White Papers

ManualsBrandsDell ManualsConverged InfrastructureServers Solution Resources

PowerEdge Product Group

Direct from

Development

Dell EMC PowerEdge R740xd Servers Demonstrate Greater

Performance with Broadcom 25GbE Ethernet Adapter

Tech Note by:

Matt Ogle

Piyush Kundra

Mike Darby

SUMMARY

Dell EMC PowerEdge product

users have become extremely

efficient at utilizing their system

hardware. To avoid wasting disk

space numerous VM’s are

assigned to drill the server with

additional tasks. Although doing

this maximizes drive usage,

performance may begin to

decrease if you are bottlenecked

by a 10GbE network adapter.

This DFD explores the extent of

degradation by comparing the

performance of a Broadcom

10GbE ethernet adapter to that

of 25GbE when dense VM

workloads are placed on Dell

EMC PowerEdge R740xd

servers. The results give

quantifiable evidence that it is

advantageous to normalize

implementing 25GbE cables for

high performance computing

data centers.

Background

An efficient system should have all pieces tailored together to produce

maximum output, with no weak links uprooting an otherwise balanced

component workload. Rapid technical advancements in one aspect of a

system can generate new bottlenecks in another.

Dell EMC recommends that Enterprise owners reevaluate their Ethernet

adapter investments to ensure networking bandwidth remains adequate.

Today the most commonly used ethernet adapter speed is 10GbE. The

next upgrade, 25GbE, boasts greater performance per dollar and is

compatible with its 10GbE counterpart. Given the cost effectiveness and

backwards compatibility, at what point does your workload outgrow

10GbE and warrant the investment to 25GbE? We decided to probe this

question further by performing a series of tests to compare the results of

each.

Test Apparatus

Numerous tests were completed to quantify the performance variance

between these two adapter speeds. These tests were run on

PowerEdge R740xd servers running VMware with Broadcom 10/25GbE

PCIe adapters acting as independent variables. The 25Gbps adapter ran

16 VM’s while the 10Gbps adapter ran 16 VM’s, 12 VM’s and 8 VM’s; a

total of 4 test groups to show the benefits of 25GbE vs. 10GbE at

different workload levels. Four types of VM workloads, ranging from

moderate to heavy,

were applied to each

group for a 2-hour

span. By the end of

the experiment

performance, latency,

CPU utilization and

throughput were

quantified and

compared.

Summary of content (4 pages)

PAGE 1
Direct from Development PowerEdge Product Group Dell EMC PowerEdge R740xd Servers Demonstrate Greater Performance with Broadcom 25GbE Ethernet Adapter Tech Note by: Matt Ogle Piyush Kundra Mike Darby SUMMARY Dell EMC PowerEdge product users have become extremely efficient at utilizing their system hardware. To avoid wasting disk space numerous VM’s are assigned to drill the server with additional tasks.
PAGE 2
Performance As previously mentioned, each configured test group applied a unique VM workload: • • • • DVDStore Fileserver Simulation on Iometer Exchange Jetstress Webserver Simulation on Iometer Each of these workloads stress the server at varying intensities giving a broad spectrum of data to analyze. Performance data collected is critiqued by completed task per time; the more tasks completed per unit of time results in greater performance.
PAGE 3
Latency The next series of tests aimed at quantifying average latency deltas between 10GbE and 25GbE. The same four workloads were run across the same four configurations. See below for comparison examples: Figure 4: Latency averages for DVDStore operations. We see that 25GbE is at parity or notably improves latency across the 10GbE VM workloads. Figure 5: Latency averages for Webserver operations. At 12 and 16VM’s the 10GbE adapter runs much slower than 25GbE.
PAGE 4
Processor Utilization Figure 6: By comparing both 16 VM averages, we can see that the lack of bandwidth at 10GbE limited processor utilization. Throughput Figure 7: In this environment, 25GbE and 10Gbe were directly compared downloading ~4TB of movie files until completed. The 10GbE ethernet cable averages about 2Gbps therefore restricting its maximum performance. On the other hand, the 25GbE ethernet caps out at 4Gbps, allowing for the movie files to complete download in 2/3 the time.