Feeds

back to article Kaminario shovels RAM, shiny new processors onto K2's peaks

Kaminario has updated its K2 all-flash array with new processors, more RAM, higher capacity SSDs and data reduction technology, claiming K2 is more efficient, scalable and cost-effective than either EMC's XtremIO or Pure Storage's FA-400 all-flash arrays. The fourth generation K2 product topped the SPC-1 random IO and SPC-2 …

COMMENTS

This topic is closed for new posts.
Anonymous Coward

Explain to me like I am 5

How is this any different than taking a beefy SAN controllers from NTAP, EMC or HDS and putting all flash behind it? Or better yet, why should I pay attention to this over Pure Storage or Skyera.

1. Performance -- I get it is fast, but how's price/performance and price/capacity?

2. Flash-ness -- is there stuff being done to make flash live longer?

3. System -- everyone has non-disruptive upgrades, VAAI support etc. Is there anything unique here?

I don't work for any of the vendors. Just curious because there is too much sales noise out there.

2
1
Anonymous Coward

Re: Explain to me like I am 5

Good job on commenting on an article for which you did not read...moreover, this really isn't the place to attempt to educate yourself on all the details. I'll just say that...yes...there are huge differences. Go to vendor sites and look at spec sheets. Call up sales and ask for a quote.

I'll tell you, NetApp EF flash Arrays do scale at all yet...limited in iops and capacity. Really, neither does their FAS systems and ClusterMode [is a joke]...although their new 8000 series systems are at least a step in the right direction....now to update antiquated wafl filesystem limits and what a waste spending ten years working on CDoT for a scale-out managed system for still a more or less traditional i/o NAS System. As for EMC and Pure...they are going to be way too expensive when compared to others. They also like to promise the world and under deliver in my experience. Moreover, EMC is very secretive...can't find much info on their site other than sale non-sense. HDS looks promising...as does Kaminario. I don't own either one yet...but I have talked to them and I would love to own one someday. I do own both EMC and NetApp products. I am also a NetApp stock holder, for now. I own an EF540 Array for an IBM GPFS filesystem metadata luns. I started talking with Pure, but stopped quickly as they are nuts with their pricing and dependance on dedupe. I guess that's what happens when you sell-out to VCs. Nimbus, Nimble, and SolidFire are all fairly nice people too...but their systems are highly limited as well for various reasons. Please remember this space is still relatively new...and a lot needs to shake out. As always...hardware is the mostly easy part....software has the long road to catch up.

0
3
Anonymous Coward

Re: Explain to me like I am 5

I think you ought to go to Fry's and built one yourself that will adhere to your requirements...

0
1

Re: Explain to me like I am 5

You are right on the spot – the storage market is awfully noisy and sales teams often beat their drums hard enough to drown out what matter. (And coincidently enough, when the product they are trying to sell lacks competence, the drums become louder.)

So let’s try and give you some real answers – some that even a five-year-old could understand:

First, the legacy HDD-based storage vendors that you mentioned – NetApp, EMC, HDS, etc. – have based their entire architectures on HDD media. In essence, that means that very few controllers (compute) handle a large number of HDDs. We have much better and consistent performance vs. the legacy and at an extremely better pricing on a $/GB basis.

Performance: The Kaminario K2 has the ability to scale out – which basically means it can linearly increase performance according to data center needs, and only buy what you need now – and grow later on. (We can also scale up.) Unlike other vendors (namely Pure Storage and Skyera, both of which can only scale up), you don’t need to pay upfront for performance you might need next year with Kaminario. (Don't believe them if they tell you otherwise. They're just trying to make noise to cover up their lack of scale-out ability.) And when it comes to XtremIO, they do have scale-out – but can't do so non-disruptively. They also have an issue with low density per controller.

Our price per capacity is ~$2/GB usable, with our price per performance running $0.28/IOPS.

Flash: Good question. Our inline deduplication and inline compression reduce the amount of actual data that is written to the SSDs; data that is written to our highly efficient and robust K-RAID is written using log-structured full stripes so that the number of updates per stripe is minimized. (OK - Not quite "child language" but this is enterprise storage after all!) Writes are fully distributed across the entire array and a scalable distributed write cache eliminates hotspots.

Features: Let’s talk about why having the entire software like Kaminario does is not trivial amongst other all-flash array vendors. Most vendors were quick to come out to the market with the essential read/write capabilities and nothing more. Sure, the legacy vendors have it all, but they also haven't really changed much in their architecture for the last 20 years, so you could argue they had time on their hands.

With Kaminario, features like global inline compression, global inline deduplication, snapshots, variable block size algorithm aligned to real workloads and the ability to scale-out and/or scale-up are inherent in K2’s architecture. These features are very hard – and we'd argue near impossible – to add in a later phase.

I hope this turns down the volume a bit and answers your questions. Check out http://kaminario.com/flash-array/ for more information.

Thanks,

Shai Maskit

Product Manager, Kaminario

0
0
Anonymous Coward

Re: Explain to me like I am 5

OP here. Shai -- thanks for the detailed reply. You did chalked up +10 as a vendor in my book for being proactive. Scale-out seems to be driven by the need to grow beyond a single controller (would I scale up to the limit of the system and then scale-out).

In scale-out is the connectivity to the LUN determined by ALUA or is it done transparently for the host (forwarding etc like Nimble)?

re: Fry's guy -- would love it if you can suggest a scale-out SW stack that matches the functionality of what the infrastructure vendors provide.

0
0

Re: Explain to me like I am 5

OP - You’re asking good questions, bring them on! :)

Let’s first set the ground with the basic terms, at least in terms of accessing LUNs.

Active/Passive – All the storage controllers are connected to the hosts, however, only one (in a dual-controller architecture, like the one Pure Storage is using) storage controller can access the backend storage – AKA the active controller. You called it “host forwarding,” which is correct since the passive controller doesn’t do much besides forwarding data from/to the active controller. This approach of active/passive doesn’t utilize half of the compute power, and if we monetize it, the customer is paying for a storage controller that is not used, and the performance is capped by a single controller.

ALUA (or Asymmetric LUN Access) – In this approach, both controllers are active in the sense of accessing the backend storage to read/write data. However, the LUNs of the storage array are distributed between the storage controllers such that per LUN, the controllers are active/passive. This gives better utilization of the compute power of the storage array but requires frequent LUN administration to achieve the right balance of LUN distribution. This balance is practically impossible to achieve since the load on different LUNs varies during the day, and when adding scalability to the game it’s even worse.

Active/Active – This is the Kaminario approach, where each LUN can be accessed from any of the storage controllers. The data is distributed across all the SSDs in the system so you get the same consistent high performance from all the LUNs, with no dependencies from which storage controller the LUNs are accessed by.

Reminding you, the Kaminario K2 has the ability to scale out. When it scales out, the data is automatically redistributed between all the new and existing SSDs, so there is no need to reconfigure host connectivity or deal with LUN affinity to the new/exiting storage controllers. This approach allows the best utilization of the array’s compute power with no compromise on high availability. This translates, of course, to real customer benefits such as cost, enterprise resiliency and easy operational management.

I encourage you to read our architecture white paper that can be found at http://kaminario.com/flash-array/. It really tells the entire story.

Thanks,

Shai Maskit

Product Manager, Kaminario

0
0
This topic is closed for new posts.