In this episode I’m with Niels Roetert (system engineer at Tintri), and we talk about:

– Difference between Tintri and other hybrid storage systems in the market
– Quality of Service

Tintri will be presenting at next TECH.unplugged event in Amsterdam next 24/9. A one day event focused on cloud computing and IT infrastructure with an innovative formula combines a group of independent, insightful and well-recognized bloggers with disruptive technology vendors and end users who manage rich technology environments.

Here the full transcript of the show

Enrico: Welcome to a new episode of Today, I am with Niels Roetert, system engineer in Europe for Tintri. Hi, Niels, how are you?

Niels: Hi Enrico, good morning, I’m really fine. How are you?

Enrico: Really good, and happy to have you. Also, I’m happy to have you for tech.unplugged event in September. We’ll have a lot of bloggers to talk about new technology, new different subjects. You are one of the sponsors of the event.

Niels: We’re really looking forward to it. We saw the lineup of speakers, and it looks really promising. Definitely looking forward to it.

Enrico: Great. I’ll go straight to the question. You know the format, three questions, five minutes. First of all, it’s about Tintri. Tintri produces an hybrid storage appliance, which is called VMstore. The name says a lot about what it is, but can you explain to us what is the difference between you and other storage system in the market?

Niels: Yeah, sure. Great question, because there’s a lot of confusion when we talk to prospects and customers in general about why Tintri is different. As you said, the name kind of spoils it already. We call it Vmstore, and that’s what we focus on. We focus on the virtual machines. We’ve built a storage appliance specifically for virtualized workloads. We support obviously hypervisors like Vmware, Hyper-V, Xenserver. We now support Open Stack as well, and RedHat enterprise virtualization, so a broad support for different hypervisors. But the focus is really on the virtual machine. In our case, the object that we manage is not a learn or a file system, it’s the actual VM. We get our statistics at the VM level, so we can give you really detailed information for each individual virtual machine, and even at the virtual disk level. Things like how many iops is a VM producing? Are those predominantly reads or writes? What are the block sizes? But also things like how many latency you are experiencing coming to a certain VM. We don’t only do that from the storage perspective, but we also do it at the network and at the host level.

We give the administrator an [entrant 00:02:31] view of the performance of a VM. The other thing is, it’s not just all about statistics, it’s also we can basically set SLA’s at the VM level. We can do things like snapshots, cloning, replication, at the VM level. With classical storage, and some of our competitors, it’s all at the [learn 00:02:52] level. Basically, what a lot of competitors are doing are putting either flash in a storage array or a combination of flash and disk. But they’re not really solving any problems. They’re just accelerating the iops, but they’re not really giving you any insights at the VM level, or helping you solve any of the problems. That’s what Tintri’s all about. It’s about giving you the visibility, giving you the control at the VM level. We are, by the way, different than some of the others. We use the NFS protocol to connect the storage to the hypervisors, except for hyper-v, obviously, where we support SMB3. One of the cool things about the Tintri technology is that we can actually take the whole capacity of an appliance, and our biggest appliance today is a hundred terabytes, and we can just present it as a single data store.

You get a dramatic reduction in the number of data stores that you have to manage, and we don’t put any SLA’s on the data store level, we just do it all at the VM level, so it makes management much, much easier. Quite a reduction in management effort.

Enrico: This raises another question about the fact that you’re talking about VM level ground running, okay, which is the key feature?

Niels: Yup.

Enrico: Okay, but many vendors are now talking about the same capabilities implemented through the VVOLs. What is your position about that?

Niels: At Tintri, we really love the VVOLs technology, because it’s basically a staking of Vmware, saying that the object to manage should not be a learn or a [falseness 00:04:30], it should be at the VM level, or even at the vdisk level. That’s what we’ve been doing since day one, since we introduced our product. Version 1.0 of our software was doing basically what VVOLs has been promising for the last couple of years. We did do a press release when Vmware said that VVOLs was going to come out in Vshpere 6. We basically said, “We will support VVols.” Force is actually quite a simple implementation because we have the functionality already. It’s basically a one on one mapping with our existing technology. We will support up to a million VVols, which is quite unique in the market, because with some of the classic storage rates, it’s going to be really, really hard to support a large number of VVOLs. Actually, on our website, we have a VVOL calculator, and it just asks you how many VMs do you have? Are you planning to take any snapshots and things like that? It will show you the number of Vfo’s that you need to be able to support that.

That number goes up quite rapidly, so if you just have like, let’s say a hundred VMs, and you do some snapshotting, it can be tens of thousands of VVOLs that you need. Not a lot of storage [rigs 00:05:48] out there will support that number.

Enrico: I see. You also talked about implementing a kind of quality of service, true SLAs for different machines, right? How do you implement that?

Niels: Quality of service at the VM level is one of the technologies that we introduced in our latest software release, version 3.2. It’s implemented for the end user in a really simple fashion, because one of the things with quality of service is that it’s a great technology, but it’s very easy to shoot yourself in the foot by doing things wrong. I mean, you can control the behavior of one VM, but basically have other VMs suffer from that. What we did is we made it really easy. You can set the maximum number of ios that a VM can do, but you can also the guarantee the number of ios another VM can do by reserving resources in the system. If you change the quality of service at things for a certain VM, you can directly see how that impacts other VMs, and how it impacts the overall performance of the appliance. Normally, we present per virtual machine or even per V disk the host latency, the network latency, and the storage latency. The storage latency used to be divided in disk latency and flash latency, since we are a hybrid. Now we actually two kinds of latency, and that’s contention and throttle.

Contention means that if an array is really busy and virtual machines are suffering getting the right amount of resources, and basically, they are fighting for the same resources, this will basically add latency, and that’s called contention. We visualized the contention that is there in the system, and we can also show you which VMs are suffering from the contention. You can basically decide which VM is more important to me, and then you can set quality of service on the one that is less important. Reducing the amount of resources that it can take, but that will directly impact the performance of that particular VM, and it will add latency to that VM. We also visualize that, so we visualize the contention, we visualize also the throttling that it gets when you start putting quality of service on it, and limiting the amount of resources.

Very, very easy to use. It’s a really simple [gui 00:08:21] that we give you. You can basically drag the lines of the maximum amount and the minimum amount. We have a nice video up on Youtube about it that shows it really, really well. But really important is that it is basically fool-proof, because if you change the quality of service on a VM, it will show you directly how that impacts all the VMs, and how it impacts the overall performance. I think we did a great job on that.

Enrico: I like that. Good. Thank you very much for all your answers, and I can’t wait to see your presentation at Tech.unplugged in September.

Niels: I think it’s going to be a great event, looking forward to meeting you and some of the other speakers. It’s going to be really interesting, and I hope we have a large crowd of people there. Really looking forward to it, and I thank you for the excellent questions.

Enrico: Thank you very much again for being with us today.