NVM Express over Fabrics

Any technologist who’s read, let alone used NVM Express (NVMe), is pretty enthusiastic about it’s capabilities and if it was not for availability and financial restrictions we’d all have at least a couple in our home systems and labs. Let Windows Server 2016 nested virtualization come, woehoe!


It seems to succeed very well in making sure the host can keep up with the performance (low latencies, high throughput) delivered by SSD drives, better than our current interfaces. Today you can drop them into your workstation or server and get going. They’ll give your home lab stellar storage performance today and Microsoft publicly talked publicly about them being supported in Storage Spaces Direct in Windows Server 2016 at Ignite 2015. But there is more to come.

Many of us are very happy with future visions on how PCIe will dislodge SAS/SATA as the preferred SSD interface. This might seem feasible for local storage right now and it can be leveraged for caching or with local PCIe RAID controllers which if shared enable Cluster in a Box (CiB) scenarios. But how to deal with this in an actual storage array, what if we want to size this to a larger scale? There are no “PCIe JBODS”. So what does one do? Well, how did we do it in the past with FC? We created a fabric. Below we see several local & remote NVMe architectures even hybrid ones with traditional SAS.


That’s exactly what NVM Express Inc. is doing, creating the specs for a fabric. This holds the promise to achieve superior results due to the elimination of SCSI translation which reduces latency significantly by delivering NVMe end to end. Not only that but we also see the following efforts in the NVM Express Specification 1.2 to give it enterprise grade capabilities beyond pure performance.

  • Enhanced status reporting
  • Expanded capabilities including live firmware updates

There have been some early demos of NVMe over Fabrics mainly focusing on the “remote” performance. While local NVMe SSDs have the edge on absolute IOPS the difference with NVMe over a fabric is not significant. The reduction in latency is measured in < 10 µs,so that’s good news. The fabric leverages RDMA (yes, yet another reason that my time spending with this technology has been a useful investment). This can be Infiniband, RoCE or iWarp. There’s also the new kid on the block “Intel Omni Scale”  (even if their early demo used iWARP). There’s also a Mellanox RoCE demo.


Now with NVMe SSD disk speeds it seems that the writing is on the wall that ever better fabric performance will be needed to support the tremendous throughput this evolution of storage can deliver. RDMA seems poised for success in regards to this. Now, yes, strictly speaking the NVMe traffic does not require RDMA but let’s just say I don’t see anyone building it without. I also think this means even iWarp fabrics will use DCB (PFC) to make sure we have a lossless network. The amount of traffics will be immense and why not optimize for the best possible performance? I hold the opinion this is beneficial for east-west traffic today in larger environments, especially when in converged networks. Unless the Intel® Omni-Path Architecture blows everyone else away that is Smile. Too early to tell.

Now does this dictate the total and absolute obsolescence of iSCSI and FC? No. There is no reason why a NVMe Fabrics storage solution cannot offer storage to hosts via FC, iSCSI, SMB 3, NFS, FCoE, … They, potentially could even offer all RDMA flavors like iWarp, RoCE or Infiniband to the hosts so you won’t lose your prior investments or get locked into one flavor of RDMA. I have no magic ball so I cannot tell you if this will happen. What I do now that when it comes to MPIO versus multichannel for load balancing and even failover and recovery, multichannel does a (far?) superior job in my honest opinion even when the hypervisor uses separate sessions per virtual machine to achieve better load balancing over iSCSI or the like. So perhaps storage vendors will finally deliver full SMB 3 support in their stacks … if not, well we’ll just abstract your storage way with SOFS. Your loss.  Anyway, I digress. One thing I do know is that I’ll keep a keen eye on what Microsoft is doing in this space, especially in regards to Windows Server 2016 capabilities & scalabilities. It’s time to up the level on scalability & support for newer state of the art technologies once again. It will ensure we get to run our stack on the very best hardware for years to come.

Shared Nothing Live Migration White Board Time – Scenario I

The Problem

Let’s say you are very happy with your SAN. You just love the snapshots, the thin provisioning, deduplication, automatic storage tiering, replication, ODX and the SMI-S support. Live is good! But you have one annoying issue. For example; to get the really crazy IOPS for your SQL Server 2012 DAG nodes you would have to buy 72 SSDs to add to you tier 1 storage in that SAN. That’s a lot of money I you know the price range of those. But perhaps you don’t even have SSDs in your SAN.To get the required amount of IOPS from your SAN with SAS or NL-SAS disks in second and respectively third level storage tier you would need to buy a ridiculous amount of disks and, let’s face it, waste that capacity. Per IOPS that becomes a very expensive and unrealistic option.

Some SSD only SAN vendors will happily sell you a SAN that address the high IOPS need to help out with that problem. After all that is their niche, their unique selling point, fixing IOPS bottle necks of the big storage vendors where and when needed. This is cheaper solution per IOPS than you standard SAN can deliver but it’s still a lot of money, especially if you need more than a couple of terabytes of storage. Granted they might give you some extra SAN functionality you are used to, but you might not need that.

Yes I know there are people who say that when you have such needs you also have the matching budgets. Maybe, but what if you don’t? Or what if you do but you can put 500.000 € towards another need or goal? Your competitive advantage for pricing your products and winning customers might come form that budget Winking smile

Creative Thinking or Nuts?

Let’s see if we can come up with a home grown solution bases on Windows Server 2012 Hyper-V. If we can this might solve your business need, save a ton of money and extend  (or even save) the usefulness of you SAN in your environment. The latter is possible because you successfully eliminated the biggest disk IO from you SAN.

The Solution Scenario

So let’s build 3 Hyper-V hosts, non-clustered, each with its own local SAS based storage with commodity SSD drives. You can use either storage pools/spaces with a non-raid SAS HBA or use a RAID SAS HBA with controller based virtual disks for this. If you’ve seen what Microsoft achieved with this during demos you know you can easily get to hundreds of thousands of IOPS. Let’s say you achieve half of what MSFT did in both IOPS and latency. Let’s just put a number on it => that’s about 500.000 IOPS and 5GB/s. Now reduce that for overhead of virtualization, the position of the moon and the fact things turn out a bit less than expected. So let’s settle for 250.000 IOPS and 2.5GB/s. Anybody here who knows what this kind of numbers would cost you with the big storage vendors their SANs? Right, case closed. Don’t just look at the cost, put it into context and look at the value here. What does and can your SAN do and at what cost?

OK we lose some performance due to the virtualization overhead. But let’s face it. We can use SR-IOB to get the very best network performance. We have hundreds of thousands of IOPS. All the cores on the hosts are dedicated to a single virtual machine running a SQL Server DAG node and bar 4Gb of RAM for the OS we can give all the RAM in the hosts to the VM. This one VM to one host mapping delivers a tremendous amount of CPU, Memory, Network and Storage capabilities to your SQL Server. This is because it gets exclusive use of the resources on the host, bar those that the host requires to function properly.

In this scenario it is the DAG that provides high availability to the SQL Server database. So we do not mind loosing shared storage here.


Because we have virtualized the SQL server you can leverage Shared Nothing Live Migration to move the virtual machines with SQL server to the central storage of the SAN without down time if the horsepower is no longer needed. That means that you might migrate another application to those standalone Hyper-V hosts That could be high disk IO intensive application, that is perhaps load balanced in some way so you can have multiple virtual machines mapped to the hosts (1 to 1, many to one). You could even automate this all and use the “Beast” as a dynamic resource based on temporal demands.

In the case of the SQL Server DAG you might opt to keep one DAG member on the SAN so it can be replicated and backed up via snapshot or whatever technology you are leveraging on that storage.

Extend to Other Use Cases

More scenarios are possible. You could build such a beast to be a Scale Out File Server or PCI RAID/Shared SAS if you need shared storage to build a Hyper-V cluster when your apps require it for high availability.


The latter looks a lot like a cluster in a box actually. I don’t think we’ll see a lot iSCSI in cluster in a box scenarios, SAS might be the big winner here up to 4 nodes (without a “SAS switch”, which brings even “bigger” scenarios to live with zoning,  high availability, active cables and up to 24Gbps of bandwidth per port).

Using a SOFS means that if you also use SMB 3.0 support with your central SAN you can leverage RDMA for shared nothing live migration, which could help out with potentially very large VHDs of your virtual SQL Servers.

Please note that the big game changer here compared to previous versions of windows is Shared Nothing Live Migration. This means that now you have virtual machine mobility. High performance storage and the right connectivity (10Gbps, Teaming, possibly RDMA if using SMB 3.0 as source and target storage) means we no longer mind that much to have separate storage silos. This opens up new possibilities for alleviating IOPS issues. Just size this example to your scenarios & needs to think about what it can do for you.

Disclaimer: This is white board thinking & design, not a formal solution. But I’d cannot ignore the potential and possibilities. And for the critics. No this doesn’t mean that I’m saying modern SANs don’t have a place anymore. Far from it, they solve a lot of needs in a lot of scenarios, but they do have some (very expensive) pain points.