SFP+ and SFP28 compatibility

Introduction

As 25Gbps (SFP28) is on route to displace 10Gbps (SFP+) from its leading role as the work horse in the datacenter. That means that 10Gbps is slowly but surely becoming “the LOM option”. So it will be passing on to the role and place 1Gbps has held for many years. What extension slots are concerned we see 25Gbps cards rise tremendously in popularity. The same is happening on the switches where 25-100Gbps ports are readily available. As this transition takes place and we start working on acquiring 25Gbps or faster gear the question about SFP+ and SFP28 compatibility arises for anyone who’s involved in planning this.

SPF+ and SFP28 compatibility

Who needs 25Gbps?

When I got really deep into 10Gbps about 7 years ago I was considered a bit crazy and accused of over delivering. That was until they saw the speed of a live migration. From Windows Server 2012 and later versions that was driven home even more with shared nothing and storage live migration and SMB 3 Multichannel SMB Direct.

On top of that storage spaces and SOFS came onto the storage scene in the Microsoft Windows server ecosystem. This lead us to S2D and storage replica in Windows Server 2016 and later. This meant that the need for more bandwidth, higher throughput and low latency was ever more obvious and clear. Microsoft has a rather extensive collection of features & capabilities that leverage SMB 3 and as such can leverage RDMA.

In this time frame we also saw the strong rise of All Flash Array solutions with SSD and NVMe. Today we even see storage class memory come into the picture. All this means even bigger needs for high throughput at low latency, so the trend for ever faster Ethernet is not over yet.

What does this mean?

That means that 10Gbps is slowly but surely becoming the LOM option and is passing on to the role 1Gbps has held for many years. In our extension slots we see 25-100Gbps cards rise in popularity. The same is happening on the switches where we see 25, 50, 100Gbps or even higher. I’m not sure if 50Gbps is ever going to be as popular but 25Gbps is for sure. In any case I am not crazy but I do know how to avoid tech debt and get as much long term use out of hardware as possible.

When it comes to the optic components SFP+ is commonly used for 10Gbps. This provides a path to 40Gbps and 100Gbps via QSFP. For 25Gbps we have SFP28 (1 channel or lane for 25Gbps). This give us a path to 50Gbps (2225Gbps – two lanes) and to 100Gbps (4*25Gbps – 4 lanes) via QSFP28. In the end this a lot more economical. But let’s look at SFP+ and SFP28 compatibility now.

SFP+ and SFP28 compatibility

When it comes to SFP+ and SFP28 compatibility we’re golden. SFP+ and SFP28 share the same form factor & are “compatible”. The moment I learned that SFP28 share the same form factor with SFP+ I was hopeful that they would only differ in speed. And indeed, that hope became a sigh of relief when I read and experimentally demonstrated to myself the following things I had read:

  1. I can plug in a SFP28 module into an SFP+ port
  2. I can plug in a SFP+ module into an SFP28 port
  3. Connectivity is established at the lowest common denominator, which is 10Gbps
  4. The connectivity is functional but you don’t gain the benefits SFP28 bring to the table.

Compatibility for migrations & future proofing

For a migration path that is phased over time this is great news as you don’t need to have everything in place right away from day one. I can order 25Gbps NIC in my servers now, knowing that they will work with my existing 10Gbps network. They’ll be ready to roll when I get my switches replaced 6 months or a year later. Older servers with 10Gbps SFP+ that are still in production when the new network gear arrives can keep working on new SFP28 network gear.

  • SFP+: 10Gbps
  • SFP28: 25Gbps but it can go up to 28 so the name is SFP28, not 25. Note that SFP28 can handle 25Gbps, 10Gbps and even 1Gbps.
  • QSFP28: 100Gbps to 4*25Gbps or 2*50Gbps gives you flexibility and port density.
  • 25Gbps / SFP28 is the new workhorse to deliver more bandwidth, better error control, less cross talk and an economical sound upgrade path.

Do note that SFP+ modules will work in SFP28 ports and vice versa but you have to be a bit careful:

  • Fix the ports speed when you’re not running at the default speed
  • On SFP28 modules you might need to disable options such as forward error correction.
  • Make sure a 10Gbps switch is OK with a 25Gbps cables, it might not.

If you have all your gear from a vendor specializing in RDMA technology like Mellanox this detects this all this and takes care of everything for you. Between vendors and 3rd party cables pay extra attention to verifying all will be well.

SFP+ and SFP28 compatibility is also important for future proofing upgrade paths. When you buy and introduce new network gear it is nice to know what will work with what you already have and what will work with what you might or will have in the future. Some people will get all new network switches in at once while others might have to wait for a while before new servers with SFP28 arrive. Older servers might be around and will not force you to keep older switches around just for them.

SFP28 / QSFP28 provides flexibility

Compatibility is also important for purchase decision as you don’t need to match 25Gbps NIC ports to 25Gbps switch ports. You can use the QSFP28 cables and split them to 4 * 25Gbps SFP28.

SPF+ and SFP28 compatibility

QSFP28

The same goes for 50Gbps, which is 100Gbps QSFP to 2 * 50Gbps QSFP.

SPF+ and SFP28 compatibility

SPF+ and SFP28 compatibility

This means you can have switch port density and future proofing if you so desire. Some vendors offer modular switches where you can mix port types (Dell EMC Networking S6100-ON)

Conclusion

More bandwidth at less cost is a no brainer. It also makes your bean counters happy as this is achieved with less switches and cables. That also translates to less space in a datacenter, less consumption of power and less cooling. And the less material you have the less it cost in operational expenses (management and maintenance). This is only offset partially by our ever-growing need for more bandwidth. As converged networking matures and becomes better that also helps with the cost. Even where economies of scale don’t matter that much. The transition to 25Gbps and higher is facilitated by SFP+ and SFP28 compatibility and that is good news for all involved.

Continuous available general purpose file shares & ReFSv3 provide high available backup targets

Introduction

In our previous two blog posts on Veeam and SMB 3 we’ve seen how and when Veeam Backup & Replication can leverage SMB Multichannel and SMB Direct. See Veeam Backup & Replication leverages SMB Multichannel and Veeam Backup & Replication Preferred Subnet & SMB Multichannel.The benefits of this are more bandwidth, high availability, better throughput and with RDMA low latency and CPU offload. What’s not to like, right? In a world where the compute and networks need keeps rising due to the storage capabilities (flash storage) pushing the limits this is all very welcome.

We have also seen earlier that Veeam B & R 9.5 leverages ReFSv3 in Windows Server 2016. This provides clear and present benefits in regards to space efficiencies and speed with many backup file related operations. Read Veeam Leads the way by leveraging ReFSv3 capabilities

When it comes to ReFSv3 in Windows Server 2016 most of the focus has gone to solutions based around Storage Spaces Direct (S2D). That’s a great solution and it is the poster child use case of these technologies.

But what other options do you have out there to build efficient and effective high available backup targets creatively except for S2D? What if you would like to repurpose existing hardware to build those? Let’s take a look together at how continuous available general purpose file shares & ReFSv3v3 provide high available backup targets

CSV, S2D, ReFSv3 & Archival Data

In Windows Server 2016, traditional shared storage (iSCSI, FC, Shared SAS, Shared RAID) with CSV are not recommend to be used with ReFSv3. Why isn’t exactly clear. The biggest impact you’ll see is the performance difference when not writing to the owner node of the CSV in this use case. Even with a well configured RDMA network that difference is significant. But that doesn’t mean that the performance is bad. It’s just that many of the super-fast meta data operations are relatively and significantly slower when compared each other, not that any of these two are slow.

clip_image002

Microsoft does state that an S2D with ReFSv3 and SOFS shares can be used for archival data. Storage spaces and ReFSv3 also have the benefit of offering automatic repair of corrupt data from a redundant copy on the fly even when needed. So yes, the best know supported scenario is this one.

Continuous available general purpose file shares and ReFSv3 provide high available backup targets

But what if we need a high available backup target and would love to leverage ReFSv3 with Veeam Backup & Replication 9.5? Well, you can have 95% of your cookie and eat it to. All this without ignoring the cautions offered.

We could set up SOFS shares on a Windows Server 2016 Cluster with ReFSv3 with traditional shared storage. Some storage vendors do state this is supported actually.

That only means you don’t have the auto repair functionality ReFSv3 combined with storage spaces offers. But perhaps you want to avoid the risk of using ReFSv3 with CSV in a non S2D scenario all together. What you could do is forgo ReFSv3 and use NTFS. How well this will work for archival data or backup is something you’ll need to test and find out how well this holds up. There is not much info is out there, only other cautions and warnings that might keep you up at night.

There is another scenario however and that is using Windows Server 2016 failover clustering to set up continuously available general purpose file shares that leverage SMB3 transparent failover.

The good news is that general purpose file shares (no CSV) do work consistently with ReFSv3 because such a share/LUN is only exposed on one cluster node at the time, the owner. By having multiple shares and setting preferred owners we can load balance the workload across all cluster nodes.

Thank to continuous availability for general file shares and SMB 3 transparent failover we can still get a high available backup target this way. The failover is fast enough to make this happen and all we see with Veeam Backup & Replication is a short pause in throughput before it resumes after failover. To put the icing on the cake, you can leverage SMB multichannel SMB Direct for both backup and restores.

I would take a sizeable whitepaper to walk through the setup so instead I’ll show you a a quick video of a POC we did in the lab here https://vimeo.com/212886392.

clip_image004

If you want to learn more come to the community & other conferences I’m speaking at and will be around for Ask The Experts time opportunities. I’ll be at the German Hyper-V community meet up, The Cloud & Datacenter Conference in Germany 2017, Dell EMC World 2017 and last but not least VeeamON 2017 (see  May 2017 will be a travelling month). 

Conclusion

What do you lose?

Potentially there is one big loss in regards to the capabilities of ReFSv3 with this solution when you are not using storage spaces. This is that you lose the capability to automatic repair of corrupt data. The ability of ReFSv3 to do so is tied into the redundant copies of Storage Spaces (parity/mirror).

What do you get?

That’s fine, the strength of this design is that you get the speed and space efficiencies of ReFSv3and high available backup targets in way more scenarios than “just” S2D. After all, not everyone is in a position to choose their storage fabric for backup targets green field or at will. But they might be able to leverage existing storage and opt to use SMB 3 for their data transport.

So even if you can’t have it all, you can still build very good solutions. It offers ReFSv3 benefits and high availability for your backup target via transparent failover with SMB transparent failover on continuous available general purpose file shares. This also only requires Windows Server 2016 Standard Edition, which is a cost saving. You get to leverage SMB Multichannel and SMB Direct. All this while not ignoring the cautions of using ReFSv3 in certain scenarios.

On top of that, if you use NTFS with this approach it will also work for Windows Server 2012 (R2) as the OS for the backup target cluster hosts.

Disclaimer

I do not work for or at Microsoft, nor am I perfect or infallible just because I’m an MVP. You’ll have to do your own testing and validation. From our testing and without ReFSv3 bugs ruining the show, to me this is a very valid and cost effective approach.

High performance live migration done right means using SMB Direct

I  saw people team two 10GBps NICs for live migration and use TCP/IP. They leveraged LACP for this as per my blog Teamed NIC Live Migrations Between Two Hosts In Windows Server 2012 Do Use All Members . That was a nice post but not a commercial to use it. It was to prove a point that LACP/Static switch dependent teaming did allow for multiple VMs to be live migrated in the same direction between two node. But for speed, max throughput & low CPU usage teaming is not the way to go. This is not needed as you can achieve bandwidth aggregation and redundancy with SMB via Multichannel. This doesn’t require any LACP configuration at all and allows for switch independent aggregation and redundancy. Which is great, as it avoids stacking with switches that don’t do  VLT, MLAG,  …

Even when your team your NICs your better off using SMB. The bandwidth aggregation is often better. But again, you can have that without LACP NIC teaming so why bother? Perhaps one reason, with LACP failover is faster, but that’s of no big concern with live migration.

We’ll do some simple examples to show you why these choices matter. We’ll also demonstrate the importance of an optimize RSS configuration. Do not that the configuration we use here is not a production environment, it’s just a demo to show case results.

But there is yet another benefit to SMB.  SMB Direct.  That provides for maximum throughput, low latency and low CPU usage.

LACP NIC TEAM with 2*10Gbps with TCP

With RSS setting on the inbox default we have problems reaching the best possible throughput (17Gbps). But that’s not all. Look at the CPU at the time of live migration. As you can see it’s pretty taxing on the system at 22%.

image

If we optimize RSS with 8 RSS queues assigned to 8 physical cores per NIC on a different CPU (dual socket, 8 core system) we sometimes get better CPU overhead at +/- 12% but the throughput does not improve much and it’s not very consistent. It can get worse and look more like the above.

image

LACP NIC TEAM with 2*10Gbps with SMB (Multichannel)

With the default RSS Settings we still have problems reaching the best possible throughput but it’s better (19Gbps). CPU wise, it’s pretty taxing on the system at 24%.

image

If we optimize RSS with 8 RSS queues assigned to 8 physical cores per NIC on a different CPU (dual socket, 8 core system) we get better over CPU overhead at +/- 8% but the throughput actually declined (17.5 %). When we run the test again we were back to the results we saw with default RSS settings.

image

Is there any value in using SMB over TCP with LACP for live migration?

Yes there is. Below you see two VMs live migrate, RSS is optimized. One core per VM is used and the throughput isn’t great, is it. Depending on the speed of your CPU you get at best 4.5 to 5Gbps throughput per VM as that 1 core per VM is the limiting factor. Hence see about 9Gbps here, as there’s 2 VMs, each leveraging 1 core.

image

Now look at only one VM with RSS is optimized with SMB over an LACP NIC team. Even 1 large memory VM leverages 8 cores and achieves 19Gbps.

image

What about Switch Independent Teaming?

Ah well that consumes a lot less CPU cycles but it comes at the price of speed. It has less CPU overhead to deal with in regards to LACP. It can only receive on one team member. The good news is that even a single VM can achieve 10Gbps (better than LACP) at lower CPU overhead. With SMB you get better CPU distribution results but as the one member is a bottle neck, not faster. But … why bother when we have …better options!? Read on Smile!

No Teaming – 2*10Gbps with SMB Multichannel, RSS Optimized

We are reaching very good throughput but it’s better (20Gbps) with 8 RSS queues assigned to 8 physical cores. The CPU at the time of live migration is pretty good at 6%-7%.

image

Important: This is what you want to use if you don’t have 10Gbps but you do have 4* 1Gbps NICs for live migration. You can test with compression and LACP teaming if you want/can to see if you get better results. Your mirage may vary Smile. If you have only one 1Gbps NIC => Compression is your sole & only savior.

2*10Gbps with SMB Direct

We’re using perfmon here to see the used bandwidth as RDMA traffic does not show up in Task Manager.

image

We have no problems reaching the best possible throughput but it’s better (20Gbps, line speed). But now look at the CPU during live migration. How do you like them numbers?

Do not buy non RDMA capable NICs or Switches without DCB support!

These are real numbers, the only thing is that the type and quality of the NICs, firmware and drivers used also play a role an can skew the results a bit. The onboard LOM run of the mill NICs aren’t always the best choice. Do note that configuration matters as you have seen. But SMB Direct eats them all for breakfast, no matter what.

Convinced yet? People, one of my core highly valuable skillsets is getting commodity hardware to perform and I tend to give solid advice. You can read all my tips for fast live migrations here in Live Migration Speed Check List – Take It Easy To Speed It Up

Does all of this matter to you? I say yes , it does. It depends on your environment and usage patterns. Maybe you’re totally over provisioned and run only very small workloads in your virtual machines. But it’s save to say that if you want to use your hardware to its full potential under most circumstances you really want to leverage SMB Direct for live migrations. What about that Hyper-V cluster with compute and storage heavy applications, what about SQL Server virtualization? Would you not like to see this picture with SMB RDMA? The Mellanox  RDMA cards are very good value for money. Great 10Gbps switches that support DCB (for PFC/ETS) can be bought a decent prices. You’re missing out and potentially making a huge mistake not leveraging SMB Direct for live migrations and many other workloads. Invest and design your solutions wisely!

E2EVC 2015 Berlin SMB Direct Slide Deck

I attended and presented at E2EVC 2015 in Berlin from June 12th to June 14th. The networking was a blast. No “marchitecure” bull shit or vendor fairy tales what so ever and lots of very open discussions on the realities we’re seeing and facing in virtualization and cloud. Most account managers and esoteric presales would die a painful (but fast) death in this environment.

image

One session was with my Hyper-V Amigo buddy Carsten Rachfahl and was pure demo extravaganza, so no slides. My own session was “SMB Direct – The Secret Decoder Ring” and was an attempt to position this technology what by looking at the why and where followed by the how by who and when.

image

I hope a lot of people had at least a better understanding of SMB Direct, RDMA and DCB. The second aim was to take away the fear many people have of this tech by showcasing it in short demos. Time constraints where a challenge so it was not a 200 level session.

Please download the presentation here if interested.

Enjoy. If you have any concerns or questions, ask, and I’ll try to answer.