Home


About Network Testing Labs

Contact Network Testing Labs

Independent Reviews of Network Hardware and Software

 

NETWORK TESTING LABS REVIEW

Expand Networks’ Accelerator Appliances Race Past the Competition


 

 



In strenuous testing, Expand’s accelerator series of appliances emerged as the fastest way to add significant horsepower to latency-challenged satellite links.
By Barry Nance


Executive Summary
Expand Network’s Accelerator WAN Optimization appliances, running the COMPASS Accelerator Operating System, dramatically outperform the competition. In our tests, the Accelerator appliances consistently and reliably sent the most data across high latency, low-bandwidth satellite connections. The appliances compensate for satellite transmission delays, increase both effective bandwidth and throughput and deliver maximum network efficiency.


Pulling applications and data through a satellite link is very much akin to Chinese water torture. The slow “drip, drip, drip” of network traffic is agony to users who desperately need to get work done. The Satellite Service Provider environment is not exactly conducive to productive, efficient workflow.

Transmissions to and from a satellite take too long, classic WAN link bandwidth categories (OC-3, T-3, T-1, etc.) don’t apply and network costs are stratospheric.

The technical limitations of satellite communications present unique challenges – and opportunities.  Most commercial communications satellites orbit the earth geosynchronously, about 22,300 miles above the equator. The time it takes for a TCP/IP packet to travel 44,600 miles source-to-destination (22,300 miles up to the satellite and then 22,300 miles down to the target) wreaks havoc with the TCP/IP protocols. The delays cause application timeouts, jitter, excessive retransmissions, disconnections and loss of synchronization. The net effect: IP networks choke, TCP window sizes shrink, VoIP calls garble and database transactions fail.

In 2001, the US Department of Defense formally adopted a vendor-neutral solution that it could apply to any DoD network: the Space Communication Protocol Standard (SCPS). SCPS has satellite-friendly protocols for file handling, communications sessions, security and link management. Unfortunately, SCPS doesn’t go far enough.

Performance Enhancement Proxy (PEP) technology can offer some help. A pair of PEPs at either end of a satellite link trick each local network into believing the remote, satellite-linked network is right next door (TCP/IP is geared toward such adjacent networks). PEPs can mitigate the effects of latency, help fill the link with data and improve network performance. However, bundled satellite modem PEPs are constrained in their capabilities, and they deliver limited results.


The ideal solution is a PEP that uses SCPS as the underlying acceleration scheme and that both implements SCPS in an extremely efficient manner and also enhances SCPS protocol handling in specific vendor-designed ways. Some manufacturers employ data compression, data caching and application–specific acceleration to help overcome satellite link delays. The result is a PEP that helps fill the network link nearly full, so you get virtually all the bandwidth you are paying for. It helps data flow smoothly, mitigating high latency and jitter. And the ideal PEP makes applications more responsive.

To determine which PEP you should be using, we evaluated pairs of Expand Networks’ Accelerator 4900 series, iDirect Network Accelerator Series 1200, Packeteer SkyX 750, ViaSat xPEP and Comtech EF Data TurboIP devices, both in our Alabama Lab and at real-world customer sites.

The Expand Networks Accelerator 4900 series appliances proved to be far better at optimizing WAN links of all speeds, and they also resulted in lower latencies for the packets traversing those links. The Expand Networks 4900 series Accelerator devices win the Network Testing Labs World Class award for Best Satellite Link Performance Enhancement.

WAN Link Acceleration Results
In our tests (see the “Testbed and Methodology” section), we funneled several different application protocol streams, singly and in various combinations, through the devices. The following table summarizes the relative performance benefits of the WAN link optimizers for different protocols.




 

Performance Optimization Factor
(Data Transmission Speed Multiplier)

 

Traffic
Type

Expand

Accelerator 4900 series

iDirect

Network Accelerator Series 1200

Packeteer

SkyX 750

ViaSat

xPEP

Comtech EF

Data TurboIP

FTP

6.2

5.1

5.2

5.4

5.4

File Sharing

-       CIFS

-       NFS

 

6.8

6.0

 

5.0

4.7

 

5.4

5.0

 

5.5

4.8

 

5.6

5.0

HTTP

6.5

4.5

4.9

4.8

4.9

Server-based (Citrix, VDI, Terminal Services)

 

 

7.6

 

 

5.8

 

 

6.2

 

 

6.0

 

 

6.5

Web Services (SOAP)

 

3.0

 

2.0

 

2.2

 

2.0

 

2.2

Database

-       Oracle

-       SQL Server

 

4.6

4.9

 

2.9

2.9

 

3.1

3.2

 

3.0

3.1

 

3.4

3.5

Backup/ Restore

8.2

5.9

6.1

6.0

6.6

E-Mail

-       Notes

-       Exchange

 

5.3

5.7

 

3.6

3.7

 

3.9

4.1

 

3.9

4.2

 

4.2

4.2

Table 1. WAN Acceleration Results.

In our tests, we also found that the Accelerator 4900 series devices did a better job of reducing latency and jitter – important criteria for real-time traffic.

We obtained these results across both real and simulated Satellite links data transmission rates of 64 Kb/sec, 512 Kb/sec and 1.5 Mb/sec (full T1).

The performance optimization factors denote by how much the optimization devices improved the performance of the WAN link. For example, a WAN optimization factor of 2 would mean the data traversed the link in half the time it did in un-optimized form. Or, expressed a different way, a performance optimization factor of 2 means the optimization device made the link carry data twice as fast.

In each test for each traffic type or mix of traffic types, we first determined the maximum amount of data the link would carry without using the optimization devices. We then added the devices to either side of the link and retested by sending exactly the same data through the link. We ran a series of tests, with each series representing a business day’s activities.


Each series of tests simulated a specific business-oriented IT operation. For example, the FTP tests transferred 1 Mb, 5 Mb and then 10 Mb files from one site to another, with each transfer representing typical end-of-day activity. Each “day,” we edited the files to mimic the kinds of changes a organization might typically make during a day.

Similarly, for the backup/restore tests, we transacted with a relational database on a “daily” basis and then used backup/restore software to make a backup copy of the updated database.

The file sharing results show the performance improvements for both Sun NFS and Windows protocol activity. Each series of Web site, Web Services, database and e-mail activities represented a separate and unique day’s worth of network traffic.

The following table shows the Mean Opinion Scores (MOS) we obtained when we used the WAN links to carry VoIP traffic. In each test, we made the link busy by devoting 75% of the link’s bandwidth to UDP packets.

Network Instruments Observer network analysis software calculated the MOS values for each set of tests.

VoIP Performance

Mean Opinion Score (MOS)

Expand

Accelerator 4900 series

iDirect

Network Accelerator Series 1200

Packeteer

SkyX 750

ViaSat

xPEP

Comtech

EF Data TurboIP

(score = 2.9 with no PEP in use)

        1 call

      
5 calls

      
10 calls

 

 
4.0

3.8

3.8




3.5

3.3

3.2


 

3.6

3.4

3.3

 
 

3.4

3.3

3.3



3.6

3.4

3.3

Table 2. VoIP MOS scores.


Finally, the table below identifies the optimization techniques employed by each of the devices we tested:

 

ExpandThe following figure illustrates the satellite-based network over which we exercised these five products.

Accelerator 4900 Series

iDirect

Network Accelerator Series 1200

Packeteer

SkyX 750

ViaSat

xPEP

Comtech EF

Data TurboIP

SCPS Implementation

X

 

 

 

X

Reno Congestion Avoidance

X

 

 

 

 

Vegas Congestion Avoidance

X

 

 

 

X

Dynamic Bandwidth Adjust

X

 

 

 

 

Fast Start

X

X

X

X

X

Automatic Window Scaling

X

X

X

X

X

Flexible Management Interfaces

X

X

X

X

X

Application Acceleration

X

 

 

 

 

Caching

X

 

 

 

 

Compression

X

 

 

 

 

Quality of Service

X

 

X

 

 

Packet Fragmentation

X

 

 

 

 

Table 3. Optimizations exhibited by each product.


Conclusion
Without question, the Expand Networks model 4900 series Accelerator turns a satellite link from a necessary evil into a useful, productive data communications medium. The model 4900 series reduces application response times and reduces costs. Its high-tech management of the satellite link dramatically accelerates traffic, and the devices are reliable, scalable, easy to install and easy to use.




Testbed and Methodology
Our test environment consisted of six Cisco- and Juniper-routed Fast Ethernet subnets linked via real and simulated satellite links. The satellite links had bandwidth settings of 64 Kb/sec, 512 Kb/sec, and 1.544 Mb/sec (full T1). In our simulations of satellite-based links, we injected 650msec latency to represent a single satellite hop.


The following figure illustrates the satellite-based network over which we exercised these five products.



We used each device’s “out of the box” preset configuration options, and we made sure that all optimizations were enabled.

Our client and server platforms included Windows NT/XP/98/2000/2003/Vista, Unix (AIX), Red Hat Linux and Macintosh OS X. Relational databases on the network were Oracle 10, Sybase Adaptive Server 12.5 and Microsoft SQL Server 2005. Windows Server shared files, while Internet Information Server (IIS), Netscape and Apache software served up Web pages.

Our tests confronted the optimization tools with several different kinds of network traffic, singly and in various combinations. The network traffic types included file sharing protocols, FTP, Web pages, Web Services Simple Object Access Protocol (SOAP) transactions,

relational database transactions, backup/restore material and e-mail. Each series of tests simulated a specific business-oriented IT operation or set of operations. We tested FTP transfers, with each transferred file a slightly-changed version of the previous test's file. We performed backups and restores of a relational database that had experienced a day's worth of transactions and updates. We shared program files and data files, and we tested with Web traffic, Web Services (SOAP/XML) traffic, database transaction traffic and e-mail traffic.

We evaluated for fast performance, support for any and all protocols (including streaming video, VoIP, routing table updates and even music-sharing protocols), ease of use, scalability, good documentation and trouble-free installation.

Copyright 2012 Network Testing Labs


  
Home

About Network Testing Labs

Contact Network Testing Labs