Studies On The Effects Of Errors On Network Performance

65 pts.
Tags:
Benchmarking
Network testing
Networking
Performance management
Protocol analysis
Does anyone know of a research study, lab test result, or white paper that addresses specifically the effects of errors on network response times? I've found hundreds that deal with the causes of errors, although without specifics, but none that correlate the degrees of errors on the network with measured transaction response times.
ASKED: September 17, 2005  1:08 PM
UPDATED: February 17, 2010  7:14 AM

Answer Wiki

Thanks. We'll let you know when a new response is added.

What type of errors are you looking at? CRC, oversized frames, packet loss, bad frames? Each would have it’s own set of results on any study. I doubt that any switch manufacturer would sponser any study like that. After searching the web, I agree that no one shows such a study.

Discuss This Question: 8  Replies

 
There was an error processing your information. Please try again later.
Thanks. We'll let you know when a new response is added.
Send me notifications when members answer or reply to this question.

REGISTER or login:

Forgot Password?
By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy
  • Ganzit
    I'd take a study on any type of errors, although I would prefer one on CRC errors over ATM links. My reason for seeking such studies is to aid in the setting of thresholds for error conditions.
    65 pointsBadges:
    report
  • Bobkberg
    Howdy Ganzit. I wasn't sure if you were looking for physical errors or administrative errors (or what). My personal take on what you're looking for is that because of switching designs, and improvements in that area, the overall objective in the industry has been to eliminate the possibility of individual errors affecting any network as a whole. Many switches (and relatively recent hubs) have jabber controls to partition the port if it exceeds a certain amount of noise. Similarly with CRC errors and the like, the "damage" is generally kept on the circuit from the port outwards with the objective of keeping problems like that from affecting the rest of the network. Some brands and models handle things better than others, but the industry as a whole has gone to great lengths to simply eliminate the problem (network-wide that is). That way when the errors do occur, the source can be indentified fairly quickly. What I would find interesting (having seen many such installations) is a study on how long it commonly takes to identify such things - as opposed to the network and server people simply blaming each other for the lousy performance. Many network admins have NO idea of how to watch out for error counts reported by (manageable) switches, and many others haven't (or don't know how to) justify to management the price differential for manageable network devices. Hope that helps, Bob
    1,070 pointsBadges:
    report
  • Posorio
    Hi there, I'm more interested in how networks perform. I'm a controls engineer, involved in aerospace and transport. This may interest you: http://www-personal.engin.umich.edu/~tilbury/papers/lmt99csm.pdf So I'll ask you back a question. What type of network are we talking about? Very often networks have errors at the physical layer and these are handled differently depending of the application. VoIP can function with a lot of errors, and maintain a good performance (audability etc...). A Control network won't...
    0 pointsBadges:
    report
  • Posorio
    Sorry I'm kinda busy, didn't read it was ATM, this may help. http://www.cs.ucsd.edu/~voelker/pubs/tcp-usenix94.pdf
    0 pointsBadges:
    report
  • Ganzit
    Where the question about detecting errors is concerned, and I'm referring to layer 2 crc errors, physical layer errors, and the like, my preference is to use threshold condition traps -- initiated by the router. Most Cisco routers support RMON lite, which is embedded in the IOS. By "lite" I mean that their implementation only uses the events and alerts function of RMON. I use this to monitor errors, sending traps to the NMS when a threshold is crossed. There's a hysteresis mechanism involved in the form of a falling threshold that re-arms trapping, which keeps the NMS from being inundated with alarms everytime the rising threshold is crossed. But I'd like a much more scientific way to set those thresholds than to simply use past instances where errors were known to cause problems. Then there's those cases where performance is degraded, some errors are found, and the debate begins as to whether those, for example, 39 errors in 10 minutes caused service degredation or not. Since the way errors propagate to the upper layers is largely dependent on a factors like line encoding, the particular layer 2 protocol, or even the type of application data in question, it would be good to see empirical data as gathered in a controlled test environment. Imagine a set of tests where synthetic application transactions are generated at regular intervals, with increasing levels of error conditions inserted. If you tested this with a variety of voice, data, and video applications, the setting of error thresholds for traps would be more scientific and the proverbial "finger pointing" contest (network or server) might be less prevalent.
    65 pointsBadges:
    report
  • Posorio
    I've worked on the Alma platform (Alcatel Management) and in the case of switched telephone networks there are simple ways of measuring the performance of a network and raising alarms. Answer to Seizure Ratio, Answer to Bid Ratio etc... The issue we are facing with non circuit switched networks is the performance of the system becomes dependant on the performance of the application using the network. The discussion is interesting but I doubt it has a simple answer. I would say it's very dependant of the type of application. This is why for "life-critical" applications a "well policed" network is prefered. Emergency telephones are still most of them circuit-switched, and token-ring or master-slave protocols are prefered for control networks. They either work or they don't...
    0 pointsBadges:
    report
  • Ganzit
    Here's the name and authors of an excellent white paper on the subject. Chasing Errors through the Network Stack – A Testbed for Investigating Errors in Real Traffic on Optical Networks Andrew W. Moore, Laura B. Jamesy, Madeleine Glickz, Adrian Wonfory Ian H. Whitey, Derek McAuleyz and Richard V. Pentyy University of Cambridge, Computer Laboratory, andrew.moore@cl.cam.ac.uk yUniversity of Cambridge, Department of Engineering, Centre for Photonic Systems flbj20, aw300, ihw3, rvp11g@eng.cam.ac.uk zIntel Research, Cambridge, fmadeleine.glick, derek.mcauleyg@intel.com
    65 pointsBadges:
    report
  • Ganzit
    Hey, Bob. I think my handle somehow got mixed up with the someone else's. Although it wasn't my question, I think your answer is enlightening..
    65 pointsBadges:
    report

Forgot Password

No problem! Submit your e-mail address below. We'll send you an e-mail containing your password.

Your password has been sent to:

To follow this tag...

There was an error processing your information. Please try again later.

REGISTER or login:

Forgot Password?
By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Thanks! We'll email you when relevant content is added and updated.

Following