Gauge network traffic for each Citrix session

Question

We are currently reviewing the bandwidth of our WAN links. How much bandwidth does a "typical" Citrix session utilize over a WAN link? JFYI - we are using Citrix Program Neighborhood V10 and each session is configured to use 256 colors.

I have set up PRTG and it appears that for a server hosting 4 users, the traffic is approximately 100k to 300k per session. Is that about right?

If you had to set a benchmark on a per-user basis, how much bandwidth would you assign to each user?

Thanks in advance.

Solution 1

That seems too high. I would have expected about 100k max and more like 60k. Are you able to log in to one of the Citrix sessions yourself? If so the client can show you stats on bandwidth. Similarly at the server end you can see this, for example in Windows counters and possibly in the zone data collector (that's from foggy memory though).

What kind of applications and window resolutions are in use, and how many at once? Or is it published desktop sessions? Are there additional local peripherals or other custom session traffic?

Solution 2

XenApp should NEVER use 100-300kbps for general usage. Printing would increase outbound usage from your servers and scanning with a scanner or using other USB peripherals with input other than typical XenApp activity would be the only way you should even come close to 100kbps.

Solution 3

100-300k seems a little high using that color level. Are you just noting the total amount of bandwidth used by the server and dividing by four? You might want to account for other traffic from the box. Edit: I'm assuming you're monitoring the switch port or the server NIC with PRTG? Do this: Monitor the traffic with no users. Write that number down. Add one user. Write that number down. Repeat. Find out what the increments are. This number will be inaccurate (in fact, very misleading) if the users' applications within the session generate LAN traffic : for example, if they're running a SQL client that connects to a local SQL server, or a local file share, that's not internet-bound traffic

You could measure the bandwidth used by ICA (TCP 1494) at the gateway if your router supports that kind of monitoring. That gives you an immediate and correct answer for total bandwidth.

What problem are you trying to solve? If you need to minimize bandwidth, Citrix has a lot of options for defining policies, as well as optimizing different parts of the ICA protocol. What version of Citrix are you running? If you're running a recent version at the highest-level edition, it has SNMP support for monitoring, the answer may be right in there and you could monitor that with PRTG.

Comments

[molecule]

Hi, Thanks for your reply. Yes, since I do not have any other method of estimating bandwidth, I am currently just noting the total amount of bandwidth and dividing by four. Completely agree that other traffic need to be accounted for as well. How much would you allocate for each session?

[molecule]

Yes, I currently have PRTG monitoring the switchport that the server NIC is connected to. I understand it is a bit difficult to capture the exact number (given my limited knowledge and tools) but was looking more for a ballpark figure. I am also not completely familiar with Citrix. As indicated, Program Neighborhood is v10 and MetaFrame Presentation Server is v4.0. As for my "problem" - we have a 6Mb point to point link which we are planning to downgrade as it appears to be under-utilized and I was going through this exercise just to make sure that I have accounted for Citrix traffic.

[mfinni]

If that's for rdp, not ica, then you're not making a valid comparison for what the question is asking.

[mfinni]

Well OK then, you should definitely read up on ica and citrix, and what you can do with them (and what they do out the box)

[joeqwerty]

@mfinni: C'mon. My answer gives the OP some info to go on and also suggests a valid method for measurement. Yes, ICA is "supposedly" less bandwidth intensive than RDP, but my answer gives the OP good info. While my answer doesn't specifically address ICA, it's hardly worth a downvote. Me thinks you're too liberal with the down button. Ease up, mate.

[joeqwerty]

@mfinni: Can't get an accurate measurement? Why not? It's not difficult to install PRTG on the server, set up a packet sniffing sensor, and filter your capture for just ICA traffic. Connect one user, capture traffic, filter for ICA, and bada-bing, bada-boom, you have your ICA bandwidth mesaurement per user. What's so hard?

[joeqwerty]

@mfinni: Note that I didn't downvote your answer, even though your suggested monitoring method is prone to producing innaccurate, misleading results. If the OP wants to know exactly how much bandwidth is consumed by a Citrix session then the best way is detailed in my previous comment.

[mfinni]

@Joe- I didn't know that you can filter for protocol types using PRTG, but that will give the same results as my suggestion for filtering on protocol types at the gateway. So, cool, he can do it with the tools he's already got.

[mfinni]

it's not "supposedly", at all - out of the box to begin with, and there's a lot of optimization and/or throttling you can do. You can optimize ICA to the point that it uses about 10-20k per session. It was created when common connection methods used phone lines - you used to be able to dial-in, not even over PPP, to a Citrix server, and could host 3-5 sessions per phone line at 56k.

[molecule]

Guys, excuse my ignorance but I am only using the free version of PRTG and as far as I know, it only gives you bandwidth in/out measurements. Joe, thanks for your feedback. What sort of packet sniffing sensor are you talking about? something like cacti or nbot? I downloaded these packages but have no clue in the world how to install them on a Windows box. Seems like I need an SQL DB and an Apache/IIS service running. mfinni, thanks for your posts. I guess my problem is that I know almost next to nothing about ICA and Citrix. Will need to read up more on that.

[molecule]

In our typical setup, we have a farm with 3 servers hosting a single published application. Whenever we set it up for the user, we use the Basic encryption, 256 Colors and Window Size is 640x480. No sounds are enabled but printers are mapped. Apps are finance related and data refreshes are very frequent.

[joeqwerty]

@mfinni: I've read the hype many, many times about how great ICA is and how much better it performs vs. RDP, but I've yet to see definitive, unbiased evidence to that effect. If you could point me to such evidence, I'd appreciate it as I'd like to finally put the question to rest for myself. Thanks.

[joeqwerty]

@mfinni: In the properties of the packet sniffer sensor, go to the channels and select only the channels (protocols) that you want to display. The sensor will capture all protocols, but only display the protocol that you've selected. The reason I would do it on the server and not the router is that you have to set up a Netflow sensor in order to see the protocol distribution in the traffic coming from the router, which requires the addidtional step of configuring the router. You also run the risk of capturing "stray" ICA traffic from, or going to, somewhere else that might skew the results.

[joeqwerty]

@molecule: Keep looking at PRTG. It has the ability to create a packet sniffer sensor. Look here at the section on Device and Sensor setup: paessler.com/manuals/prtg7

[joeqwerty]

@molecule: Just so you know, no actual application data ever passes through the ICA (or RDP) session. The only "data" that passes through the session is print data, drive mapping data, sound data (if sound is being redirected), mouse clicks, key strokes, and screen draws\refreshes, in a nutshell.