All-flash arrays conquered the market of high-performance storage. But the arrays are still expensive for many business and spindle drives are too weak to handle the task. But it’s still possible to get all-flash performance without actually spending a lot of money and StarWind and Intel know how to do it. Watch the record of the webinar to learn how to acquire the technology you could previously only get with high-end hardware SANs, using just StarWind HyperConverged Appliance armed with commodity Intel flash.
How to Get All-Flash Performance with Intel SSD and StarWind HyperConverged Appliance
Fill in the Form to Continue
Allen Scheer, Solutions Architect, Intel
Max Kolomyeytsev, Product Manager, StarWind
Published: February 11th, 2016
Key points of the webinar:
- Introduction into Intel flash technologies
- How to achieve sky-high IOPS with StarWind HCA all-flash configuration
One of the SSD myths that we have in this marketplace has been around for a while because the first flash products were not tremendously reliable. They had a problem of wearing out very quickly. The problem is that once this idea gets in the mind of the public, it is hard to overcome. It is important to recognize that the typical failure rate for hard drives is somewhere between 2,5%-5% of the drivers currently installed per year. The SSDs and including non-Intel SSDs are about ten times better than that. After testing our SSDs as well as those of our competitors, we can actually claim that our failure rate is twenty times better than that of the hard drives.
The general idea is that all-flash technology, with the exclusion of some new things that are just coming on the market, operate by trapping electrons within a silicon structure or some other kind of structure and it can be hit by actual cosmic ray. And your flash storage is now non-reliable. Now this happens rarely. But it happens often enough that there have been some pretty interesting cases in the past few years with Yahoo and some other companies that have actually validated it by using flash solutions that do not protect against this problem. They have actually exhibited millions of dollars of losses and downtime. So we test for this kind of thing rather aggressively and we can definitely say that silent error rate is up to 225x times better than the next major competitor. It is just an actual cosmic ray or something like that hitting an electron in exactly the right way. Such things are rare but they happen often enough to be a serious problem, and it is one of the ways that we try to make our drives much better than our competition.
In addition, to having some products that feature SSDs, StarWind is going to support NVMe. NVMe is a storage technology that takes that RAID controller out of the equation or your HBA card and uses as a much thinner protocol. SSD storage has lot of advantages over the traditional hard drives: it uses less power, it is much faster, there are no moving parts because it is flash, and it is definitely rugged and reliable.
NVMe is considerably faster. The controller and software in SAS and SATA technologies add a lot to the latency and the efficiency and the through-put of the media. The media bar is the blue bar that we call the drive. That is the same on SATA, SAS and NVMe. It is the same memory that we have been using, but we are retaking that protocol out of that controller and that software. In the result, we get things 2.3 times more efficient and definitely 1.5 times better latency.
Intel SSD P3700 is definitely a much better performing product. We guarantee our drives operate at 90% consistently throughout the life of the drive and the five-year warranty. If the drive is not performing within spec, that is the cause for a warranty claim.
3D Xpoint is a technology that we have developed along with Micron. StarWind is getting ready to use this technology. 3D Xpoint is completely new technology and it is not based on trapping electrons to store bits and bytes. This is going to be the memory fabric. It could be a thousand times faster than any current flash memory product. The endurance is going to be a thousand times greater. It also 10 times denser than current DRAM and RAM memory technologies. So we have got a non-volatile technology which is very close to memory speed. But it has the non-volatile capabilities that flash had, and in the same time it has incredible endurance and it is a lot faster than any current NAND technology. So you are going to see a mix of products coming out. Some of them will be in NVMe type of package. StarWind is definitely on the map getting ready for this technology coming on down the line. It means that normally you will have the ability to run all-flash not only at 20 TB for server but 200 TB per server. You will also be able to enjoy your 1TB smartphone or your 128 GB RAM in your laptop. So it is going to become possible to pack more stuff in a smaller space while making it at the same time incredibly fast and reliable.
What is the common first thing people want to do when they think flash and the enterprise world? We have got new SSD capabilities and we need SSDs for solving our problems. The problem is that with virtualization, the storage is under the heaviest stress it had ever been. And one of the easiest way to solve this problem is to transition from spindle disks to all-flash solution. There is one small thing that may spoil the joy of the all-flash SAN here.
From the first glance everything looks just as it would typically look in the SAN environment: everything is redundant and reliable. And with flash we have got tons of performance and a microsecond latency. At this point, all the entire networking stack becomes a huge bottleneck. And at this point we do get the IOPS, but we do not get the latency. So in order to get rid of the latency, we can only get rid of the networking itself, which means we need to change our approach of storing the data on flash.
So SANs are good, but when you got closer to light speed, you got a whole new laws of physics. There is a much better place for flash in this world. It is not the SAN, it should be a Virtual SAN: sitting closer to your virtual machines and sitting closer to your application. So right now we have nodes which are equipped with local flash and we do not do a dedicated layer of storage anymore. And we omit the entire networking layer with this step. The process was like this: going from the virtual machine to the hypervisor and then going to the initiator, then going to the networking stack, then there will be probably some Ram caching or NV RAM caching on the storage unit, and only then we get to the flash drive. That is tremendously slow. We could not do that with the amount of IOPS we have put in our all-flash array and the amount of money we have spent on it. It will be very slow. We transit the flash to our compute notes. At this point, we are becoming hyperconverged as we are merging storage and compute layers. And we are working from the hypervisor layer directly to RAM and flash, which gives us the ability to maintain microseconds latency and high IOPS. And of course we need to make sure that our data is protected so we synchronize the information between our notes. So this is exactly what StarWind all-flash HyperConverged Appliance is about.
We took this approach from just spindles and tiered storage as there are multiple types of customers, multiple types of workload and multiple problems we need to solve. Some problems can be solved just with spindles, some problems can only be solved with flash. We are trying to make that all-flash performance and all-flash appliances to be available not only to the huge enterprises but also to the commodities.
StarWind HCA all-flash L model – this is the basic available model. It has from 24 to 40 cores, up to 512 GB of RAM, and can get you up to a sixteen terabyte a flash storage. It already features 10 GB and 40 GB Ethernet. The good thing is that for the Ethernet you do not need 10 GB or 40 GB switch to start. So that is a little easier on the budget to get you a high performance computing cluster in your environment without actually rebuilding the whole environment. A lot of people who had this issue previously were investing in the Fiber Channel infrastructure, and then Fiber Channel moved to the enterprise, but SMB did not really accept it. So there is no such issue as rebuilding infrastructure with StarWind HyperConverged Appliance. You can start just with putting into servers and migrating your applications there. Now there is of course a bigger player in our team which is the XL flash.
StarWind HyperConverged Appliance all-flash XL model features up to 56 compute cores, up to 1 TB of RAM and about 38 TB of flash available in these units. There will be more flash as we are working on a technology where we deliver better performance by tiering between NVMe and SSD. And there will be also one more tier in this technology. But it is currently outside of this picture. So stay tuned for the news from StarWind and see how we change the way to store data.