Converged infrastructure, Nutanix

[Funk]

bhstewie said:

swerni said:

There are lots of technologies that can offer cloning, replication etc.
Dedupe is an interesting one. There is a massive overhead for running it. If you're using expensive disk or SSD as your storage media then it's worth it, if you aren then it does't usually pay for itself.

Their management layer is okay, but again there are much better ways of managing and supporting an infrastructure.

I agree entirely, and for most folks it's probably cheaper to just buy more disk space.

I struggle with this myth that hyperconverged means you don't need IT people and you don't have to do any management.

The biggest downside with hyperconverged as mentioned earlier is that it assumes linear growth for both compute and capacity and that's not always right for many organisations.

It definitely has its place (and I agree with you about the misconceptions about the management of it) but for most SMEs the traditional server/SAN model probably offers a far greater degree of flexibility, especially when it comes to differing scaling requirements.

[bitchstewie]

It was me that mentioned it. If you're storing dumb file data that doesn't compress/dedupe well you're going to be a world of financial hurt with any hyperconverged system when you need to expand (or to be fair any traditional array that only allows you to expand by the shelf like Nimble and Tegile and some others do).

[devnull]

There's plenty of scope for a well thought out DIY approach to this, assuming you use the right orchestrator.

My labs alone at work have 10+ compute hosts - whatever I get my hands on really from "obsolete" kit - by obsolete, I mean Dual Processor DL360G8s (!). Cheap and easy to do it this way.

I've played with the nutanix stuff. Its great for a few days titting about with some new shiny kit, but I wouldn't pay for it out of my own pocket.

HP's helion is also a similar proposition.

In my field, we work with ultra critical real-time apps, and we've seen our products barf on the hyper converged boxes when stressed.

[Vaud]

Funk said:

Make three visits and bring a different one each time!

You may need sunglasses on for one of them.

Edited by Vaud on Friday 22 April 18:00

[Vaud]

devnull said:

In my field, we work with ultra critical real-time apps, and we've seen our products barf on the hyper converged boxes when stressed.

Sure, they don't meet every workload. Often good for stateful apps that have a relatively wide window of acceptable performance (vdi, mail, sap) or some stateless (e.g. private cloud PAAS), but a high performance, time critical load - I'd be careful.

[AClownsPocket]

This is all interesting stuff.

Although the company I work is far from an SME, this SME size solution may just work for us in this case. Our mission critical ERP infrastructure is in a Co-Lo, so the HCI stuff is to replace file servers, some small scale SQL deployments (2012 with AlwaysOn) and regional app servers. We are in the process of removing a load of servers from the infrastructure as they are no longer needed as we've gone Co-Lo, so its really to manage the localised requirements and provide infrastructure for file services, domain controllers, etc, nothing exciting

Funk, thank you for the email. Between you and swerni I have a lot to think about. I have reached out via Insight (who we use) to setup some meetings with Nutanix, Nimble and EMC. I may come to you guys directly if I'm not happy with the info I get.

Cheers

ACP

[Sheets Tabuer]

How well can these one box vendors be believed, I dropped two 100TB SANs in each datacentre and these guys walk in and say the can sell me three boxes for 150k that can do the same thing.

I gotta say I feel like I'm buying a second hand car meeting all these reps.

[Sheets Tabuer]

Indeed but saving 5-700k seems pie in the sky, something deep down is telling me I'd have to go and apologise to the board

I almost bought a tintri box to test but NFS and and no fibre put me right off.

[Vaud]

If you want agnostic, buy from an SI who have to stand by their SLA and managed service offering...

[Vaud]

swerni said:

You're not agnostic, but you do decouple technology from the solution and as you say, deliver an SLA.
Not that an SI has ever failed to deliver on it's promises.......

I couldn't possibly comment. Like Manuel, "I know nothing Mr Fawlty"

[bitchstewie]

Sheets Tabuer said:

How well can these one box vendors be believed, I dropped two 100TB SANs in each datacentre and these guys walk in and say the can sell me three boxes for 150k that can do the same thing.

I gotta say I feel like I'm buying a second hand car meeting all these reps.

I know of few other industries where the price can literally half from their "best offer" simply by saying "We'll make our decision once we've spoken with <rival>".

The HC stuff absolutely works, but it's not suitable for all use cases and isn't the magic bullet they will have you believe (reviews are hard to come by because Nutanix used to forbid discussing performance in their EULA.

Bottom line is with any of them when you have them £30-50k for a node you're getting about £8k worth of Dell/SuperMicro/Cisco server so what they're essentially selling you is a software defined storage product that happens to be tied to the hardware they supply.

Same with Nimble, never heard a bad word about them but the hardware is regular SuperMicro (nothing wrong with that) with their software on it.

If you just need some storage for a few hosts it's one hell of a jump from an entry level EMC/HDS/3PAR box with a couple of hosts looped directly in, to several nodes of any HC product (with the exception of Scale and vSAN which does seem reasonably priced).

It's use case and with all respect to the OP we still have no idea if we're talking 2TB or 200Tb and 200 IOPS or 20,000 IOPS.

[bitchstewie]

I'm no knocking Nimble, they're one of a few vendors I don't think I've ever heard a bad word about, but it's interesting (to me) that if you go under the hood it is just regular tin (and again SuperMicro is good stuff, not knocking that) backed with some apparently amazingly good telemetry/monitoring and support feeding info back to base.

I'm just curious who's going to buy them since Dell have EMC and HP have 3PAR and most of these newer vendors seem destined to either burn or be acquired and I don't see Nimble burning.

[Funk]

AClownsPocket said:

Funk, thank you for the email. Between you and swerni I have a lot to think about. I have reached out via Insight (who we use) to setup some meetings with Nutanix, Nimble and EMC. I may come to you guys directly if I'm not happy with the info I get.

Cheers

ACP

No worries, you're most welcome.

Insight aren't a Nimble reseller so if you want info about them drop me a line and I can work with swerni to get you whatever you need to know.

[Funk]

ACP - how are you getting on..?

[AClownsPocket]

Hi

I've had a quick call with Nutanix today and have meetings planned in May with Nutanx, Nimble and EMC. Those will be technical discussions so hopefully have more insight and look to make a recommendation then.

[Ved]

swerni said:

The challenge with hyper converged is, how to scale it?
You're tied into modular building blocks, if you need more compute, you buy more compute and storage.
If you need more storage you buy more compute and storage.
Buying in is fairly cheap, scaling it ain't.

They are great for organisations who won't grow and have no IT skills.
Also the storage functionality is pretty crap as well


I should add, I am very biased.

Not true on how it scales. If you want to add more overall storage capacity to a Nutanix cluster you just add in storage nodes. These particular nodes have minimal CPU (1x 6 core) to run the storage controller VM but have 30TB of RAW storage on them (14TB ish usable before dedupe/compression etc). They don't incur a hypervisor cost on top as they don't run guest VMs so it's not the same cost as buying 'regular' nodes that can have any CPU or storage config on them. The nice thing is you add what you need when you need it so it's a glove-fit rather than a bin liner.

I'm also biased

[AClownsPocket]

Ved said:

Not true on how it scales. If you want to add more overall storage capacity to a Nutanix cluster you just add in storage nodes. These particular nodes have minimal CPU (1x 6 core) to run the storage controller VM but have 30TB of RAW storage on them (14TB ish usable before dedupe/compression etc). They don't incur a hypervisor cost on top as they don't run guest VMs so it's not the same cost as buying 'regular' nodes that can have any CPU or storage config on them. The nice thing is you add what you need when you need it so it's a glove-fit rather than a bin liner.

I'm also biased

This is appealing and makes me more keen to demo the Nutanix kit. Next week will be interesting with the various manufacturers. As much as we've used EMC before, I'm sceptical their product is going to fit what we want.

[Fastdruid]

Sheets Tabuer said:

I almost bought a tintri box to test but NFS and and no fibre put me right off.

Many years ago when the choice was 1GbE or 2+Gb FC I'd have agreed with you, now with 10GbE and DCB for all except the real high performance stuff I wouldn't bother with FC.

swerni said:

Why would you want NFS?

Why wouldn't you? Stable mature technology, Way easier to manage, stateless and handles clustered storage. Performance is no different to iSCSI and it's only beaten by FC because of FC having 16G instead of 10GbE!

[Fastdruid]

AClownsPocket said:

I like Nutanix because I can use our existing VMware licenses with them, but it also supports Hyper-V which is where we see ourselves moving to in the next 2 years.

bhstewie said:

Also why are you switching from ESXi to Hyper-V as it's not that often you see that.

Don't. Just don't.

Hyper-V is just abysmal. It's like Microsoft took a good look at VMware and thought "nah, we're not going to make it that easy".

Everything that is easy in VMware isn't and even the stuff that is tricky in VMware has been made far worse. Cluster Management is terrible. Networking is horrible. Filesystem management is slapdash and random. Storage is a joke (new "datastore"? 18 step process!). DR can be painful. SCVMM is an abortion of terrible badly thought out ideas on top of badly thought out ideas.

The only reason for going with Hyper-V is cost (or if you really hate the vCenter web console ) and even then IMO it's just not worth it.

[AClownsPocket]

Fastdruid said:

Don't. Just don't.

Hyper-V is just abysmal. It's like Microsoft took a good look at VMware and thought "nah, we're not going to make it that easy".

Everything that is easy in VMware isn't and even the stuff that is tricky in VMware has been made far worse. Cluster Management is terrible. Networking is horrible. Filesystem management is slapdash and random. Storage is a joke (new "datastore"? 18 step process!). DR can be painful. SCVMM is an abortion of terrible badly thought out ideas on top of badly thought out ideas.

The only reason for going with Hyper-V is cost (or if you really hate the vCenter web console ) and even then IMO it's just not worth it.

We use ESXi 5.5 at the moment and it works brilliantly. However, its bloody expensive and having an MS EA, we want to leverage the benefits. Is Hyper-V that bad in Server 2016, or are you referring to an earlier version? We've always stuck by the mantra, use an MS product only on version 3

Edited by AClownsPocket on Friday 13th May 11:06