end to end qos

followurself · ‎07-24-2009

hello

will appreciate if someone can help to clarify

this is the set up

site a router-- isp routera---mpls cloud---isp router b---site b

the site a router lan interface does marking for citrix traffic as precedence 3.. isp router a does shaping for both precedence 3 and 2. the isp router a see citrix traffic marked as 2 treats it shaping forwards it..isp router b sees it as 2 forward it to wan interface of site b router. site b router wan interface route it to its lan interface and then to the citrix server. on reply back from the citrix server for the same packet , it comes as precedence 2. but at site b router lan interface , policy is as such any citrix traffic mark it as 3.

does the router remark the packet

will this cause a delay

the issue is citrix client at site a are facing slow response from citrix server at site b.

will this remarking cause the problem. the isp router b is configured to understand both 2 and 3 and give shaping.

thanks

u1kumar2002 · ‎07-24-2009

I was also having same problem for slow response from citrix.

I had done a qos with dscp & CBWFQ.

Just define the servers in a access-list, Call that ACL in Class map & set the dscp value of af 41 or anything which u feel is appropriate.

Applying that acl in POP end (ISP router site B).

It has to work fine. the problem is low high utlisation of other bandwidth hungry applications due to which citrix will not run properly.

Or

you can define specific bandwidth for that class map. & fair queue for class default

apply on ISP End ( POP End). Then also it will work. The reservation will be only work at the time of congestion. it will be not all time.

u1kumar2002 · ‎07-24-2009

I was also having same problem for slow response for citrix application in locations where congestion problem was more.

I had done a qos with dscp & CBWFQ.

Just define the servers in a access-list, Call that ACL in Class map & set the dscp value of af 41 or anything which u feel is appropriate.

Applying that acl in POP end (ISP router site B).

It has to work fine. the problem is low high utlisation of other bandwidth hungry applications due to which citrix will not run properly.

Or

you can define specific bandwidth for that class map. & fair queue for class default

apply on ISP End ( POP End). Then also it will work. The reservation will be only work at the time of congestion. it will be not all time.

u1kumar2002 · ‎07-24-2009

I was also having same problem for slow response for citrix application in locations where congestion problem was more.

I had done a qos with dscp & CBWFQ.

Just define the servers in a access-list, Call that ACL in Class map & set the dscp value of af 41 or anything which u feel is appropriate.

Applying that acl in POP end (ISP router site B).

It has to work fine. the problem is high utlisation of link by other bandwidth hungry applications due to which citrix will not run properly.

Or

you can define specific bandwidth for that class map. & fair queue for class default

apply on ISP End ( POP End). Then also it will work. The reservation will be only work at the time of congestion. it will be not all time.

followurself · ‎07-24-2009

Thanks for the response

we are going to change it to very similar kind what you have mentioned

prior to this we need to explain why it has been 2 at one end and 3 at other router. because both isp router understand 2 and 3 it shd not matter.

but if site b router sees 3 and then remark it looking at the class , is this is the reason of the current problem

Laurent Aubert · ‎07-24-2009

Hi,

If your SP treats IPP 2 and 3 the same way, the remarking should not be an issue.

Do you allocate enough Bw to your citrix traffic in case of congestion ?

Maybe the shaper value are too aggressive and delay many packets. You should check with your SP if the shaping is always on or not.

HTH

Laurent.

Joseph W. Doherty · ‎07-24-2009

Slow Citrix response is most likely due to latency. From the information you provide, it's very difficult to say how remarking would impact your traffic.

For example, if only these two sites were in communication across the "cloud", then you might not even need packet ToS markings. If there are more than two sites involved, then markings, within your MPLS provider's QoS model, can be crucial.

You also mention shaping of IP Precedence 3 and 2 traffic, but if there's any other traffic, shaping could work against you. Normally you would shape for far side bandwidth, assuming it's less then near side, and then allocate bandwidth/priorities for all possible traffic.

Also with Citrix, you want to treat the remote terminal (screen scraping) traffic almost like VoIP, but Citrix printing or disk-to-disk copying traffic should not be given the same high level treatment. (NB: The later Citrix protocols include a Citrix traffic type that can be analyzed using the later NBAR variants.)