VIRT2550BU Reducing Latency in Enterprise Applications or ... · Reducing Latency in Enterprise Applications Caused by Hair-Pinning with VMware NSX VMworld 2017 Content: Not for publication

Ryan BradleyMohammad HannanBob Goldsand

VIRT2550BU

#VMworld #VIRT2550BU

Reducing Latency in Enterprise Applications Caused by Hair-Pinning with VMware NSX

VMworld 2017 Content: Not fo

r publication or distri

bution

Deloitte Case Study with NSX Overview by VMware | VIRT2550BUReducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSXSolutions Network (SNET) | Deloitte Consulting LLP | August 2017



bution

#VIRT2550BU Reducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSX Copyright © 2017 Deloitte Development LLC. All rights reserved. 3

Presenters

Ryan BradleyArchitecture ManagerDeloitte Consulting LLP

Mohammad HannanBusiness Technology AnalystDeloitte Consulting India Private Ltd.

Bob GoldsandSenior Staff ArchitectVMware



bution

• This presentation may contain product features that are currently under development.

• This overview of new technology represents no commitment from VMware to deliver these features in any generally available product.

• Features are subject to change, and must not be included in contracts, purchase orders, or sales agreements of any kind.

• Technical feasibility and market demand will affect final delivery.

• Pricing and packaging for any new technologies or features discussed or presented have not been determined.

Disclaimer

#VIRT2550BU CONFIDENTIAL 4



bution


• Problem Statement – Mutli-tiered Applications and Hairpinning

• VMware NSX Overview (Presented by VMware)

• Hypothesis and Testing Methodology

• Testing Results

• Conclusions

• Questions & Answers

Reducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSX

Agenda



bution

#VIRT2550BU Reducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSXCopyright © 2017 Deloitte Development LLC. All rights reserved. 6

Problem StatementTraditional Multi-Tiered Applications and Hairpinning



bution


Multi-Tiered Applications Defined

Problem Statement

• Applications are often logically or physically separated into multiple application tiers

• Driven by architectural best practices surrounding scalability, security, segregation, 24/7 availability

• A common application architecture utilizes 3 tiers:

– Presentation Tier

– Logic or Business Tier

– Data Tier

• In some scenarios the presentation tier is hosted on or part of the client machine, e.g., an SAP client installed on end user’s laptop

Data Tier

Logic Tier

Presentation TierTop most level of the application. This tier will produce something that the user can interact with and/or understand.

This tier makes logical decisions through application code and processing commands. It is also responsible for moving data between the Presentation and Data Tiers.

This tier contains the back end database or filesystem were information is stored.



bution


• Enterprise application architectures will often utilize multiple tiers. Tier counts will vary, vendor application specific communication methods will vary,

Your mileage may vary…

• Two & three tier are most common

• Some use cases call for more tiers

• For the purpose of our discussion, and to simplify our testing, these exact vendor application scenarios were not considered or evaluated. Instead, a generic approach was taken

Enterprise Application Architecture Examples

Problem Statement

Tier 2 & 3 Tier 4 Tier 5

SAP Applications Java EESAP Business

Objects BI

OracleApplications

Ruby on Rails JAVA 2EE

Sage ERP X3 Enterprise SOADBMS-Senko Five Layer Architecture



bution

# VIRT2550BU Reducing Latency in Enterprise Applications Caused by Hairpinning with VMware NSX Copyright © 2017 Deloitte Development LLC. All rights reserved. 9

• Application tiers are usually isolated by VLAN requiring routing for cross-tier communication

• Communication between tiers requires network traffic to traverse physical network devices

− Top of Rack (ToR) & Core Switches, Routers

− Firewalls as inter-VLAN communication typically involves firewall policy

Multi-Tiered Applications – Intra Tier Communication

Problem Statement



bution


Hairpinning and Multi-Tiered Applications

Problem Statement

• The National Institute of Standards and Technology (NIST) defines “hairpinning” as:

“the performance penalty due to increased latency involved in routing the virtual

network traffic to the physical network outside the virtualized host and then back to the virtual network inside the virtualized host.”

• Problem directly correlated with tier count

− i.e., more application tiers leads to more VLAN traversal which requires additional routing, more physical hops

• This presentation will focus on the impact of network latency in hairpinning scenarios using a simulated multi-tiered application

Source: http://csrc.nist.gov/publications/drafts/800-125B/sp800_125b_draft.pdf

WEB APP

10.1.1.0/24 10.1.2.0/24 10.1.3.0/24

DB

Can hairpinning be avoided?



bution

VMware NSX OverviewSoftware Defined Networking

11



bution

VMware NSX

• Lots Of Key Features With NSX

• Micro-Segmentation, Workload Isolation, Security

• But Application/Database Architects - expect above features

• Why Should Application Owners, dba’s, CIO’s Care

• Better performance with NSX

• Seems counterintuitive…

– Virtualization performance penalties??

• NSX Changes The Game

• Day 2 Operations & Intelligent Enterprise Management

• Typically 20% to 40% faster

• 2x to 3x faster depending on operation

• Reduces time to value in the enterprise

Software Defined Networking Highlights




bution

VMware NSXArchitecture and Components




bution

VMware NSXEliminating Hair-Pinning with VMware NSX

Communication Never Leaves the Host with NSX Reduce Latency and Network Hops with NSX




bution

Performance Gains with NSX and vRealize Network Insight

• Eliminating Hair-Pinning – Design Considerations

– NSX + Workload Mobility – Live Migrations vMotion

– Intelligent Operations – vRealize Network Insight

– Optimizations Not Possible in Physical World; Only in the SDDC

• Typical Day 2 Operation Which Can Benefit from NSX

– Enterprise Data Warehouses – data loads from multiple source systems

– Optimization of Scale Out Databases – Identify busy node or nodes

– Backup/Recovery

– Node reconfigurations

– IOT operations, ETL/ELT

– Essentially any data movement within the enterprise

– Latency Sensitive Workloads


VMware NSX

Live Migrations -vMotion

vRealize Network InsightVMworld 2017 Content: Not fo


bution

• VMware NSX

– Abstract Layer 2 thru 7 Network Service

– VXLAN IP overlay technology

• Install 3 vSphere Installation Bundles

• Foundational Components

– Logical Switches

• VXLAN port group on a Distributed Switch

– Distributed Logical Routers (DLR)

• The distributed router allows routing decisions to be made inside the hosts

• Design Goal or Takeaway

– NSX Minimize Network Latency – elimination of hair-pinning

– Contain Network Traffic within ESXi Kernel

– Not Possible In Physical World

VMware NSX Components & Services


WebTier AppTier DBTier

172.16.10.0/24

172.16.20.0/24

172.16.30.0/24.1

.1

.1

Network Traffic Flow NSX

Logical Switches

Distributed Logical Router



bution

VMware NSX

• Results: Oracle 12c Business Intelligence Deployment

– NSX/vRNI Workload Management techniques can be used on any mission critical application/database

Measured: Application Performance 40-60% faster with NSX




bution


Testing the HypothesisReducing Latency in Multi-Tier Application Architectures



bution


What Did We Expect to Find?

Testing the Hypothesis

What we heard:

The NSX implementation of VXLAN overlay is designed to reduce physical network hops, thus helping to:

• Reduce application latency which will…

• Improve application performance by an unknown factor of “X” which will…

• Reduce overall job execution time

Some initial thoughts:

This “X” improvement will increase/vary based on:

• Amount of data being transferred across the network connection

• Number of network hops removed from the application

• Placement of VMs on hosts (all VMs on same host vs. VMs on different hosts)

Note: VMware NSX is not required nor is it the only product that can deploy VXLAN. This case study only evaluated VMWare NSX’s implementation of VXLAN.

Can VMware NSX overcome some hairpinning challenges in multi-tiered applications?

We put it to the test…



bution



Automating and Executing the Tests

− ApacheBench (ab)

− With Varying:

◦ # Requests

◦ File Size retrieved

Simulating a Multi-tiered Application

− Apache Redirects

− With Varying:

◦ Tier Count (number of VMs and redirects) – 2 through 6 tiers

◦ VM Placement (all VMs same host or each VM different host)

Capturing the Results

− Data Recorded and Compared

◦ NSX with VXLAN

vs.

◦ Non-NSX with VLAN

Testing Methodology



bution


Physical and Virtual Environment Specifications


• Physical Environment Specifications

– Compute - Cisco UCS

– Networking – Cisco Nexus

– Storage – EMC VNX

• Virtual Environment Specifications

– VMware vSphere 6.0

– NSX for vSphere 6.3.2

• Virtual Machine and Operating System Specifications

– 2 vCPU x 8GB vRAM per App VM

– RedHat Enterprise Linux 7.3VMworld 2017 Content: N

ot for publicatio

n or distribution


Capturing Apache Benchmark Results


2 Tier 3 Tier 4 Tier

5 Tier 6 Tier



bution


Benchmark Testing ResultsReducing Latency in Multi-Tiered Application Architectures



bution


ApplicationTiers

Different Hosts

Same Host

2 tier 1.94x 4.14x

3 tier 2.06x 4.95x

4 tier 2.79x 6.33x

5 tier 3.03x 6.71x

6 tier 4.81x 7.64x

”X” Improvement of NSX vs non-NSX - Averaged Results View

Benchmark Testing Results

1.94 2.06

2.793.03

4.81

4.14

4.95

6.33

6.71

7.64

0

1

2

3

4

5

6

7

8

9

2 3 4 5 6

Avera

ge "

X"

Impro

vem

ent

Application Tiers

Different Hosts Same Host

Source: Unless specified, all data is from Deloitte internal testing.

Average X factor of improvement increased linearly as tiers were added



bution


1.14

2.192.50

1.43

2.34 2.41

1.09

3.383.90

1.14

3.53

4.42

1.27

5.82

7.34

0

2

4

6

8

10

12

14

0.184 4.3 13.7

"X"

Impro

vem

ent

File Sizes (MB)

2 Tier 3 Tier 4 Tier 5 Tier 6 Tier

Same Host, #Requests Averaged

1.72

5.38 5.31

3.70

5.45 5.70

2.35

7.31

9.32

2.70

7.58

9.86

2.39

8.02

12.50

0

2

4

6

8

10

12

14

0.184 4.3 13.7

"X"

Impro

vem

ent

File Sizes (MB)


Different Hosts, #Requests Averaged

“X” Improvement - Varying File Sizes View


More data transferred = More improvement



bution


1.66 1.71 1.80

3.80 3.78 3.52

2.66 2.58

1.822.16

3.112.81

2.182.57 2.42

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent

Number of Apache Benchmark Requests


Same Host, File Size: 0.184 MB

1.13 1.14 1.161.35 1.44 1.501.11 1.06 1.101.09 1.23 1.081.23 1.27 1.31

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent



Different Hosts, File Size: 0.184 MB

“X” Improvement - Varying Requests View – Small File Size


Number of requests did not affect factor of improvement



bution


“X” Improvement - Varying Requests View – Medium File Size


Different Hosts, File Size: 4.3 MB Same Host, File Size: 4.3 MB

2.15 2.19 2.242.37 2.31 2.32

3.37 3.49 3.293.54 3.60 3.44

5.96 5.73 5.77

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent



5.775.09 5.295.41 5.42 5.52

7.27 7.077.587.80

7.43 7.51

8.77

7.377.91

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent






bution


“X” Improvement - Varying Requests View – Large File Size


Different Hosts, File Size: 13.7 MB Same Host, File Size: 13.7 MB

2.50 2.48 2.512.45 2.37 2.43

3.95 3.96 3.794.40 4.36 4.51

7.28 7.33 7.40

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent



5.09 5.33 5.525.64 5.71 5.75

9.579.10 9.28

9.83 9.77 9.99

12.14 12.24

13.12

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent






bution


“X” Improvement - 3 Tier Environment View



(File Sizes) (File Sizes)

1.35 1.44 1.50

2.37 2.31 2.322.45 2.37 2.43

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

3.80 3.78 3.52

5.41 5.42 5.525.64 5.71 5.75

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

Linear increase in “X” improvement with an increase in file size



bution






1.11 1.06 1.10

3.37 3.49 3.293.95 3.96 3.79

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

2.66 2.58

1.82

7.27 7.077.58

9.579.10 9.28

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB




bution






1.09 1.23 1.08

3.54 3.60 3.44

4.40 4.36 4.51

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

2.16

3.112.81

7.807.43 7.51

9.83 9.77 9.99

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB




bution






1.23 1.27 1.31

5.96 5.73 5.77

7.28 7.33 7.40

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

2.182.57 2.42

8.77

7.377.91

12.14 12.24

13.12

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB




bution


ApplicationTiers

Different Hosts

Same Host

2 tier 1.94x 4.14x

3 tier 2.06x 4.95x

4 tier 2.79x 6.33x

5 tier 3.03x 6.71x

6 tier 4.81x 7.64x

”X” Improvement of NSX vs non-NSX - Averaged Results View


1.94 2.06

2.793.03

4.81

4.14

4.95

6.33

6.71

7.64

0

1

2

3

4

5

6

7

8

9

2 3 4 5 6

Avera

ge "

X"

Impro

vem

ent

Application Tiers


Average “X” factor of improvement increased linearly as tiers were added



bution


Conclusion



bution


Our Testing Demonstrated NSX with VXLAN vs. Non-NSX with VLAN

• Displayed overall faster test results

• Specific applications will vary. Look at the trends instead of the numbers. The “X” factor of improvement increased as:

• Data transfer size increased – generally linear

• Number of application tiers increased – generally linear

• All VMs of application were on same host

• “X” factor of improvement did not increase as the number of the requests increased. Number of requests did not affect performance increases


In Summary

• Multi-tiered enterprise applications could see performance improvements with NSX with VXLAN compared to more traditional non-NSX with VLAN environments.

• NSX can eliminate some network hops, which could improve page load times, job execution times, and application performance overall.

• The elimination of some network hops may help deliver better overall applications performance, leading enterprises to achieve value for their Software-Defined Data Centers

Lessons Learned & Conclusions



bution


Questions



bution


Thank You



bution


Appendix



bution





1.13 1.14 1.16

2.15 2.19 2.242.50 2.48 2.51

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB

1.66 1.71 1.80

5.775.09 5.295.09 5.33 5.52

0

2

4

6

8

10

12

14

500 1000 1500

"X"

Impro

vem

ent


0.184 MB 4.3 MB 13.7 MB(File Sizes) (File Sizes)

Linear Increase in “X” Improvement with an increase in file size



bution

About Deloitte

As used in this document, “Deloitte” means Deloitte Consulting LLP, a subsidiary of Deloitte LLP. Please see www.deloitte.com/us/about for a detailed

description of our legal structure. Certain services may not be available to attest clients under the rules and regulations of public accounting.

This publication contains general information only and Deloitte is not, by means of this publication, rendering accounting, business, financial, investment,

legal, tax, or other professional advice or services. This publication is not a substitute for such professional advice or services, nor should it be used as a

basis for any decision or action that may affect your business. Before making any decision or taking any action that may affect your business, you should

consult a qualified professional advisor. Deloitte shall not be responsible for any loss sustained by any person who relies on this publication.

Copyright © 2017 Deloitte Development LLC. All rights reserved.



bution

Documents

VIRT2550BU Reducing Latency in Enterprise Applications or ... · Reducing Latency in Enterprise Applications Caused by Hair-Pinning with VMware NSX VMworld 2017 Content: Not for publication