Last week I was invited as a delegate at Storage Field Day one. On Wednesday there was an open symposium on the SSD/Flash theme (Solid State Storage Symposium), Thursday and Friday were 2 days of 4 to 5 two hour presentations by some vendors to 12 independent storage specialists. Most of the vendors we spoke to were startups, because it is these people that push the market with new ideas and solutions without the boundaries of legacy hardware/software (and internal politics). One of these startups is Kaminario.
The business model of Kaminario is a Scale-Out SSD SAN software on “off the shelve” hardware. By off the shelf I do NOT mean that you can purchase just licenses. It is based on standard components delivered by enterprise servers & storage vendors. The configurations are preset and not changeable by the reseller/customer. For now the solutions are entirely based on DELL hardware but nothing holds Kaminario from working with other vendors (HP/IBM/…). In fact I guess they would love to have more OEM partners.
Architecture and Configurations
Two (or more) controllers, called IO directors, handle the frontend connectivity to the servers through 4/8/16GB FC. It presents classical block storage through a proprietary MPIO driver. At the backend we have multiple (at least three) storage nodes, called the data nodes. The backend connectivity between the nodes and the directors is at 10GbE. There is also some kind of 1GbE backend connectivity but I guess this is metadata or administrative workloads like performance monitoring.
* Model 1) DRAM only: super performance – heavy writes
* Model 2) Hybrid DRAM + MLC: mixed workloads
* Model 3) MLC only: transactional read
The core model is the second one with a mix of super-speed DRAM and high-speed MLC.
hardware: DELL M1000e Enclosures, full height blades wth 6 PCIe SSDs each. One of the data nodes is not used and will kick in on a failure. Example: because these are blade server solutions, you’ll have to pull a blade on each hardware maintenance (MEM, SSD, fans, …).
If you look at the lab report ESG did last september you’ll see that they managed to get a multi-user database workload on 6 nodes (5 active, 1 standby) and hit up to 200.000 IOPs. Once they added nodes, the performance scaled linearly over 800.000 IOPs by scaling to 16 data nodes. And in here is the value prop of the Kaminario. If one looks at the tiering or caching solutions other vendors tend to use, they always fail to mention that it is only for the size of their flash that you get higher performance, but if the data does not reside (yet) on the flash disks it still hits the spinning disks, hence slowing down.
- I would love to do some crazy whiteboarding on the physical architectures of Kaminario. I think the choice for blade enclosures is good because it is reliable hardware and you do not have to worry about cabling etcetera. But I really think you might get better density in a scale-up/out environment with decent 1U rack servers and a PCIe backend architecture like XSIGO or even just with CNA’s from Emulex or Brocade.
- I also have some thoughts on the second/third/fourth generation integrations. What if they would like to integrate other protocols/media (frontend 10GbE, Infiniband, DCB, FCoE, FCoTR, … ). What if SSDs get 6TB big in a year? What if a vendor gets 48 SSDs direct connected to PCIe per server (and the CPU can handle it 🙂 ? Are they still going to be able to scale inside the same system? Will you get 2 systems? Would you have to do rip-and-replace? If you look at the posts I write on storage you’ll see that I tell people not to buy today what they’ll only need by tomorrow. Not for the lower CAPEX but because of the rapidly changing market. This however requires in-system, non-disruptive hardware expansion through generations! Not many vendors get this picture.
- DELL has no real scale-out all flash solution in the portfolio. Could the current co-operation lead to an acquisition? If you look at all the startups we saw this week, I guess this is one to keep in mind for acquisition, just because of the fact that it is only software!
One of the speakers at the #SSSS was asked what would be next after flash. The answer was: “ask Apple. By the time they have it running for 2 years, we’ll be implementing it in Enterprise IT”. And he was/is right. If you look at the fact that Apple makes more devices of a single product then any other company in the world, they have the leverage to move the focus of engineering new technologies.
disclaimer: all transportation, catering and stay on TechFieldDay was taken care off by the sponsors you can find here through the organization of GestaltITBe Social and Share: