Category Archives: Networking

DELL Has Great Windows Server 2012 R2 Feature Support – Consistent Device Naming–Which They Help Develop

Posted on by
6

The issue

Plug ‘n Play enumeration of devices has been very useful for loading device drivers automatically but isn’t deterministic. As devices are enumerated in the order they are received it will be different from server to server but also within the system. Meaning that enumeration and order of the NIC ports in the operating system may vary and “Local Area Connection 2” doesn’t always map to port 2 on the  on board NIC. It’s random. This means that scripting is “rather hard” and even finding out what NIC matches what port is a game of unplugging cables.

Consistent Device Naming is the solution

A mechanism that has to be supported by the BIOS was devised to deal with this and enable consistent naming of the NIC port numbering on the chassis and in the operating system.

But it’s even better. This doesn’t just work with on board NICs. It also works with add on cards as you can see. In the name column it identifies the slot in which the card sits and numbers the ports consistently.

In the DELL 12th Generation PowerEdge Servers this feature is enabled by default. It is not in HP servers for some reason, you need to turn in it on manually.

I first heard about this feature even before Windows Server 2012 Beta was released but as it turns out Dell has been involved with the development of this feature. It was Dell BIOS team members that developed the solution to consistently name network ports and had it standardized via PCI SIG.  They also collaborated with Microsoft to ensure that Windows Server 2012 would support all this.

Here’s a screen shot of a DELL R720 (12th Generation PowerEdge Server) of ours. As you can see the Consistent Device Naming doesn’t only work for the on broad NIC card. It also does a fine job with add on cards of which we have quite a few in this server.image

It clearly shows the support for Consistent Device Naming for the add on cards present in this server. This is a test server of ours (until we have to take it into production) and it has a quad 1Gbps Intel card, a dual Intel X520 DA card and a dual port Mellanox 10Gbps RoCE card. We use it to test out our assumptions & ideas. We still need a Chelsio iWarp card for more testing mind you Winking smile

A closer look

This solution is illustrated the in the “Device Name column” in the screen shot below. It’s clear that the PnP enumerated name (the friendly name via the driver INF file) and the enumerated number value are very different from the number in Name column ( NIC1, NIC2, NIC2, NIC4) even if in this case where by change the order is correct. If the operating system is reinstalled, or drivers changed and the devices re-enumerated, these numbers may change as they did with previous operating systems.

image

The “Name” column is where the Consistent Device Naming magic comes to live. As you can see you are able to easily identify port names as they are numbered consistently, regardless of the “Device Name” column numbering and in accordance with the numbering on the chassis or add on card. This column name will NEVER differ between identical servers of after reinstalling a server because it is not dependent on PnP. Pretty cool isn’t it! Also note that we can rename the Name column and if we choose we can keep the original name in that one to preserve the mapping to the physical hardware location.

In the example below thing map perfectly between the Name column and the Device Name column but that’s pure luck.image

On of the other add on cards demonstrates this perfectly.image

Windows NLB Nodes Misconfigured after Simultaneous Live Migration on Windows Server 2012 (R2)

Posted on by
5

Here’s the deal. While Windows NLB on Hyper-V guests might seem to work OK you can run into issues. Our biggest challenge was to keep the WNLB cluster functional when all or multiple node of the cluster are live migrated simultaneously. The live migration goes blazingly fast via SMB over RDMA nut afterwards we have a node or nodes in an problematic state and clients being send to them are having connectivity issues.

After live migrating multiple or all nodes of the Windows NLB cluster simultaneously the cluster ends up in this state:

image

A misconfigured interface. If you click on the error for details you’ll see

image

Not good, and no we did not add those IP addresses manually or so, we let the WNLB cluster handle that as it’s supposed to do. We saw this with both fixed MAC addresses (old school WNLB configuration of early Hyper-V deployments) and with dynamic MAC addresses. On all the nodes MAC spoofing is enabled on the appropriate vNICs.

The temporary fix is rather easy. However it’s a manual intervention and as such not a good solution. Open up the properties of the offending node or nodes (for every NLB cluster that running on that node, you might have multiple).

image

Click “OK” to close it …

image

… and you’re back in business.

image

image

Scripting this out somehow with nlb.exe or PowerShell after a guest gets live migrated is not the way to go either.

But that’s not all. In some case you’ll get an extra error you can ignore if it’s not due to a real duplicate IP address on your network:

image

We tried rebooting the guest, dumping and recreating the WNLB cluster configuration from scratch. Clearing the switches ARP tables. Nothing gave us a solid result.

No you might say, Who live migrates multiple WNLB nodes at the same time? Well any two node Hyper-V cluster that uses Cluster Aware Updating get’s into this situation and possibly bigger clusters as well when anti affinity is not configured or chose to keep guest on line over enforcing said anti affinity, during a drain for an intervention on a cluster perhaps etc. It happens. Now whether you’ll hit this issue depends on how you configure and use your switches and what configuration of LBFO you use for the vSwitches in Hyper-V.

How do we fix this?

First we need some back ground and there is way to much for one blog actually. So many permutations of vendors, switches, configurations, firmware & drivers …

Unicast

This is the default and Thomas Shinder has an aging but  great blog post on how it works and what the challenges are here. Read it. It you least good option and if you can you shouldn’t use it. With Hyper-V we and the inner workings and challenges of a vSwitch to the mix. Basically in virtualization Unicast is the least good option. Only use it if your network team won’t do it and you can’t get to the switch yourself. Or when the switch doesn’t support mapping a unicast IP to a multicast MAC address. Some tips if you want to use it:

  1. Don’t use NIC teaming for the virtual switch.
  2. If you do use NIC teaming for the virtual switch you should (must):
    • use switch independent teaming on two different switches.
    • If you have a stack or just one switch use multicast or even better IGMP with multicast to avoid issues.

I know, don’t shout at me, teaming on the same switch, but it does happen. At least it protects against NIC issues which are more common than switch or switch port failures.

Multicast

Again, read Thomas Shinder his great blog post on how it works and what the challenges are here.

It’s an OK option but I’ll only use it if I have a switch where I can’t do IGMP and even then I do hope I can do two things:

  1. Add a static entry for the cluster IP address  / MAC address on your switch if it doesn’t support IGMP multicast:
    • arp [ip] [cluster multicast mac*] ARPA  > arp 172.31.1.232  03bf.bc1f.0164 ARPA
  2. To prevent switch flooding occurs, as with the unicast configure your switch which ports to use for multicast traffic:
    • mac-address-table static [cluster multicast mac] [vlan id] [interface]  > mac-address-table static 03bf.bc1f.0164 vlan 10 interface Gi1/0/1

The big rotten thing here is that this is great when you’re dealing with physical servers. They don’t tend to jump form switch port to switch port and switch to switch on the fly like a virtual machine live migrating. You just can’t hardcode all the vSwitch ports into the physical switches, one they move and depending on the teaming choice there are multiple ports, switches etc …it’s not allowed and not possible. So when using multicast in a Hyper-V environment stick to 1). But here’s an interesting fact. Many switches that don’t support 1) do support 2). Fun fact is that most commodity switches do seems to support IGMP … and that’s your best choice anyway! Some high end switches don’t support WNLB well but in that category a hardware load balancer shouldn’t be an issue. But let’s move on to my preferred option.

  • IGMP With Multicast (see IGMP Support for Network Load Balancing)

    This is your best option and even on older, commodity switches like a DELL PowerConnect 5424 or 5448 you can configure this. It was introduced in Windows Server 2003 (did not exist in NT4.0 or W2K). It’s my favorite (well, I’d rather use hardware load balancing) in a virtual environment. It works well with live migration, prevents switch flooding and with some ingenuity and good management we can get rid of other quirks.

    So Didier, tell us, how to we get our cookie and eat it to?

    Well, I will share the IGMP with Multicast solution with you in a next blog. Do note that as stated above there are some many permutations of Windows, teaming, WNL, switches  & firmware/drivers out there I give no support and no guarantees. Also, I want to avoid writing a  100 white paper on this subject?. If you insist you want my support on this I’ll charge at least a thousand Euro per hour, effort based only. Really. And chances are I’ll spend 10 hours on it for you. Which means you could have bought 2 (redundancy) KEMP hardware NLB appliances and still have money left to fly business class to the USA and tour some national parks. Get the message?

    But don’t be sad. In the next blog we’ll discuss some NIC teaming for the vSwitch, NLB configuration with IGMP with Multicast and show you a simple DELL PowerConnect 5424 switch example that make WNLB work on a W2K12R2 Hyper-V cluster with NIC teaming for the vSwitch and avoids following issues:

    • Messed up WNLB configuration after the simultaneous live migration of all or multiple NLB Nodes.
    • You avoid “false” duplicate IP address goof ups (at the cost of  IP address hygiene management).
    • You prevent switch port flooding.

    I’d show you on redundant Force10 S4810 but for that I need someone to ship me some of those with SFP+ modules for the lab, free of cost for me to keep Winking smile

    Conclusion

    It’s time to start saying goodbye to Windows NLB. The way the advanced networking features are moving towards layer 3 means that “useful hacks” like MAC spoofing for Windows NLB are going no longer going to work.  But until you have implement hardware load balancing I hope this blog has given you some ideas & tips to keep Windows NLB running smoothly for now. I’ve done quite few and while it takes some detective work & testing, so far I have come out victorious. Eat that Windows NLB!

    • Hot Iron, Cold Steel & Cables Are Still Paramount In The Era Of The Cloud

    Posted on by
    Reply

    Cloud, virtualized, on premise, hosted … the people in the field offices need to connect to them and as such hardware is not dead yet Winking smile. Commodities don’t mean obsolete or “in the cloud” only.

    clip_image002

    Some nice DELL PowerConnect 5548P switches. We’ve been using this line of switches (since the 53XX series) for many years now and with great success for in the datacenter (before we switched to 10Gbps) and campus/client access. They’ve never let us down at a price/value point that make the economies of using them to good to ignore.

    Once in a while, we’re out in the field making sure the people can access their apps, services, servers in the cloud, the data center or at a hosting provider. Meaning we get to play with some hardware and we all enjoy that still Smile. Whilst at work at several sites I’m once again confronted with commodities being treated like specialties with the following results:

    • Overly expensive
    • Very little value & capabilities (under delivery)
    • Slow delivery
    • Churning

    To avoid wasting you money or allowing it to be wasted you need to use common sense. If you use advisors get a consigliore, not a racketeer.

    1Gbps to the desktop and get some extra ports

    I’ve talked about getting affordable 10Gbps without compromising capabilities before so here I’ll look at the access/campus side of the story. I still find many organizations rolling out 100Mbps to the desktop for cost reasons and counting ports in orders of one. Two things to keep in mind. Buy 1Gbps and buy some extra.

    Buying vast quantities of something you don’t use but does it power is not a good idea. But being a complete scrooge and not having some extra ports is ridiculous. I have seen many thousands of € wasted in meetings about 10 to 40 switch ports too few in new building projects that have > 5000 outlets. The only real saving I see in in electricity used, if that is a major concern where you are at. Organizations spend tens of thousands of  € discussing something that would be fixed by spending a few thousand which would give extra benefits on top. That’s churning people. Creating work and billable hours by overinflating issues & crying wolf to justify the expenditure that’s supposedly needed to stave of disaster.

    On top of that when you do ask those architects to do some modern designs like SMB Direct  & DCB they freak out & repeat the above ritual. Chances are you’ll spend 20.000 to 30.00 euro on a 6 month study that says it can’t be done because of cost & the probability the sky will fall on your head, leaving you empty handed an poorer. You should have taken the money and just done it. Their scams defer responsibilities to untraceable entities, lines the pocket of consulting houses and, as no one is going to take responsibility to stop this madness, it just goes on forever whilst on paper everything is done by the book and compliancy to the rules is achieved.  Until the day some joker, frustrated at the lack of a few ports, attached a cheapo 8 ports switch to the outlet, creates a loop and brings down the buildings network affecting many thousands. Because the design didn’t handle that to well … been there, seen it.

    I also disagree with the practice of dropping in 100Mbps unless you have really good reasons. Structural cabling is being put in at Cat6A specifications nowadays and CAT5E has been put I for many years. 1Gbps is not a luxury if you do lots of data transfers within an office and have image intensive needs (more and more that is all of us with video, images, all in high res). Google fiber is coming to residential homes … guess what that could mean to services that can be delivered … Heaven forbid you buy 100Mbps because those fancy overpriced VOIP phones only do 100Mbps & you can’t afford the replace them.

    With QoS for VOIP and other use cases some extra bandwidth comes in handy as well for. Also don’t forget software installations & automated rollouts of desktops & laptops. Last but not least it helps deal with crappy network behavior of way to many software packets.

    On the number of ports and the price per port. We buy the most minimal support on switches possible. They hardly ever die on you and if something goes bad it’s a port perhaps, and even that is rare. So don’t waste money on support contracts. Buy some extra ports. For one you need some wiggle room and you have spare capacity to deal with port or even switch failures. If you need 400 ports, by 10*48 port switches. You have spare capacity and can even afford to lose a switch. If one really fails you most have a “lifetime warranty”. You finance 1Gbps to the desktop by dumping support you won’t need, buying value commodity switches and avoiding the racketeers mentioned above. If you need a network engineer, hire one, a good one.

    Than inevitably the cry comes: “you’ll saturate the uplinks”! Not a big issue for the small office (+/- 60 people) setup we did recently but what about a bit larger environments? Todays commodity switches had dual fiber uplink port,10Gbps capable, for a redundant lag. If you build a star design and not a cascade to a more capable core/top switch & you’re golden. It’s also great future proofing as we use access switches for a long time, over 7 years is not an exception, so give yourself some wiggle room.

    Cost you say? Again, forgo the expensive market leaders and you’ll get better value for less money that get the job done very well. Cables, even OM3 fiber, is affordable compared to the labor, construction and maintenance of a  > 1000 employee building. Put in enough cabling to allow for 21st century network traffic and make sure working on it is easy. Good principles used at the wrong place in the wrong way are no good to anyone except for the ones making money of this scam.

    RDMA Over RoCE With DCB Requires Tagged Non Default VLANs

    Posted on by
    3

    It’s DCB That Requires This

    For those of you who are experimenting with the RoCE variant of RDMA for SMB Direct in Windows Server 2012 (R2), make sure you have a VLAN tag in your configuration if this is more than a simple RDMA over two NICs. The moment you get DBC with PFC & ETS involved you’ll need non default tagged VLANs. Do note that PFC alone is good enough, ETS is strictly speaking not a requirement, but I’d consider doing it if you can.

    With Enhanced Transmission Selection (ETS) the network traffic type is classified using the priority value in the VLAN tag of the Ethernet frame. The priority value is the Priority Code Point (PCP), which is described in the IEEE 802.1Q specification and uses a 3-bit field in the VLAN tag with eight possible priority values (0 to 7).

    Priority-based Flow Control (PFC) allows to individually pause priorities of tagged traffic and helps to provide lossless or “no drop” behavior for a certain priority at the receiving port. As  above, each frame transmitted by a sending port is tagged with a priority value (0 to 7) in the VLAN tag. So for the traffic pause and resume functionality to work we need a VLAN tag to carry the priority value.

    Does It Work Without?

    But you’ll tell me that, as you may be lacking a DCB capable switch for lab purposes, you used a direct cable between your two RoCE NICs. And guess what RoCE, might have indeed worked for you without a VLAN tag. You can test & get a feel for what RoCE/RDMA can do for you with just the NICs. But as there is no switch involved you’re not using DCB for PFC/ETS and without that the need for the tagged VLAN isn’t there. Also see https://blog.workinghardinit.work/2013/05/03/smb-direct-roce-does-not-work-without-dcbpfc/.

    So there you go. Design your RoCE/RDMA network based on DCB with PFC( and ETS) and not just on the tests with an direct cable or you might miss a few details that are quite important. Happy testing!