“…if the application is at the other end of a cable, I don’t see how you get rid of the inherent latency in the cable.”
Many applications in wide use in enterprise today still haven’t been optimized to take advantage of flash array-type speeds – especially an array with a very large cache in front that is even faster than the SSD array itself. In other words, the application itself is often a bottleneck, not the wire speed. But even in cases where it isn’t, aggregated “wires” (more ports in use in parallel) and higher-speed caching often make the wire latency negligible. 40GbE ports are in mainstream use now, and you can aggregate many of them together fairly easily. 100GbE ports are available will be mainstream soon as well.
Also, don’t underestimate flash offerings from EMC or NetApp or HPE outright in your free love of PSTG. There are large buckets of money there that can be used for marketing, acquisition or even later entry into the market with a better v2.0 product, or one that matches performance but also carries all the other intrinsic benefits at a similar cost.
Again (full disclosure), although I work for a large storage vendor, I am also eagerly looking for a good entry point on PSTG myself, if for no other reason than to hedge the former. Also, storage isn’t going away anytime soon… the winners in this kind of space will typically have to offer (mostly in order of necessity): excellent support when needed, an advantage in predictive failure analysis, tools for storage management as well as performance analysis and trending, ease of use, market-leading performance, easy buying/sales options, and strong ROI and price/performance metrics. Smaller vendors typically can only offer one or three of those, which gets them into a given niche, but often prevents them from larger/wider adoption.
But even in cases where it isn’t, aggregated “wires” (more ports in use in parallel) and higher-speed caching often make the wire latency negligible. 40GbE ports are in mainstream use now, and you can aggregate many of them together fairly easily. 100GbE ports are available will be mainstream soon as well.
To be sure, performance problems are often traceable to the software … no argument on that one from me! (Not my software, of course! )
But, a 40GbE port or multiple ones provides you with very high bandwidth … but I don’t see that it does anything for latency. Whether that matters or not depends on the application, of course, but there are many high transaction rate applications such as ERP systems where the limiting performance factor for many tasks is highly rapid random access to individual chunks of data. It is the randomness that turns the task into a large number of isolated tasks and makes latency the limiting factor in performance.
If there is a way around this, I would be very interested in hearing what it was.
<<<the winners in this kind of space will typically have to offer (mostly in order of necessity): excellent support when needed, an advantage in predictive failure analysis, tools for storage management as well as performance analysis and trending, ease of use, market-leading performance, easy buying/sales options, and strong ROI and price/performance metrics.>>>
That is what makes PSTG unique, like NTAP before it, PSTG offers all of that, and appears to be the best in the industry on all of these categories when it comes to flash storage.
And yes, EMC or NTAP can come out with new product. But can they actually survive business model wise by changing the expected replacement life to 10 years from 5 years, by removing the requirement for “forklift” replacements, by replacing flash for free, and all the other things that PSTG does?
Will it not require EMC or NTAP or HP to re-architecture their hardware and software to create the modularity and simplicity of what PSTG offers?hear
I don’t know, but this seems like the case since neither NTAP or EMC or HP have really emulated what PSTG does as of now. They have given here and there, but nothing really that similar.
Interesting, it shows that the processing itself is 6% more efficient, but that it is 3005+ more efficient when the data download is taken into account. I do not understand this aspect of their slides.
That slide, and the otherwise identical one before it, confused me to. Maybe I figured it out.
I think it assumes that your operation has vast amounts of data stored, more than can fit on the internal SSDs of a specific computer. That requires a step to copy data from whatever massive (slower, cheaper) storage you have to the SSDs of the system that will be processing it, followed by running the computations on that system. What they have done is make the massive storage system fast enough that you don’t need to copy anything to internal drives. All data is directly available, eliminating the performance hit from a mixed disk/SSD configuration.
If I get this right, the use of SSDs has been hobbled by them being replacements for disks. Connectivity has been based on or derived from disk connectivity. PURE started with a clean design that takes advantage of SSDs strengths. Disks are, inherently, optimized to give best performance from a point sequentially forward, but slow reaching that point. SSDs can start from any point, and then jump to any other point. BUT, SSDs configured as replacements for disks are responding to an interface built for disks. The clean design allows SSDs to be treated as direct-access devices at every stage of the process.
“My prediction is that we will see a 50 percent decline in flash pricing in about three years and a 75 percent decline in six years…But the important thing to realize is that flash doesn’t actually have to be cheaper than magnetic disk for the full transition to happen — it just has to be cheap enough for folks to be able to justify the conversion in terms of saving power, space, cooling, and manpower, and for folks to understand the business upside in moving to flash. … That ROI equation already works for the vast majority of important business applications today, and we think the pivot for unstructured data applications will happen much faster than people anticipate.”
Through our discussions and research it appears likely that Flash is the future of data storage across all markets. Great news for all creators of flash. Except, is it? Anybody excited to learn that Pure’s CEO believes what they sell will be for 50% less in three years and 75% less in six years. That answers the question as to whether flash will be commoditized in the coming years. According to Pure’s CEO, you betcha.
Hi darthtaco, I think the point he was making is that there are still huge amounts of simple things that people are still buying magnetic disks for… because they are CHEAPER. He’s saying that when the price of flash gets reasonably close to the price of magnetic disks, the magnetic disk market will collapse, and everyone will use flash, for everything…because it’s BETTER!
Saul
Darth,
thanks for that link to the British Communication company blog. There was an interesting tidbit:
Pure Storage has recently announced that their Net Promoter Score (NPS) has increased to 83.5…
Wikipedia defines NPS as :
An index ranging from -100 to 100 that measures the willingness of customers to recommend a company’s products or services to others. It is used as a proxy for gauging the customer’s overall satisfaction with a company’s product or service.
Pure’s increased NPS means that they rank in the top 1% among all B2B brands in terms of customer satisfaction. The industry average is only 16, so the fact that Pure ranks more than 65 points above it proves once more Pure’s dedication to their customers.
…but if the application is at the other end of a cable, I don’t see how you get rid of the inherent latency in the cable.
The cable isn’t going to introduce a lot of latency, at least not within a datacenter. Cat-7 twisted pair carries signals at about 75% of the speed of light. Cat-6a is around 65%, and plain old Cat-5e is 64%. I would assume that the Ethernet Mesh mentioned in the FlashBlade material would use cables no slower than those. The electronics that process the signals at either end of the wire will be the bottlenecks. https://en.wikipedia.org/wiki/Velocity_factor
Hlygrail, …but also carries all the other intrinsic benefits at a similar cost.
Do you really think EMC-Dell, HPE, NatApp will adopt the PSTG evergreen pricing model? Color me skeptical.
As for v2.0 product, Pure has already planned for this. They will swap out old flash for new. They will continue to enhance their s/w controller. And their customers will receive all the upgrades to their installed Pure storage free of charge. They only pay for new storage.
BTW, where I worked (big aerospace firm) the single biggest IT budget line item for several years running was for more storage. Storage requirements were not only the largest single budget item, but the fastest growing budget item. Still true? I don’t know, I retired 8 years ago, but my gut tells me if anything, it’s accelerating especially with IOT in its infancy. Yes, even when I retired my company was embedding devices in our products in order to monitor our products in the field.
Was this just where I worked, or is it a megatrend? Do you suppose Giancarlo knows this?
from it The all-flash array (AFA) market share still remains relatively small with only 6% of respondents declaring their organisations had deployed all-flash arrays.
plenty of room to grow. Spinning discs are an anachronism in an electronic world.
Tanhas,
Again, my info is dated but maybe still relevant. Some years ago I was one of several managers involved in a major technology insertion project. In a nutshell, the project was focused on migrating our mainline applications and databases from centrally located big blue boxes to distributed UNIX boxes.
The job I had was to develop the standards for application and database design in order to optimise for this environment. Invariably (I mean 100% of the time) when there were performance problems, and there were lots of them, it was due to bad design of applications and databases. I don’t recall a single instance of network latency or insufficient bandwidth ever being identified as a performance issue. And I worked closely with the manager who was in charge of the network side of the project.
OK, it’s stale information I know. Server performance has improved a lot and of course that ameliorates problems resulting from poorly designed apps and DBs. But, I am confident that the underlying problem persists. Only now, with much more COTS in the environment, it’s much harder to implement performance enhancements.
Saul, …it just has to be cheap enough for folks to be able to justify the conversion in terms of saving power, space, cooling, and manpower, and for folks to understand the business upside in moving to flash. … That ROI equation already works for the vast majority of important business applications today…
Actually, Giancarlo is saying that the long-term installed cost of flash is already cheaper than spinning disc. The only inhibition is that not everyone who makes a purchasing decision is aware of it yet. Probably a good thing, if they were suddenly inundated with orders they couldn’t fill, Pure’s customer ratings would falter.
Also, is anyone not going to buy Pure today because of what might happen 3 to 6 years down the road? OK, if you’re a dedicated buy and hold investor maybe pass it by. But aside from the reduction in input costs to PURE in that time frame, look at what’s going on today with video and music streaming. Think about what’s going on with “big data.” Think about the explosion in storage needs that will arise from IOT which is currently in its infancy. Think about how many captive customers they will have due to their evergreen pricing. Do you think those customers are going to abandon Pure for HPE due to a slightly better acquisition cost (if competitors can undercut Pure in the first place)?
There’s a lot more to consider than just the cost of flash in the future.
Incredible in depth knowledge on this board from various posters, more so than I have seen on some Companies own website. Can’t say I understand a lot of the very advanced technical stuff when it comes to flash and storage(do know Flash Gordon though), way over my head, however, it actually beats reading a book at times. Certainly been an education. Thank you.
Pure Storage has recently announced that their Net Promoter Score (NPS) has increased to 83.5… Pure ranks in the top 1% among all B2B brands in terms of customer satisfaction. The industry average is only 16.0.
I was thinking about this, and realized that Pure’s 83.5 is included in that industry average, pulling it up. That means the rest of the industry has aa average rating even significantly lower than that 16.0 score, implying that Pure has VERY happy customers, compared to the rest of the industry.
When I say “cable” I am referring to end-to-end. Compared to a local database, going across a cable involves building the message, getting it on the cable, going across the cable, and unpacking at the other end. The actual cable transit is going to be the least significant part of that.
Indeed, with the database that I have had the most experience, if a client is on the same system as the database, one can choose to connect via what is called a “shared memory” client, where the client reads and writes directly into the buffers of the server software, or via a client which communicates with the server using the same TCP/IP layers used for running the client remotely. The performance of the latter is dramatically lower than the former, even though both are on the same computer.
Like I said, Brittlerock, I agree enthusiastically that a great many performance problems are just bad code, but I work with DBAs that find and fix those problems and then find that there is a residual issue and with notable frequency that has to do with the company having implemented some form of networked storage. Amusingly, the networked storage is far, far more expensive than a set of in box SSDs.
What continues to bother me about the claims for Pure is that it is still networked and physics tells me that no matter what you do at the end of the cable and no matter how many parallel cables you provide, that true random access is going to be dramatically slower than SSDs which are in the box with the processor (being immensely cheaper, as well). For huge volumes of poorly structured data this may not matter, but for relational databases, it is almost certainly going to be a triumph of standard practice over performance.
I am a little confused what this has to do with PURE. You are bothered and yet every other business measure suggests the actual industry seems to know something more about its performance and value.
Their revenue is growing impressively, NTAP is wanting to compete and is actively converting its legacy systems, customers seem to be particularly pleased by their product and the technology is improving.
So the obvious question is how does this very knowledgeable industry, those actually in this industry, have it so wrong?
Pure, like NTNX, has grown faster than any company in storage industry history. Clearly both are doing something that is quite disruptive and missing from the incumbents. With Nutanix we know only VMWare stands in their way. With Pure, seems to be more competitors, but to date none of them have been able to come close to what Pure offers. What they do have is large customer bases who will be reluctant to leave their vendors, and they will make their products just enough better to try to fight off the new benefits that a Pure brings.
The question becomes, can the incumbents do enough to be good enough so as to remove Pure’s obvious advantages?
Over time I think the value proposition for Pure will grow, and that like with Arista, the technological limiters will be removed and maximized to the extent possible. Musk did this with Tesla, Arista has done this in SDN against one of the largest and meanest companies in the world (what, can’t Cisco just recreate its product? Yeah, and then have to lay off 50% of its sales force).
Can’t EMC or NTAP just redesign their entire business models? Answer that question yourselves.
So the obvious question is how does this very knowledgeable industry, those actually in this industry, have it so wrong?
Well, they have had it “wrong” in this particular sense for a long time. Various kinds of networked storage are sold to companies as universal good solutions and while they may in fact provide cost and management savings for word processing, spreadsheets, and the like they provide significant compromises for high transaction volume database applications. That doesn’t make it a bad product if one is knowledgeable about its characteristics and uses it accordingly, but it can mean that it is a bad product for some companies for their database applications.
In some cases, these characteristics don’t matter because the volume and critical speed of transactions does not require highest performance. Or, as been found by a number of companies using the old rotating rust type of networked storage with RAID to cuts costs and a cache to reduce apparent write time, they work OK during normal processing and then are terrible when it comes to an operation like restoring a backup which swamps the cache. Pure seems likely to not have that problem and so should work for a larger number of companies.
I don’t know that this is a factor which should retard the adoption of Pure in a meaningful degree … even if the industry were to actually wake up and pay attention to these issues, i.e., stop drinking their own koolaid. I have raised the issue because Pure is claiming latency speeds which seem dubious to me unless they are measured within the box rather than from the other end of the cable. It doesn’t help that Pure is missing from sources like the SPC-1 benchmarks http://spcresults.org/ apparently because the benchmark rules do not allow some of the “tricks” which Pure uses to achieve their results. That doesn’t mean that those “tricks” don’t work in practice, but it does mean we don’t have a standardized comparison.
I don’t know that this is a factor which should retard the adoption of Pure in a meaningful degree … even if the industry were to actually wake up and pay attention to these issues, i.e., stop drinking their own koolaid. I have raised the issue because Pure is claiming latency speeds which seem dubious to me unless they are measured within the box rather than from the other end of the cable. It doesn’t help that Pure is missing from sources like the SPC-1 benchmarks http://spcresults.org/ apparently because the benchmark rules do not allow some of the “tricks” which Pure uses to achieve their results. That doesn’t mean that those “tricks” don’t work in practice, but it does mean we don’t have a standardized comparison.
Which again gets back to the original question, you question the investment and the benefits of the company’s technology…are you the only one who gets this and yet NTAP and all its customers are being duped and all of the rapid growth PURE customers are being duped, etc.???
Your “koolaid” comment seems to suggest that all those database experts are in fact idiots.
Can you state your exact expertise in this arena and what publications you have on this matter and what companies you own that have dealt with this issue and what specific software you created that deals with this topic specifically?
)