My understanding, and it is only *my* understanding, is that vswitches pull and put data very low in the driver, and thus bypass some of those features. For example, you can only assign a vswitch to a NIC, not a VLAN (e1000g3, not e1000g123003). You can set PVID and VID when you create the vswitch, but must use the 'raw' NIC. Since VLANs can be implemented as part of a VNIC with the Crossbow features, I would expect a vswitch to be below where Crossbow ties in as well.

Steffen

On 10/04/10 13:24, Octave Orgeron wrote:
Yeah that feature is only in OpenSolaris. Once S10 goes away, I'm sure a lot of the VSW infrastructure will leverage Crossbow, which changes things and makes networking easier in virtualized environments. *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Octave J. Orgeron
Solaris Virtualization Architect and Consultant
Web: http://unixconsole.blogspot.com
E-Mail: [email protected]
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*


------------------------------------------------------------------------
*From:* Steffen Weiberle <[email protected]>
*To:* [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
*Sent:* Mon, October 4, 2010 11:44:32 AM
*Subject:* Re: [ldoms-discuss] Network related questions concerning vsw spanning tree

The design document at http://hub.opensolaris.org/bin/view/Project+rbridges/ states that the bridging support would be responsible for loop detection. However, at this time it does not do so. So care should be taken if enabling this non-default feature in a domain.

Steffen

On 10/04/10 11:52, Steffen Weiberle wrote:
I am not a certified network engineer, however, as Octave says, only one NIC can be attached to a vswitch. Thus there is no L2 forwarding of broadcasts in current domains, only layer 3, via IP forwarding. As said, that would involve a different MAC address.

With the bridging bridging work that is in progress (or may be completed) on Nevada (don't know the current state), it would be possible for a domain to forward a broadcast. Only then could an ethernet frame sent in on one interface actually come out on another. Not sure if that would require the bridging framework to do loop detection. I would think that is not an LDom function at that point--the forwarding is being done as a service higher up.

Steffen

On 10/04/10 10:53, Octave Orgeron wrote:
Hi,

I could be wrong about this, but my understanding of the VSW implementation is that it can only be attached to one physical NIC, VNIC (with Crossbow), or Link Aggregation to run upon. As a result, there shouldn't be an issue of a loop from that end. Now in the case of using IPMP in the guest domains where you have two separate VSWs that are attached to their own physical NIC on the same network, you won't have a loop either as the MAC addresses of those physical links are different. So again, no looping happening.

Now is it possible to connect multiple VSWs together? Yes it is, but it requires an I/O domain to act as a router between them for traffic to be routed. Otherwise, it would be no different than a multi-homed server connected to multiple networks, but providing no routing between them. It is important to understand that the VSW is a dumb layer-2 switch, it's like going to Best Buy and getting a cheap Linksys, Netgear, or DLink switch. The big difference is that you're not limited on the number of ports, no special management features, and it's at the speed of memory and LDCs. So unlike VMware which tries to emulated a switch with ports and settings, VSWs are much simpler.

If you want to understand it better, probably a good place to look is the UltraSPARC Hypervisor API document on our community site which talks about some of the low-level functionality happening there. *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
Octave J. Orgeron
Solaris Virtualization Architect and Consultant
Web: http://unixconsole.blogspot.com
E-Mail: [email protected]
*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*


------------------------------------------------------------------------
*From:* "[email protected]" <[email protected]>
*To:* [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
*Sent:* Mon, October 4, 2010 6:28:46 AM
*Subject:* [ldoms-discuss] Network related questions concerning vsw spanning tree

Hi there thank you for your answers below. Still some questions:

My concern regarding virtual switches is the fact that from a network perspective I don’t have control, but I am responsible if network attached servers cause network issues.

Primary concern is the fact that in theory it is possible to create layer-2 loops with vswitches. Therefore I would like to know from Oracle/Sun what measures they have taken in order to avoid the creation of Layer-2 loops with virtual switches. I would like to have some documentation of Sun regarding the virtual network implementation within T5240. Please see the splendid vmware documentation reqarding “virtual networking concepts” for example.

Here my requirements for virtual switch implementations within servers:

-virtual switch isolation is required. It should not be possible to connect 2 or more virtual switches in the same server with each other. What mechanisms are in place on the T5240 to prevent this?

-suppose that a virtual switch has 2 fysical uplinks, what mechanisms are in place to prevent the forwarding of a frame coming in on

uplink1 to uplink2 (thereby creating a layer-2 loop)

It would be apriciated, for our network guys to get a satisfied answer for this.

Thanks in adv

Anne Adema

Martin Qoute:”

*Hi,

For a PoC we're running with LDOMs 1.3 and LDOMs 2.0 I have some questions that hopefully someone can help me with. We're using Solaris 10 9/10 (u9) for control/service domain and guest domains.

1. Does the vswitch participate in Spanning-tree?
  If yes, can this be disabled?
  Any differences for LDOMs 1.3 and 2.0 here?*

*No idea, I never observed spanning- tree behaviour of vswitches*

*

*

*
2. How can the bandwidth that a VNIC consumes be limited?
  (Case: we're creating multiple non-global zones in 1 LDOM guest and
want to prevent that one non-global zone can take up all bandwidth for a
  single VNIC).*

*OTOH no way to do this in an I/O domain only without Crossbow. *

*

*

*
3. How can we configure "IP Multipathing" for the vswitch with LDOMs 1.3?
  I know LDOMs 2.0 has this capability...*

*Question: do you want to connect a guest redundantly to the outside world or do you want to connect whole vswitch redundantly to the outside world?*

*

*

*
4. Using VLAN tagging on the vswitch, van I assign one VLAN to a VNIC in a guest domain? So that there's a VLAN between the vswitch and the guest,
  accessible as a VNIC from the guest?*

*Yes. IIRC possible since v1.2, described in the slides on LDOMs under https://sunspace.sfbay.sun.com/x/c9VnDg** (sorry only accessible from the Oracle intranet)*

“


*Stefan Quote”:*

* *

*1. Does the vswitch participate in Spanning-tree?
  If yes, can this be disabled?
  Any differences for LDOMs 1.3 and 2.0 here?*

*No idea, I never observed spanning- tree behaviour of vswitches*

* *

*It doesn't have to. It can not be configured to create loops, so we don't need to spanning-tree to protect against them. Externally, it appears as a normal ethernet port, so the real switches at the other end of the cable can do their spanning tree stuff, if they're so inclined.

Turning it off completly should'nt be supported on any switch. I assume you mean turning off the spanning tree checking before taking a port online. Cisco switches used to do this, and you could turn this off, as in "go online first, check later, then, if loop detected, take port offline". This would speed up the onlining of ports, but not disable spanning tree checking completely.

*

*2. How can the bandwidth that a VNIC consumes be limited?
  (Case: we're creating multiple non-global zones in 1 LDOM guest and
want to prevent that one non-global zone can take up all bandwidth for a
  single VNIC).*

*OTOH no way to do this in an I/O domain only without Crossbow. *

* *

*you'll need crossbow for that, as martin already said”*

* *

* *

*  *


------------------------------------------------------------------------

_______________________________________________
ldoms-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss




_______________________________________________
ldoms-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/ldoms-discuss
  • [ldoms... anne.adema
    • R... Octave Orgeron
      • ... Steffen Weiberle
        • ... Steffen Weiberle
          • ... Octave Orgeron
            • ... Steffen Weiberle
    • R... Sriharsha Basavapatna - Sun Microsystems - Menlo Park United States

Reply via email to