Hi Praveen, I think the idea is to be able to have QoS in a way that the traffic from one spoke cannot overwhelm the Hub and lead to issues there. We need to maintain QoS policies per spoke (or spoke type) on the Hub, as well as to be able to push some QoS policies to the spokes from the Hub.
Do I make sense? Thanks, Vishwas On Mon, May 6, 2013 at 9:30 AM, Praveen Sathyanarayan <[email protected] > wrote: > Hi Toby, > > When you say QoS policy, could you elaborate what it really means? I > mean what kind of information does it need to have or exchanged? > > -- Praveen > > From: Toby Mao <[email protected]> > Date: Sunday, May 5, 2013 8:49 AM > To: Yoav Nir <[email protected]> > Cc: IPsecme WG <[email protected]>, "[email protected]" <[email protected]>, Paul > Hoffman <[email protected]> > Subject: Re: [IPsec] One comment to this draft//Fwd: I-D Action: > draft-ietf-ipsecme-ad-vpn-problem-06.txt > > Hi Yoav. > > The QoS implementations in ADVPN are : > > 1. In the star topology, the QoS policy is implemented > individually for each spoke in the hub, all the traffic through the Hub can > be regulated by QoS policy in the hub. > > 2. In the full mesh topology, when the two spokes establish the > direct connection, each spoke should also have the QoS policy for each > other. The QoS policy can be obtained from the Hub, or other control device > , which has the individual QoS policy for each spoke. > > I think your understanding is the same as the QoS implementation in the > full mesh topology. > > Toby > > > On Fri, May 3, 2013 at 4:11 AM, Yoav Nir <[email protected]> wrote: > >> Hi Toby. >> >> Let's see if I understand the issue. I'll describe this with an >> example. Please let me know if I got it. >> >> Suppose we have satellite gateways A, B, C, D, and E. A through D each >> have a bandwidth of 10 Mb/s, while E has 20 Mb/s. >> >> The center gateway, Z, has plenty of bandwidth and the appropriate QoS >> policy. So if A, B, and C are simultaneously sending traffic to E through >> Z, Z will do the QoS magic (maybe by dropping packets or playing with TCP >> ACKs) to make sure the QoS goals are met. >> >> Now add ADVPN to the mix. A and E discover each other, and are able to >> bypass Z. Initially A had no IPsec policy about E. There's no reason to >> think it had a QoS policy about E, and the same is true in the other >> direction. Unless the QoS policy from Z somehow gets transmitted to the >> satellites, they may reach congestion and have the QoS targets miss. >> >> So whereas before ADVPN the center gateway could be counted on to >> handle the QoS (because everything goes through it), as soon as you add >> ADVPN, that policy has to be enforced on the spokes, or not at all. >> >> I'm not sure whether we can or should solve this issue as part of >> AD-VPN, but I want to make sure that we understand the issue. >> >> Yoav >> >> On May 2, 2013, at 6:02 PM, Toby Mao <[email protected]> wrote: >> >> >> On Sat, Apr 27, 2013 at 10:57 PM, Paul Hoffman <[email protected]>wrote: >> >>> These requirements might be useful to add in the next draft, but they >>> need to be refined. >>> >>> On Apr 26, 2013, at 8:10 PM, Toby Mao <[email protected]> wrote: >>> >>> > The ADVPN solution SHOULD be able to implement Quality of Service >>> (QoS) to regulate the traffic in the ADVPN topology. >>> >>> Why is this statement needed? Do you see situations where an ADVPN >>> solution would be *prevented* from implementing some sort of QoS because it >>> was an ADVPN? >>> >> >> [Toby]: There is no situation that ADVPN solution could be prevented >> from implementing Qos. Actually, Qos is crucial on ADVPN, such as sharing >> network bandwidth, meeting the application latency requirement. Especially >> in the Hub, for each spoke, the Qos policy should be implemented >> individually , because different spoke has different link speed and data >> processing capability. Thus, in the ADVPN solution, the small spoke can not >> be overrun by hub by sending too much traffic, also the spoke which has >> large bandwidth cannot hog the hub's resources and starve other spokes. In >> addition, a unique Qos policy for each spoke in the hub could be cumbersome >> for administrator, some improvement could be implemented, such as the >> spokes with the same bandwidth can belong to the same group, the Qos policy >> can be implemented on a basis of group. >> >>> >>> > ADVPN peer SHOULD NOT send excessive traffic to the other members of >>> ADVPN. >>> >>> How would you define "excessive"? Where would that measurement be done? >> >> >> [Toby] The traffic to the ADVPN peer exceeding the actual peer bandwidth >> can be defined as "excessive". To solve this problem, the other ADVPN peer >> should apply Qos policy for this ADVPN peer. >> >> > The traffic for each ADVPN peer CAN be measured individually for >>> shaping and policing. >>> >>> Why is this statement needed? Do you see situations where an ADVPN >>> solution would be *prevented* from measuring individually? >> >> >> [Toby] The reason is explained in the first answer. >> >>> >>> --Paul Hoffman >> >> >> >> >> Email secured by Check Point >> >> _______________________________________________ >> IPsec mailing list >> [email protected] >> https://www.ietf.org/mailman/listinfo/ipsec >> >> >> > > _______________________________________________ > IPsec mailing list > [email protected] > https://www.ietf.org/mailman/listinfo/ipsec > >
_______________________________________________ IPsec mailing list [email protected] https://www.ietf.org/mailman/listinfo/ipsec
