bench marking servers guide

8/6/2019 Bench Marking Servers Guide

http://slidepdf.com/reader/full/bench-marking-servers-guide 1/3

TEST NETWORK PERFORMANCE:

NB: User should perorm network benchmarking and tuning in order to

get the maximum throughput or iSCSI SAN environment. The tests need

o be perormed beore putting StarWind iSCSI SAN into a live production

environment in order to have pure test results and avoid environment’s

econguration.

StarWind Sotware has no strict recommendations or the network

benchmarking utilities which should be used, however the most common

network benchmarking tools according to StarWind Sotware’s clients are

NTTTCP and IPer. The most common disk benchmarking tool is IOMeter.

The diagram bellow shows an optimal and recommended conguration or

StarWind HA based SAN inrastructure.

Logical understanding o networking assumes that all links should have almost

dentic perormance; otherwise HA SAN will work at the perormance o the

lowest link.

Connect the RAM disk device to the server via the link you would like to

benchmark.

Next Step is to measure the perormance o the device.

Use IOMeter rom the client side, choose the ollowing Access Specicat

64kb, 100% Write, 100% Random.

Results less than 80% o the links saturation (even i only one) will not be

suitable or HA implementation. StarWind HA does not use all the availab

bandwidth. In case o bad network test results you should x the link be

implementing HA.

Send and receive tests need to be perormed on each server.

Please reer to the appropriate vendor technical documentation when u

network benchmarking tools.

SCSI COMMANDS THROUGHPUT TESTING:

Once you have achieved optimal results you can proceed to next step

which is checking the SCSI commands throughput. This can be perorme

by measuring the perormance o an iSCSI connected RAM device. RAMdisk is based on the server’s local RAM thus it will have the perormance

exceeding the network speed, as a result user can easily see what is the

perormance o the iSCSI throughput or the particular link.

First, create a RAM disk device in StarWind management console by ope

Add target Wizard and going Hard Disk →Basic Virtual.

FOLLOWING LINKS SHOULD BE CHECKED:

1. Synchronization link between StarWind 1 and StarWind 2.

2. The link between StarWind 1 and client 1. 3. The link between StarWind 1 and client 2.

4. The link between StarWind 2 and client 1.

5. The link between StarWind 2 and client 2.

StarWind is SAN Software to build a SAN for VMware and SAN for Hyper-V

Storage for Business Continuity in Virtual Environment

Benchmarking ServerS guide



StarWind Support team can provide you with preset IOmeter conguration le.

Same test needs to be perormed on Client 2.

• IOmeter test results charts should not be lower than the results o the

network benchmarking.

MAGE FILE DEVICE TESTING:

Once you have reached optimal results with RAM disk device you can

checkthe perormance o an image les target.

First, create an Image le device:

Open the Add target Wizard and go to Hard Disk →Basic Virtual→Image File

device→Create new virtual disk.

Image le device used or testing needs to be at least 2GB. Device cac

mode to use or benchmarking can be normal or write-back. Write-

caching will increase the perormance.

Once the device is created connect it to client 1.

Use IOMeter to test the perormance. Access specications are ollow

16kb, 32kb, 64kb; 100% Read, 100% Write, 50/50 Read/Write; 100% Seque

100% Random.

Same test needs to be perormed on client 2.

• Tests results o the write-back cached image le device

are usually an 80-95% speed or the GbE channel.

I using 10GbE network interaces the perormance will be 110-120%

the disk array’s perormance or 80-95% i the disk array is aster than

network.

Depending on the network, disks perormance and conguration, the Im

le device perormance can be less o the network perormance.

HA DEVICE TESTING:

For creating HA device please reer to the appropriate technical docume

in the embedded help le.

Following tests are based on act o using a recommended conguratio

shown on the diagram in the beginning o this document.

HA device perormance directly depends on your network and disks.

HA device perormance will not be higher than your network and disks

perormance.

MPIO policy on the client side directly afects the HA device perormanc

“Round Robin” provides better results than “Failover only” (xed path).

Caching highly increases the perormance. It is recommended to use aminimum o 512MB or cache, in write-back mode (increases write spee

in write-through mode (increases read per ormance).

To test HA speed a customer has to mount the HA volume to the client

server and run test rom and to this VMA.

StarWind is SAN Software to build a SAN for VMware and SAN for Hyper-V

Storage for Business Continuity in Virtual Environment

IN CASE OF SIGNIFI CANT DIFFERENCE YOU SHOULD CHECK

WHATCAN BE CAUSING THESE:

• Firewalls

• Jumbo rames

• Antivirus

• Improper Stripe Size in RAIDs.

bench marking servers guide

Documents