ABCs of QoS

If there is a more appropriate forum for this post, please point me in the right direction. But you guys are generally such a font of knowledge, I thought I'd start here.

Some of our users use a web-based application, and they are constantly complaining of sluggish performance and lock-ups. It is a Citrixy, terminal services kind of app, and the vendor recommends something like 24k - 36k of bandwidth available per user. We have an aggregate of over 6MB internet bandwidth, 3MB from one provider (say Sprint), and 3MB from another provider (say AT&T), going through a load balancing device that does round-robin distribution.

We have used various measurement tools and have determined that the problem does not seem to be bandwidth. We never "peg-out," and there always seems to be bandwidth available according to our measurement tools.

We told our application vendor this, and they replied that besides bandwidth, they require a minimum QoS of 90%. They pointed us to a site called where we can run bandwidth tests. Sure enough, the QoS readings from this site are consistently less than 90%. We have also isolated the circuits, and interestingly we found that AT&T consistently gives poorer results than Sprint. With 100 samples for each circuit, AT&T averages roughly 40% QoS, while Sprint Averages 80%.


  1. Are there other ways to measure QoS for Internet access? I don't know how reliable "InternetFrog" is. If I could at least compare it with a few other sites, I'd feel more confident before taking this to our ISPs.

  1. Is there anything we can do on our end to improve QoS? Our path through the Internet goes through a Cisco switch, a Cisco router, a CheckPoint firewall, and a load balancer. I'm not real sure which one might be tunable to provide better QoS.


Reply to
Loading thread data ...

If you remove the load balancing device and have a session stick to one circuit, does that improve matters? How many users do you have? Do you actually have 24-36k per user?

Are you monitoring your circuits to keep a record of utilisation? This may be useful to correlate poor performance with circuit utilisation.

It's not obvious from InternetFrog how they measure QoS - I've just ran a test and it says my 15M DSL gets 922kbps through with 33% QoS. I would take it with a pinch of salt. 33% QoS obviously doesn't mean 67% packet loss, as that would be insane.

Ping is the simplest. For example you could run a 100-repeat ping to the same destination from either circuit and compare packet loss.

If you're not maxing out the link at your end, then I can't really see what you can do about it, although it may be worth implementing QoS for when you inevitably do run out of bandwidth.

Reply to

If you are mixing time sensitive applications with regular internet usage without QoS set up on your provider's edge router, you will have problems. You can expect to experience glitches with even "apparently" low bandwidth usage. How granular is your monitoring software? Typically, the bottleneck is at the egress interface of the provider to the customer, assuming your provider is a top level ISP...and yours are.

Those QoS metrics from the frog site are dubious. I would recommend doing a traceroute to the site and compare the path through both providers as well as get some info on how those numbers are derived.

Reply to

i think you need you app supplier to define what they mean by QoS in terms of things you can measure - common ones would be end to end latency, packet loss, and jitter.

that way you should be able to instrument the client and get some stats that you can correlate against "good" and "bad" user experiences, or maybe come up with a simple test script that throws packets around.

Note that the killer is often latency since you dont have much control over how far a packet has to go to be useful in most cases - each packet has to travel between the 2 end points which are usually fixed, and all you can do is play around with things like choice of ISP that affect it indirectly.

you may just have an app that doesnt work properly in your environment - maybe it needs the stability of a private network where you can use QoS directly, or maybe it just does not work effectively over the required distances.

looking at the site it is intended for "voip" testing.

however - they give round trip time, upload / download speeds and max pause time - which are much easier quantities to check elsewhere than the "QoS %age they do not define.

qos is all about controlled unfairness of forwarding for some packets. The only bit you can control directly is outbound forwarding.

But you can make sure that the return traffic doesnt hit any unnecessary bottlenecks as it enters your central site..

a lot depends on what they are measuring - but you might want to put your "citrixy app" thru the best internet connection, and push everything else on

Reply to


When you say this is a 'Citrixy, terminal services kind of app' do you mean that you are publishing it through Citrix?

If there is such a high QOS requirement, you may want to consider changing the delivery method to distributed meaning that the web browser runs directly from the client machine as opposed to the Citrix/Terminal Server. Then you can look at products similar to Citrix NETScaler/WANScaler to apply a more efficient level of control to the data stream. You should be able to trial the setup to see if it will suit your needs before making any kind of financial investment.


JohnD wrote:

Reply to
Mike Silverman Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.