Communications layer quality requirements.TCP/IP on a VPN 128kbps connection. Latency. Packet loss

[danielmaxwell]

We are planning installations in parts of the world where quality and noise might not be good. We need documentation or standards to show internet service providers what level of latency or packet loss (jitter, noise, etc...) we are not able to accept in order for our VPN connections at 128kbps actual data throughput to work.

We expect to use standard TCP/IP, HTML, small FTP files, etc...

Please point us in the direction of documents or white papers that talk about standards in this area. This is not to troubleshoot Cisco hardware but to use Cisco hardware in places where we need to troubleshoot the ISP, not the Cisco equipment.

Here are some expected expectations:

• Web pages of 1MB typical in size will load within 60 seconds without a timeout.
• VPN connected longer than 30 minutes.
• VPN will be able to reestablish within 30 seconds.
• Alerts of 50kB are emailed or posted to a server every hour.
• Data usage per site ranges from 10MB to 100MB per month with a cap of 500MB per month.
• Potential for 2 simultaneous VPN tunnels.

INFORMATION REQUESTED:

1. What quality standards in terms of packet loss rate, jitter, latency, and BER should we require for the following:
   1. IPSEC VPN
   2. 128kbps maximum throughput of communications layer 1 link (128kbps IDSN for example).
   3. UDP and TCP/IP most of the time with sparse FTP requests.

[Philip D'Ath]

I've done a bit of work with VPNs and satellite links, including some low bandwidth connections like you mention.

First to reset one specification above - you will never be able to download a 1MB web page in 60s over a 128Kb/s link.  Not even close.  It is technically not possible.

On the whole, it is usually a question of how well the TCP stack handles the link parameters rather than IPSec - which is OS specific.

[Philip D'Ath]

If you use a Cisco router at each end of the link you could also consider using "Rate Based Satellite Control Protocol".

http://www.cisco.com/en/US/docs/ios/interface/configuration/guide/ir_rt-bsd_sat_support_TSD_Island_of_Content_Chapter.html

Rate Based Satellite Control Protocol (RBSCP) was designed for wireless or long-distance delay links with high error rates, such as satellite links. Using tunnels, RBSCP can improve the performance of certain IP protocols, such as TCP and IP Security (IPSec), over satellite links without breaking the end-to-end model.

[Philip D'Ath]

Another option you should also seriously consider is using Cisco WAAS.  This guide talks about the steps needed to optimise the circuit for a higher Bandwidth-Delay-Product (BDP) link.

http://www.cisco.com/c/en/us/td/docs/app_ntwk_services/waas/waas/v4013/configuration/guide/cnfg/policy.html

Using Cisco WAAS is probably the only way you'll be able to download a 1MB web page in 60s over a 128Kb/s link - maybe.  You really are asking more than is reasonably likely to happen.

[danielmaxwell]

Thank you, Philip. We would like to know specific packet loss rates, latency etc... where even Rate Based Satellite Control Protocol (RBSCP) will break down. By the way, that link on RBSCP is quite useful. Thank you.

We need some numbers. We must be able to troubleshoot with Internet service providers when connections fail and we need some acceptable standards in hand.

Sorry, I was not correct for 1MB / 60s. I timed about 850kB in 90 seconds over a GSM cell phone gateway. You are right on that. Sorry for that blunder.

We need to keep trying to find out at what levels of "noise" in the line do we "fail" even if extra techniques such as RBSCP are used. I try to define what "fail" means but for now it means a web page timing out with a reasonable amount of data that it takes to load in today's average web sites and equipment embedded GUI pages with data gauges and bar graphs etc...

Is there such a thing as a "MOS (mean opinion score)" for TCP/IP web pages? If so, we are searching for a fairly high opinion score since managers with little time to waste want to see fairly responsive web pages loading and no timing out.

Are there additional places or documents to review to help get these answers?

[Philip D'Ath]

Its hard for me to give specific numbers - as the point at which the system fails is far different to the point where a user would consider it unacceptable.  TCP is very robust at handling problem circuits.

Could I perhaps suggest using a free WAN emulator, and try different delay and loss parameters, and then determine for your specific case what is "acceptable".

http://wanem.sourceforge.net/

Otherwise, plucking numbers out of thin air, I wouldn't like to use a circuit with these parameters (and I would personally call it a fail):

• Latency greater than 2000ms
• Packet loss greater than 10%