Re: [Users] Service Chaining for vCenter-Only in contrail3.1

[Harshad Nakil]

It is not needed in service chasing V2. Launching of the VM with correct 
interfaces is not contrail responsibility in V2. 
Users get more control over how the VMs are launched in V2. 

Regards
-Harshad


> On Oct 17, 2016, at 7:06 PM, fubinghao686  wrote:
> 
>i find a new speciality in contrail3.1 which is service chaining for 
> vcenter-only,the original text in contrail3.1 release note is below:
> Service chaining for Layer 3 services for vCenter-only systems is supported, 
> only by means of port-tuples.
> 
> This feature allows the user to configure service chaining Version 2 in the 
> vCenter-only mode of operation. Both in-network and in-network-nat service 
> chains are supported. Not supported is service chain Version 1.
> 
> when i finish install contrail3.1 of vcenter version,i create service 
> template,i can't find the place of img name, the page screenshot is below
> 
> <截图1.png>
> 
> could anyone can help me why 
> 
> 
> 
>  
> 
> ___
> Users mailing list
> Users@lists.opencontrail.org
> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] DPDK vRouter failling to start

[Harshad Nakil]

This lists is for open Contrail users. If you are using juniper Contrail than 
you need to contact juniper support. 



Regards
-Harshad


> On Oct 14, 2016, at 6:32 AM, shyam yeduru  wrote:
> 
> Hi Team,
> 
> We want to check the following combination in Juniper Contrail 
> 1.0 Contrail + dpdk,
> 2.0 Contrail+SRIOV 
> 3.0 Contrail+dpdk+sriov  
> 
> We have successfully, configured 1.0 Contrail+SRIOV on ubuntu 14.0.4 along 
> contrail 3.0.0.0.2725 version , with the same version of both we want to try 
> and another combination with dpdk. 
> The issue, vrouter is crashing and we are unable to get the Vrouter and other 
> components running on the host.
> 
> Please help in getting the understanding to configure the contrail with dpdk 
> and with both dpdk + SRIOV, as One 10 Gig interface is getting added to 
> vRouter as vhost0 interface and the same was not allowing to work as SRIOV 
> interface (VF's).
> 
> Thanking you all ... Please find the attached screenshots from that server.
> 
> Cheers
> SY
> 
> 
> ___
> Users mailing list
> Users@lists.opencontrail.org
> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] How to configure High Availability for the vRouters on Compute node

[Harshad Nakil]

It's not clear how exactly will you achieve it. So clarity on that might help. 

Regards
-Harshad


> On Sep 30, 2016, at 12:58 PM, Sudeep Batra <sudeep.ba...@ericsson.com> wrote:
> 
> I am thinking vrouters in active active.
> If one fails other can keep working the forwarding table operational.
> No impact on end user traffic. 
> 
> Regards,
> Sudeep Batra
> +91 84201 96572
> 
> On 30-Sep-2016, at 11:51 PM, Harshad Nakil <hna...@gmail.com> wrote:
> 
>> Sudeep,
>> 
>> Can you also suggest how redundant vrouters overcome this issues?
>> 
>> Because it's not very clear to me. 
>> 
>> Regards
>> -Harshad
>> 
>> 
>> On Sep 30, 2016, at 8:53 AM, Sudeep Batra <sudeep.ba...@ericsson.com> wrote:
>> 
>>> Well I am not sure there could be scenario :
>>>  
>>> · Process can get stuck
>>> · vRouter overload due to tenant traffic
>>> · stuck during interaction with agent
>>> · any package update
>>> · etc
>>>  
>>> I am trying to think of a scenario with huge number of Compute nodes (1000s 
>>> being used) and we really like to design with High-Availability principle 
>>> in mind.
>>>  
>>> Regards,
>>> SUDEEP BATRA
>>>  
>>>  
>>> -Original Message-
>>> From: Harshad Nakil [mailto:hna...@gmail.com] 
>>> Sent: Friday, September 30, 2016 7:09 PM
>>> To: Sudeep Batra
>>> Cc: users@lists.opencontrail.org
>>> Subject: Re: [Users] How to configure High Availability for the vRouters on 
>>> Compute node
>>>  
>>> Sudeep,
>>>  
>>> Currently there are no plans to have redundant vrouters.
>>>  
>>> Currently the direction is to support hitless restart of vrouter agent some 
>>> time in future.
>>> Vrouter data plane itself is single point of failure when part of kernel.
>>> In DPDK mode we will support restart with traffic hit sometime in future.
>>> We feel there are diminishing returns wrt efforts and complexity to do 
>>> anything more.
>>>  
>>> What kind of failures are you trying to guard against?
>>>  
>>> Answer to this may explain need for redundancy.
>>>  
>>> Regards
>>> -Harshad
>>>  
>>>  
>>> > On Sep 30, 2016, at 5:58 AM, Sudeep Batra <sudeep.ba...@ericsson.com> 
>>> > wrote:
>>> >
>>> > Hi All,
>>> >
>>> > Is it possible to deploy Vrouters on the Compute node in HA/Redundant 
>>> > mode ?
>>> > Like two instances of vRouters on same Compute node running in 
>>> > active/active.
>>> > I understand its not there, Can it be done ?
>>> > It looks like a useful and necessary feature..
>>> >
>>> > Let me know any suggestions please.
>>> >
>>> > Regards,
>>> > SUDEEP BATRA
>>> > Ericsson India Global Services Private Limited
>>> > sudeep.ba...@ericsson.com www.ericsson.com
>>> >
>>> > ___
>>> > Users mailing list
>>> > Users@lists.opencontrail.org
>>> > http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrai
>>> > l.org
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] How to configure High Availability for the vRouters on Compute node

[Harshad Nakil]

Sudeep,

Can you also suggest how redundant vrouters overcome this issues?

Because it's not very clear to me. 

Regards
-Harshad


> On Sep 30, 2016, at 8:53 AM, Sudeep Batra <sudeep.ba...@ericsson.com> wrote:
> 
> Well I am not sure there could be scenario :
>  
> · Process can get stuck
> · vRouter overload due to tenant traffic
> · stuck during interaction with agent
> · any package update
> · etc
>  
> I am trying to think of a scenario with huge number of Compute nodes (1000s 
> being used) and we really like to design with High-Availability principle in 
> mind.
>  
> Regards,
> SUDEEP BATRA
>  
>  
> -Original Message-
> From: Harshad Nakil [mailto:hna...@gmail.com] 
> Sent: Friday, September 30, 2016 7:09 PM
> To: Sudeep Batra
> Cc: users@lists.opencontrail.org
> Subject: Re: [Users] How to configure High Availability for the vRouters on 
> Compute node
>  
> Sudeep,
>  
> Currently there are no plans to have redundant vrouters.
>  
> Currently the direction is to support hitless restart of vrouter agent some 
> time in future.
> Vrouter data plane itself is single point of failure when part of kernel.
> In DPDK mode we will support restart with traffic hit sometime in future.
> We feel there are diminishing returns wrt efforts and complexity to do 
> anything more.
>  
> What kind of failures are you trying to guard against?
>  
> Answer to this may explain need for redundancy.
>  
> Regards
> -Harshad
>  
>  
> > On Sep 30, 2016, at 5:58 AM, Sudeep Batra <sudeep.ba...@ericsson.com> wrote:
> >
> > Hi All,
> >
> > Is it possible to deploy Vrouters on the Compute node in HA/Redundant mode ?
> > Like two instances of vRouters on same Compute node running in 
> > active/active.
> > I understand its not there, Can it be done ?
> > It looks like a useful and necessary feature..
> >
> > Let me know any suggestions please.
> >
> > Regards,
> > SUDEEP BATRA
> > Ericsson India Global Services Private Limited
> > sudeep.ba...@ericsson.com www.ericsson.com
> >
> > ___
> > Users mailing list
> > Users@lists.opencontrail.org
> > http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrai
> > l.org
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] How to configure High Availability for the vRouters on Compute node

[Harshad Nakil]

Sudeep,

Currently there are no plans to have redundant vrouters. 

Currently the direction is to support hitless restart of vrouter agent some 
time in future. 
Vrouter data plane itself is single point of failure when part of kernel. 
In DPDK mode we will support restart with traffic hit sometime in future. 
We feel there are diminishing returns wrt efforts and complexity to do anything 
more. 

What kind of failures are you trying to guard against? 

Answer to this may explain need for redundancy. 

Regards
-Harshad


> On Sep 30, 2016, at 5:58 AM, Sudeep Batra  wrote:
> 
> Hi All,
> 
> Is it possible to deploy Vrouters on the Compute node in HA/Redundant mode ?
> Like two instances of vRouters on same Compute node running in active/active.
> I understand its not there, Can it be done ?
> It looks like a useful and necessary feature..
> 
> Let me know any suggestions please.
> 
> Regards,
> SUDEEP BATRA 
> Ericsson India Global Services Private Limited
> sudeep.ba...@ericsson.com
> www.ericsson.com 
> 
> ___
> Users mailing list
> Users@lists.opencontrail.org
> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] same MPLS for multiple interfaces

[Harshad Nakil]

No, labels are property of the source of the route. Source is free to choose 
any value. 
We restrict it to smaller value so that it is direct index lookup for 
performance. 

There are many cases but above is true in case of contrail. 

Regards
-Harshad


> On Jun 27, 2016, at 10:34 PM, kevin parrikar <kevin.parker...@gmail.com> 
> wrote:
> 
> Thanks for the very good explanation... 
> i have one more doubt here could you explain that 
> 
>> Since there is no central authority to allocate labels it scales well.
> 
> isn't controller which allocates labels 
> 
> 
> 
>> On Mon, Jun 27, 2016 at 7:18 PM, Harshad Nakil <hna...@gmail.com> wrote:
>> MPLS label is local to compute node (or source of the route (from SDN GW)). 
>> It represents what happens to packet once it reaches the compute node (MPLS 
>> label represents “next-hop” in the vrouter). Most of the time it represents 
>> tap interface in the compute node.
>> Since there is no central authority to allocate labels it scales well.
>> 
>> Regards
>> -Harshad
>> 
>> > On Jun 27, 2016, at 5:04 PM, mad Engineer <themadengin...@gmail.com> wrote:
>> >
>> > I don't see any forwarding issues,things are working,i am trying to
>> > understand how this is achieved :)
>> >
>> > On Mon, Jun 27, 2016 at 4:03 PM, Harshad Nakil <hna...@gmail.com> wrote:
>> >> It is okay if
>> >> 1. two ip(s) belong to same interface(port).
>> >> or
>> >> 2. two ip(s) are on different compute node.
>> >>
>> >> Are you seeing any forwarding issues? Until I know your specific case I 
>> >> cannot comment.
>> >>
>> >>
>> >> Regards
>> >> -Harshad
>> >>
>> >>
>> >>> On Jun 27, 2016, at 10:43 AM, mad Engineer <themadengin...@gmail.com> 
>> >>> wrote:
>> >>>
>> >>> Harshad i understand that mpls labels will be unique per compute
>> >>> node,but here in my setup i see 2 ip address with same MPLS label,do
>> >>> you have any idea on that.
>> >>>
>> >>> On Tue, Jun 7, 2016 at 2:22 PM, mad Engineer <themadengin...@gmail.com> 
>> >>> wrote:
>> >>>> Thanks that answers my question
>> >>>>
>> >>>> On Mon, Jun 6, 2016 at 10:20 PM, Harshad Nakil <hna...@gmail.com> wrote:
>> >>>>> MPLS labels are unique for given compute node. Can be reused in 
>> >>>>> different compute.
>> >>>>>
>> >>>>> Regards
>> >>>>> -Harshad
>> >>>>>
>> >>>>>
>> >>>>>> On Jun 6, 2016, at 7:34 PM, mad Engineer <themadengin...@gmail.com> 
>> >>>>>> wrote:
>> >>>>>>
>> >>>>>> hello All,
>> >>>>>> I have 4 compute nodes with contrail 2.20
>> >>>>>>
>> >>>>>> On my compute node  i noticed that MPLS label is same for two
>> >>>>>> interfaces,i was under the impression that each interface will be with
>> >>>>>> unique  label but thats not always the case.Can you tell me when will
>> >>>>>> interfaces will be in same MPLS and when not.Thanks for any help .
>> >>>>>>
>> >>>>>> ___
>> >>>>>> Users mailing list
>> >>>>>> Users@lists.opencontrail.org
>> >>>>>> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
>> >>
>> 
>> 
>> ___
>> Users mailing list
>> Users@lists.opencontrail.org
>> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
> 
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] same MPLS for multiple interfaces

[Harshad Nakil]

It is okay if 
1. two ip(s) belong to same interface(port).
or
2. two ip(s) are on different compute node.

Are you seeing any forwarding issues? Until I know your specific case I cannot 
comment.


Regards
-Harshad


> On Jun 27, 2016, at 10:43 AM, mad Engineer <themadengin...@gmail.com> wrote:
> 
> Harshad i understand that mpls labels will be unique per compute
> node,but here in my setup i see 2 ip address with same MPLS label,do
> you have any idea on that.
> 
> On Tue, Jun 7, 2016 at 2:22 PM, mad Engineer <themadengin...@gmail.com> wrote:
>> Thanks that answers my question
>> 
>> On Mon, Jun 6, 2016 at 10:20 PM, Harshad Nakil <hna...@gmail.com> wrote:
>>> MPLS labels are unique for given compute node. Can be reused in different 
>>> compute.
>>> 
>>> Regards
>>> -Harshad
>>> 
>>> 
>>>> On Jun 6, 2016, at 7:34 PM, mad Engineer <themadengin...@gmail.com> wrote:
>>>> 
>>>> hello All,
>>>> I have 4 compute nodes with contrail 2.20
>>>> 
>>>> On my compute node  i noticed that MPLS label is same for two
>>>> interfaces,i was under the impression that each interface will be with
>>>> unique  label but thats not always the case.Can you tell me when will
>>>> interfaces will be in same MPLS and when not.Thanks for any help .
>>>> 
>>>> ___
>>>> Users mailing list
>>>> Users@lists.opencontrail.org
>>>> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org


___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org

Re: [Users] LBaaS poor performance with more then one member

[Harshad Nakil]

Yes fat flow will help. 
You can also try increasing the default flow table size from 512 to 2M. And 
decreasing flow timeout from 3 minutes to 5seconds. 

Regards
-Harshad


> On Jun 9, 2016, at 5:31 AM, Jacek Czerniak  wrote:
> 
> Hi All!
> 
> We are testing LBaaS in OpenContrail 2.22.2 version and we saw very strange 
> behevior.
> 
> 
> When we attach 1 vm to lbaas pool the performance is pretty good (it's about 
> 20k transaction/sec per LBaaS,  it's almost equal pure vm without LBaaS).
> 
> When we add second or third VM to the lbaas pool performance is decreasing 
> dramatically (it's about 10k transaction/sec per LBaaS) also the vrouter 
> flows is increasing about 5 times 
> 
> 
> Is it normal behevior?
> 
> I find the commit for LBaaS VIP fat flow but only for master repo not for 
> 2.22 (https://bugs.launchpad.net/juniperopenstack/+bug/1583247)
> 
> 
> I send my observations below. I will be grateful for your help!
> 
> 
> 
> 
> LBaaS with 1 vm
> 
> 
> Test
> 
> siege -c 100 -b http://lbaas_ip
> 
> 
> ** SIEGE 2.70
> ** Preparing 100 concurrent users for battle.
> The server is now under siege...
> Lifting the server siege...  done.
> Transactions:   13872683 hits
> Availability:  99.19 %
> Elapsed time: 744.60 secs
> Data transferred: 476.28 MB
> Response time:  0.01 secs
> Transaction rate:   18631.05 trans/sec
> Throughput: 0.64 MB/sec
> Concurrency:   99.73
> Successful transactions:13872683
> Failed transactions:  113078
> Longest transaction:5.05
> Shortest transaction:   0.00
> 
> 
> flow -r
> 
> New = 10554, Flow setup rate = 18982 flows/sec, Flow rate = 18982 flows/sec, 
> for last  556 ms
> New =  -46, Flow setup rate =  -84 flows/sec, Flow rate =  -84 flows/sec, for 
> last  546 ms
> New =   64, Flow setup rate =  117 flows/sec, Flow rate =  117 flows/sec, for 
> last  546 ms
> New =  -34, Flow setup rate =  -62 flows/sec, Flow rate =  -62 flows/sec, for 
> last  546 ms
> New =   -3, Flow setup rate =   -5 flows/sec, Flow rate =   -5 flows/sec, for 
> last  546 ms
> New =  106, Flow setup rate =  191 flows/sec, Flow rate =  191 flows/sec, for 
> last  553 ms
> New =   64, Flow setup rate =  110 flows/sec, Flow rate =  110 flows/sec, for 
> last  579 ms
> New =   -8, Flow setup rate =  -14 flows/sec, Flow rate =  -14 flows/sec, for 
> last  563 ms
> New =  -13, Flow setup rate =  -23 flows/sec, Flow rate =  -23 flows/sec, for 
> last  545 ms
> New =   73, Flow setup rate =  133 flows/sec, Flow rate =  133 flows/sec, for 
> last  545 ms
> New =   51, Flow setup rate =   93 flows/sec, Flow rate =   93 flows/sec, for 
> last  547 ms
> New =6, Flow setup rate =   11 flows/sec, Flow rate =   12 flows/sec, for 
> last  545 ms
> New =   44, Flow setup rate =   80 flows/sec, Flow rate =   78 flows/sec, for 
> last  546 ms
> New =  157, Flow setup rate =  289 flows/sec, Flow rate =  289 flows/sec, for 
> last  543 ms
> New =  -28, Flow setup rate =  -50 flows/sec, Flow rate =  -50 flows/sec, for 
> last  554 ms
> New =  -18, Flow setup rate =  -31 flows/sec, Flow rate =  -31 flows/sec, for 
> last  577 ms
> ...
> 
> flow -s
> 
> Flow Statistics
> ---
> Total  Entries  --- Total =   11576, new =  16
> Active Entries  --- Total =   11576, new =  16
> Hold   Entries  --- Total =   0, new =   0
> Fwd flow Entries  - Total =   11574
> drop flow Entries - Total =   2
> NAT flow Entries  - Total =   0
> 
> Rate of change of Active Entries
> 
> current rate  =   29
> Avg setup rate=20823
> Avg teardown rate =0
> Rate of change of Flow Entries
> --
> current rate  =   29
> 
> 
> 
> dropstat 
> 
> NO - Flow Action Drop
> 
> PKT0 int rate 380 pps
> 
> Comments:
> Vrouter create about 20k new flow per minute and aged about 20K flow per 
> minute.  Vrouter CPU usage ~70%, haproxy usage ~90%
> 
> 
> 
> 
> LBaaS with 2/3 vm's
> 
> Test
> 
> siege -c 100 -b http://lbaas_ip
> 
> ** SIEGE 2.70
> ** Preparing 100 concurrent users for battle.
> The server is now under siege...
> 
> Lifting the server siege...  done.
> Transactions:1832187 hits
> Availability: 100.00 %
> Elapsed time: 173.98 secs
> Data transferred:  62.90 MB
> Response time:  0.01 secs
> Transaction rate:   10531.02 trans/sec
> Throughput: 0.36 MB/sec
> Concurrency:   99.47
> Successful transactions: 1832187
> Failed transactions:   0
> Longest transaction:5.06
> Shortest transaction:   0.00
> 
> 
> flow -r
> 
> New = 141242, Flow setup rate = 250429 flows/sec, Flow rate = 250508 
> flows/sec, 

Re: [Users] OpenContrail Installation and Configuration for Centos7

[Harshad Nakil]

Rory,

Looks like, may be you have some misunderstanding. Vrouter is replacing 
functionality of Linux bridge OR OVS for virtualized resources and it provides 
distributed virtual router. 

Routing between two physical subnets on different physical interfaces is a use 
case that is not productized yet.  

Regards
-Harshad


> On May 2, 2016, at 9:32 PM, Rory Charlton  wrote:
> 
> Awesome - thank you!
> 
> Do you have any advice for configuring a vrouter so that it binds to two 
> independent NICs on different subnets?
> 
> -Rory
> 
>> On Tue, May 3, 2016 at 2:12 PM, Jakub Pavlik  
>> wrote:
>> Hello,
>> 
>> currently you can use our packages for ubuntu trusty:
>> 
>> deb [arch=amd64] http://apt.tcpcloud.eu/nightly trusty oc30 extra tcp
>> 
>> Jakub
>> 
>> 
>> 
>>> On 2.5.2016 19:06, Rory Charlton wrote:
>>> Hello!
>>> 
>>> I am interested in running a benchmarking evaluation (on Centos 7) of 
>>> OpenContrail as a part of my thesis project and was wondering if I would be 
>>> able to obtain access to the OpenContrail rpms that are listed here 
>>> (currently I don't have permission to download these).
>>> 
>>> On a somewhat related note, I'm looking to try to set up a virtual router 
>>> that can route traffic between two different NICs on the same physical 
>>> host, and was wondering if you might be able to offer some advice/resources 
>>> on the best way to configure OpenContrail to facilitate this.
>>> 
>>> Thanks very much for your time.
>>> 
>>> Regards,
>>> Rory
>>> 
>>> 
>>> ___
>>> Users mailing list
>>> Users@lists.opencontrail.org
>>> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
>> 
>> -- 
>> Jakub Pavlik
>> CTO
>> 
>> [tcp ◕ cloud]
>> 
>> +420 602 177 027
>> jakub.pav...@tcpcloud.eu
>> 
>> tcp cloud a.s.
>> Thamova 16
>> 186 00 Praha 8 - Karlin
>> Czech republic
>> http://tcpcloud.eu
>> http://opentcpcloud.org
> 
> ___
> Users mailing list
> Users@lists.opencontrail.org
> http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org
___
Users mailing list
Users@lists.opencontrail.org
http://lists.opencontrail.org/mailman/listinfo/users_lists.opencontrail.org