Tintri VM-aware storage, anyone have experience?

T

Thuleman

Supreme [H]ardness
Joined
Apr 13, 2004
Messages
5,833
Just came across these guys http://www.tintri.com while researching VDI storage. A vExpert gave them two thumbs up in a customer case study though I am always weary of these public endorsements.

Has anyone used these arrays? Are they basically "just" running ZFS with their own version of a linux OS like Coraid does?
 
No...it's a lot more. They have in-line deduce and great VMware integration/awareness. Well worth checking out. I don't sell it but I have a good friend that is one of their SEs and it's good. The downside is limited scaling..meaning, once you run out of space you buy a second one and create another NFS datastore. But it's fast... Everything lands on flash and cold data goes to spinning disks.

Also check out Nimble and if you want a true converged solution look at Nutanix. We just picked up Nutanix.
 
Tintri is good stuff and I like their product and integration with VMware but my main concern is the fact that all the storage is presented as one big 13TB NFS datastore. That's an awfully big failure domain. I hear now they offer replication between nodes.

If you want to use them you'll be impressed with how simple it is, easy to manage, and how well it performs.
 
You guys have any idea how much it is? Are we talking 20k, 30k, 40k, 50, 60?
 
Yes and Yes and Yes - since you have to buy nodes to expand capacity it depends on your need. The big push for Tintri is VDI. Their methodology for spawning instances (from gold master image) is basically like a snapshot in that only the specific VDI changes are recorded but on a per user/VM basis. So while the device does have good IO the amount of activity necessary to create (not boot) a new VDI instance is almost instantaneous since it is just a journal pointer.
 
Sure, but how much is one node? I don't really want to go though with registering with them, getting the sales call, yadda yadda yadda, just looking for a ballpark number to get some sense of how it compares to basic storage (with the understanding that a normal storage array doesn't have the VMware integration and that this integration is where they add value above basic tiered storage).
 
Yeah at that price point you better have the 1,000 VMs it can handle to put on it. Totally out of my range for the ~200-250 VDI VMs I will end up with if the 25 seat test goes well this summer.

This is actually a great example of why VDI should be done on an institutional level rather than a department level so that one can take advantage of technology that's made for larger deployments.

Sadly there isn't enough buy-in and funding to do it on the institutional level so we monkey around on a departmental level.
 
Nimble comes in cheaper than that IIRC and also works very well. Also a simple solution but only uses iSCSI.
 
Hello everyone, John Phillips from Tintri here. I am not in sales, and have spent 15 of my 20 high-tech years in storage (including 12+ years at NetApp), so I'll be candid with you. :)

I help drive our strategic alliances, and also have technical marketing duties, which means that I have a lab. Wee! Most of all, I love helping out where I can.

Thuleman, if you would like we can hop on an short, impromptu WebEx later this week (I am packed today) with just us techie nerds, you can ask questions, poke around, at your convenience. Not just a canned demo (those aren't as fun, right). In my setup I have 4x R820's with 256GB of RAM (+ a new UCS chassis, yet to be racked), XenDesktop, Horizon View, Exchange, SQL, SharePoint, etc., etc. We rock equally as hard for server virtualization as well as VDI, although we get more press in the VDI space because it's a hot topic in the industry. I'm a Msft developer/enterprise apps and virtualization guy, so VDI is just another workload.

No pressure of course, I just wanted to offer my help if that sounds useful to you. Feel free to ask questions here, and I'll answer them. I am not however in sales, so keep that in mind. I barely know the list price of our boxes (somewhere in the $90K's), but I can go very deep on the technical side of things. ;)

Warmest regards,

-Jp
 
One other thing that made my heart jump -- we DO NOT use ZFS. When Dr. Kieran Harty (our CEO) left VMware to found Tintri, he secured the best in the business to build a file system that was built 100% from the ground up for flash (SSD) technology (not for spinning disks), and virtualization. Tintri VMstore was conceived and is built right here in Silicon Valley. There's nothing else like it, which is precisely why I am here. It is a brilliant technology.
 
Welcome to [H] John! Classy that you signed up to answer some questions, we have a decent and quite diverse crowd here so there's bound to be some interest in talking to the source. ;)

As for me personally, my departmental deployment on the server and VDI side are just too small to justify the capital cost regardless of the technology benefit. The institution I work for dropped 1M+ into a Hitachi EVA array a couple (few?) years back because someone thought it would be a good idea to buy monolithic storage. The results were mixed.

I think that the Tintri concept is interesting and if I were to run a bigger setup I'd definitely take a serious look at it.
 
Thank you! I was browsing around and checking out a lot of kewl stuff here.. *nice*

$1M wow, that's a lot of $$ for some big [storage] iron. I'd love to drop that kind of coin on a bunch of electronics instrumentation and heavy fabrication/welding machinery. I'm a geek with no bounds, brother. ;)

By all means, if you need help with anything let me know. I'm constantly revving through a bunch of test cases, researching apps, platforms, automating/codifying tedious tasks, etc. I love being part of a community and sharing. It's in my DNA, so to speak.

Summer is here -- everyone have fun!

Cheers!

-Jp
 
It's good stuff. Reliable, fast - only issue is the one I saw already - single failure domain. But if they've got replication now, problem solvedish.
 
Thuleman said:
Welcome to [H] John! Classy that you signed up to answer some questions, we have a decent and quite diverse crowd here so there's bound to be some interest in talking to the source. ;)

As for me personally, my departmental deployment on the server and VDI side are just too small to justify the capital cost regardless of the technology benefit. The institution I work for dropped 1M+ into a Hitachi EVA array a couple (few?) years back because someone thought it would be a good idea to buy monolithic storage. The results were mixed.

I think that the Tintri concept is interesting and if I were to run a bigger setup I'd definitely take a serious look at it.
Click to expand...

HP produces EVA's, not Hitachi (they make VSP/USP/AMS). If you already have monolithic storage, you may want to think about putting something like Atlantis on top.

Keep in mind that HP EVA's are the biggest POS arrays I have ever dealt with.
 
Hi lopoetve, you are absolutely correct (in a given rack)! Just remember that each Tintri VMstore is fully redundant (controllers, disks, networking, power supplies, etc.). You can lose an entire controller, a bunch of disks, etc., and things keep rockin'. In fact, because we're active/standby, you do not suffer any performance degradation when controllers failover. Our *sweet* (sorry, I am spoiled) workload capacity stays at 100%. Lots of disk I/O bandwidth for your apps, etc. Code upgrades are also non-disruptive. Rock and roll, straight through...

If the entire rack in this example is taken out by a flood, thanks for recognizing ReplicateVM! You've been watching (*points*). ;)

I recorded some videos at http://www.tintri.com/resources/videos that highlight ReplicateVM with some apps (e.g. SQL, SharePoint, etc.). It's the "Data management series", by yours truly. Just trying to show why I love Tintri. Yes, it is my goofy voice behind the vids. See if you can detect the lisp from biting my tongue (ouch). ;) If something else is interesting, let me know and I can create and post more videos.

By some fate, HardForum is the only forum I am participating in at the moment (contextually). I have lots of "mad science", creative interests. Everyone just seems cool here. Anyways, thank you sincerely for the "vendor guy" hospitality. I'm true as the driven snow.

REDYOUCH -- As a deep storage nerd, I will take the high road. I'm not a competitive guy (a lover, not a fighter). haha ;)

Cheers to a super kewl group!

-John
 
REDYOUCH said:
HP produces EVA's, not Hitachi (they make VSP/USP/AMS). If you already have monolithic storage, you may want to think about putting something like Atlantis on top.
Click to expand...

The thing is that *I* don't have it, and *they* don't let us use it (meaning our servers cannot access their storage), which is crazy considering that everything is running in the same datacenter, but it is what it is. *They* say that only their stuff can run on it, and they are happy to provide us with VMs on a recharge basis, ~$1,500 per VM per year, with 20GB storage, extra storage is extra at crazy prices to match the cost of the crazy expensive storage they bought. Nevermind that we are all working at the same place.

So it didn't make any sense at all for our department to pay central IT when we can just deploy our own gear for a fraction of the cost.
 
Thuleman said:
The thing is that *I* don't have it, and *they* don't let us use it (meaning our servers cannot access their storage), which is crazy considering that everything is running in the same datacenter, but it is what it is. *They* say that only their stuff can run on it, and they are happy to provide us with VMs on a recharge basis, ~$1,500 per VM per year, with 20GB storage, extra storage is extra at crazy prices to match the cost of the crazy expensive storage they bought. Nevermind that we are all working at the same place.

So it didn't make any sense at all for our department to pay central IT when we can just deploy our own gear for a fraction of the cost.
Click to expand...

I thought I was the only one with that problem...

Your in education aren't you.

I can one up you I'm a branch of central IT and have the same issues. IT is charging IT for VMs
 
ivnext said:
Hi lopoetve, you are absolutely correct (in a given rack)! Just remember that each Tintri VMstore is fully redundant (controllers, disks, networking, power supplies, etc.). You can lose an entire controller, a bunch of disks, etc., and things keep rockin'. In fact, because we're active/standby, you do not suffer any performance degradation when controllers failover. Our *sweet* (sorry, I am spoiled) workload capacity stays at 100%. Lots of disk I/O bandwidth for your apps, etc. Code upgrades are also non-disruptive. Rock and roll, straight through...

If the entire rack in this example is taken out by a flood, thanks for recognizing ReplicateVM! You've been watching (*points*). ;)

I recorded some videos at http://www.tintri.com/resources/videos that highlight ReplicateVM with some apps (e.g. SQL, SharePoint, etc.). It's the "Data management series", by yours truly. Just trying to show why I love Tintri. Yes, it is my goofy voice behind the vids. See if you can detect the lisp from biting my tongue (ouch). ;) If something else is interesting, let me know and I can create and post more videos.

By some fate, HardForum is the only forum I am participating in at the moment (contextually). I have lots of "mad science", creative interests. Everyone just seems cool here. Anyways, thank you sincerely for the "vendor guy" hospitality. I'm true as the driven snow.

REDYOUCH -- As a deep storage nerd, I will take the high road. I'm not a competitive guy (a lover, not a fighter). haha ;)

Cheers to a super kewl group!

-John
Click to expand...

What lopo and I see as a concern is that while the hardware is redundant there is still a single huge data store. What if the file system becomes corrupt? While a minor concern if it did happen all 13tb of data is gone.

I think what would make us feel better is if Tintri would allow users to carve up that 13tb into multiple datastores.
 
DeChache said:
I thought I was the only one with that problem...

Your in education aren't you.

I can one up you I'm a branch of central IT and have the same issues. IT is charging IT for VMs
Click to expand...

This isn't an uncommon practice.
 
Last edited:
QHalo said:
This isn't an uncommon practice. Shit ain't free.
Click to expand...

This I know but when the use politics to take my server budget to buy the vm cluster and then charge me to use it after I paid for half of it.......
 
We have a few Tintris including a 540. I find them pretty consistent and responsive with large numbers of VMs which is all we are looking for. On the system I use the most, we have a Dell 16 blade chassis with 2 cpu x 8 core intel blades.
 
Hi Child of Wonder, how are you?

We've worked hard to make VMstore crazy-reliable, but I totally understand where you are coming from. Carving up meaningful "slices" in this context equates to creating multiple physical disk/raid groups where each "slice" has a separate file system. That significantly changes the deployment, configuration, management, and operational scenarios, which is part of what separates us from the masses. Tintri VMstore's balance of capacity and tremendous I/O and VM density is (imho) not only unique, it delivers flash-first low-latency I/O rates (99%-100% of all read and write I/O from flash) while also leveraging the available capacities of larger SATA disks, etc., for “cold” stuff like aging snapshots, etc., on a granular 8K block basis.

Thanks to Kieran’s vision and our engineering talent, we’ve started a bit of a revolution. Other companies will throw marketing terms around, and eventually start to bolt-on “VM aware” features, etc., into their systems, but until I see something other than marketing jargon out there I’ll call if for what I see it: lipstick on a...limited legacy technology. ;) [Did I just define a new acronym there? "LLT"? haha!] I am humbled (strictly as a technologist) to experience what our engineers have built. It is why I am here.

On that note, I want to emphasize that my comments are not “coached”, and that all of my comments here are my own. Hopefully, no one will frown at me. I doubt it though, we’re an open and fun group. My name has been around longer than Tintri, and I have many years ahead. I wouldn’t risk my credibility for any company. Candidly, Tintri is simply “the real thing.” We are not general purpose storage. Just VMs, baby!

Warmest regards,

-Jp
 
jimh425, Wow! That's awesome. I have Dell servers in my lab right now (R820's) as well, but nothing as cool as that 16-blade chassis. *Sweet!*

Sooooooo glad to hear that we're delivering what you need. We all *love* to hear things like that. It validates all of the hard work and passion. We're not slowing down!
 
REDYOUCH said:
HP produces EVA's, not Hitachi (they make VSP/USP/AMS). If you already have monolithic storage, you may want to think about putting something like Atlantis on top.

Keep in mind that HP EVA's are the biggest POS arrays I have ever dealt with.
Click to expand...

:confused:Correct me if i'm wrong, but why would anyone ever put something like Atlantis on top of their current storage system. If I remember correctly our quote came in at $80,000 for atlantis. You could get the same if not better performance for VDI with new players like Tegile, at a substantially less price point.
 
ivnext said:
Hi lopoetve, you are absolutely correct (in a given rack)! Just remember that each Tintri VMstore is fully redundant (controllers, disks, networking, power supplies, etc.). You can lose an entire controller, a bunch of disks, etc., and things keep rockin'. In fact, because we're active/standby, you do not suffer any performance degradation when controllers failover. Our *sweet* (sorry, I am spoiled) workload capacity stays at 100%. Lots of disk I/O bandwidth for your apps, etc. Code upgrades are also non-disruptive. Rock and roll, straight through...

If the entire rack in this example is taken out by a flood, thanks for recognizing ReplicateVM! You've been watching (*points*). ;)

I recorded some videos at http://www.tintri.com/resources/videos that highlight ReplicateVM with some apps (e.g. SQL, SharePoint, etc.). It's the "Data management series", by yours truly. Just trying to show why I love Tintri. Yes, it is my goofy voice behind the vids. See if you can detect the lisp from biting my tongue (ouch). ;) If something else is interesting, let me know and I can create and post more videos.

By some fate, HardForum is the only forum I am participating in at the moment (contextually). I have lots of "mad science", creative interests. Everyone just seems cool here. Anyways, thank you sincerely for the "vendor guy" hospitality. I'm true as the driven snow.

REDYOUCH -- As a deep storage nerd, I will take the high road. I'm not a competitive guy (a lover, not a fighter). haha ;)

Cheers to a super kewl group!

-John
Click to expand...

Given where I work, I'm intimately familiar with what you all do (and impressed too) - I'll see you all at VMworld again this year!

Now, I just need my SE friend there to finish arranging the loaner, so I can find the holes in your caching I promised I'd find ;)
 
lopoetve, that sounds great! I'll see you there!

Let me know if you want me to "rough up" an SE for you (haha).

What caching? I *know* you're not grouping Tintri VMstore into the category of storage systems that use flash as a cache... Please, say it isn't so. *sniffle*
 
DeChache said:
I thought I was the only one with that problem...

Your in education aren't you.

I can one up you I'm a branch of central IT and have the same issues. IT is charging IT for VMs
Click to expand...

LOL, yes I am in edu.

Going a bit OT; I am actually not opposed to recharge at all, but the combined recharge for all the services we need has to be lower than the TCO of just buying our own equipment, licenses, and effort, on a 3-4 year basis.

The problem is that my institution heavily invested into technology at a scale that was simply not needed at the time and now they are trying to recoup those costs by attempting to charge outlandish recharge rates.

I would happily pay a different department to host our stuff at rates that make economic sense for them and us. $1,500 per VM per year is probably an awesome rate if you only need a couple, but you quickly hit a VM count at which that's just not a viable rate to pay in comparison to just buying your own stuff.
 
Perhaps a Tintri and [H] brain picking session is in order at VMworld? My company is a partner and have sold several units.
 
ivnext said:
lopoetve, that sounds great! I'll see you there!

Let me know if you want me to "rough up" an SE for you (haha).

What caching? I *know* you're not grouping Tintri VMstore into the category of storage systems that use flash as a cache... Please, say it isn't so. *sniffle*
Click to expand...

Lol no, I used the term rather generically (and inaccurately) - was typing fast ;)

I break arrays - it's what I do (just ask a few of the other vendors - you'll probably see them cowering at VMworld) - that being said, they're always better when I'm done ;)
 
Child of Wonder: That sounds fun! We'll be there, armed with choice beverages. :)
lopoetve: I love it. One of my friends at Microsoft (awesome guy) has a notorious reputation for doing the same. One of his favorite "test instruments" is a screwdriver...
 
ivnext said:
Child of Wonder: That sounds fun! We'll be there, armed with choice beverages. :)
lopoetve: I love it. One of my friends at Microsoft (awesome guy) has a notorious reputation for doing the same. One of his favorite "test instruments" is a screwdriver...
Click to expand...

I just use sustained 8MB scsi commands over a 500G test set with ~2000 outstanding IO.
 
2K outstanding IOs? *wowza* That would sure kill spinning media (or require *a lot* of spindles)! Custom workload generator, or a known tool (e.g. IOmeter, etc.)? You don't have to say -- just being my performance-analysis-geek self. haha
 
Well, a lot of hardware these days is "just Supermicro" boxes. The magic is always in the software. Their code isn't anything off the shelf. Far from it.
 
The problem with Tintri is its price. Did you look at that T520? For the price I can built myself a full SSD supermicro box with the same capacity (with some nice Seagate 1200 SAS 800GB), add Windows Server 2012 R2, FreeBSD, whatever and still have some money left.
Then I do not need this software doing some magic...
 
snottz said:
The problem with Tintri is its price. Did you look at that T520? For the price I can built myself a full SSD supermicro box with the same capacity (with some nice Seagate 1200 SAS 800GB), add Windows Server 2012 R2, FreeBSD, whatever and still have some money left.
Then I do not need this software doing some magic...
Click to expand...

And you'll have about half the performance, none of the interoperability or integration, and none of the support. The software is a huge part of how modern arrays work.

Lets be honest here - if you're in the market to build a storage server with BSD/Solaris/*NIX of your choice, you're ~not~ who Tintri is selling to.
 
lopoetve said:
And you'll have about half the performance, none of the interoperability or integration, and none of the support. The software is a huge part of how modern arrays work.

Lets be honest here - if you're in the market to build a storage server with BSD/Solaris/*NIX of your choice, you're ~not~ who Tintri is selling to.
Click to expand...

Yup.
 
You must log in or register to reply here.
Back
Top