how much of a certain resource can be used at once

Dynamic Slots

Cloud Computing with MR

G

G

rid

MRG integrates internal clouds, hybrid clouds, and public

clouds from one interface

Internal Cloud:

Build your own cloud with Red Hat Enterprise Virtualization

and Red Hat Enterprise MRG

Hybrid Cloud:

Augment your internal cloud with resources from remote

grids/clouds, idle resources, virtual machines, bare metal, etc

Public Cloud:

Schedule jobs to Amazon EC2 from the same interface as for

local clouds

Next Session at 4:30: Building and Leveraging Compute

Clouds with Red Hat Enterprise MRG

QMF: AMQP Messaging-Based Management

Red Hat Enterprise MRG's entire

management/monitoring system is

AMQP messaging-based

Asymmetric, Efficient, Scalable, and Secure Any messaging client can manage

QMF: AMQP Messaging-Based

Management Framework

Agent-defined management model (self-describing)

Objects (properties, statistics, and methods/controls), Events

Ease of development and extensibility

AMQP

Network

Server EventStorage

procfs Console

M

RG

M

essaging Roadmap

MRG 1.2 Highlights

Protocol independent API for c++, python and ruby clients

iwarp support for RDMA

Durable stores will function in clustered environments

Performance enhancements

Beyond MRG 1.2

AMQP 1.0

Unix Domain Sockets Enhanced performance Enhanced tools

M

R

G

R

ealtime Roadmap

MRG 1.2 Highlights

2.6.31 kernel

Improved Throughput Performance Rteval realtime evaluation toolkit New perf tools

New Performance Counter System

interface to underlying hardware registers that track events that affect performance. e.g. cache misses, number of context

switches, etc.

Profile at hardware speeds New perf command for interfacing

with the Performance Counter System. e.g. perf top, perf record, perf report

Beyond MRG 1.2

Enhanced performance

Continue to track upstream kernel and merge realtime into the mainline trunk Enhanced tools

Ftrace user space latency tracing Enhance rteval to provide qualitative

assessments

Integrate TUNA with Web management console Additional hardware support and

MR

G

G

rid Roadmap

MRG 1.2 Highlights

New User Tools, including GUI Submit Updated Admin Tools

Windows Execute Node Support Resource restrictions

Updated Amazon EC2 Enhanced Support Enhanced scalability

Beyond MRG 1.2

Fine-grained SELinux integration & policies (SELinux Sandbox)

Configuration for diskless Condor

Dynamic provisioning preemption support Pre-built configurations and job templates Integration with RHEV

Additional Information

http://www.redhat.com/mrg

Next Session at 4:30:

Building and Leveraging Compute Clouds with Red Hat

Enterprise MRG