MPLS vs Point-to-Point over Citrix Performance Difference

Question

We run Citrix at our remote locations. We have two circuits at each location. One is a point to point for backup and the other is an MPLS circuit that is our primary. Both circuits are T-1 speeds.
We have all thin clients at our remote locations. When communication goes through the point to point circuit it utilizes much more bandwidth. Maybe 1.2 Mbps on average but when we communication through the MPLS circuit only 800 Kbps of bandwith is actually being used.
Can anyone explain this? I was thinking that maybe MPLS does a faster job of switching the packets across the WAN and that the Citrix does not need to use as much bandwidth because of this. This analyis was completed across all of our sites and in each case Citrix uses more bandwidth on a point-to-point vs an MPLS circuit. I have not had any users complain when accessing either ciruits.
What do you think is causing this?

Answer 1

Hi,
I fully agree with Swaroop. Being an instructor teaching many MPLS classes I was frequently confronted with the opinion MPLS is "faster" as it is "switching". This is not true and I always countered this - provocantly - stating that MPLS is reducing throughput, so it is slower! What I mean writing this: given a certain topology for IP forwarding and turning on MPLS on it will increase the overhead (additional overhead by adding labels) and thus reduce end to end IP throughput. The lookup is done by the same algorithm (CEF) at wire speeds for IPv4 and labeled packets - there is no speed gain for either technology.
Do not get me wrong, this does not mean MPLS is "bad" and in fact the difference between pure IPv4 forwarding and MPLS forwarding is marginal and most likely irrelevant for any real environment. The advantages of MPLS are plenty and thus a marginal throughput difference is not the most important thing to consider.
I guess the idea of "switching is faster than routing" stems from the fact that there were times, when IPv4 forwarding ("routing") was done in CPU, thus was slow, whereas L2 forwarding ("switching") was done in hardware and thus was faster. It dates back to those days where we used AGS+ (an old router, which is EOS, EOL and most likely even EOeBay ;-) and f.e. Cat5000.
Now coming back to the observed behaviour in the original post there might be some reasons to explain it:
1) different L2 overhead as pointed out by Swaroop, especially as I would assume rather small average packetsizes.
2) Additional traffic on the P2P link not sent through the MPLS cloud - check your routing, if it is exactly the same for both links.
3) Measurement artefacts - as Swaroop pointed out. Is the load interval the same for both interfaces? I would rather use a packet analyzer than only go for a "show interface" to get precise values.
Hope this helps!
Regards, Martin