Ten years ago high performance computing (HPC) vendors included the likes of Cray, SGI, Compaq/DEC, NEC, and several other companies. Today for large servers we are primarily down to IBM, Intel-based platforms, and the emerging new Cray, Inc. This huge change for HPC over just 10 years is characteristic of almost everything we use in the data center, from HBAs to tapes to RAID, you name it.
Let's examine exactly why some storage companies fall while others rise and explore the process of what happens to the winners and losers of the technology race. It's surprising just how straightforward the process really is (baring a few bits of randomness here and there) and how it can be applied to almost all of our industry segments.
What Is the Mark of a Winner?
While picking future technology winners and losers is no simple task, there are a few points that I believe are universal constants in our industry:
- The “haves” (big companies with deep pockets but not necessarily the best technology) tend to win over the “have nots”:
- They win when times are bad as companies become more risk adverse
- They also win over a short time period, as they can price the “have nots” out of the market (I live in an area where a single airline dominates the market, and they continually price the competition out of the market)
- The “haves” often buy “have nots” when they (the “haves”) are far behind technologically
- Long-term R&D capital expenditure is critical to staying as one of the “haves”. Over the short term a company can market around a mediocre product, whereas long term that just doesn't work
- Not understanding the market, customers, and/or the competition can quickly turn a “have” into a “have not”
The market is unforgiving. Consider a couple good examples of companies that missed huge opportunities:
Xerox - The Palo Alto Research Center not only had the mouse, but also a windowing operating system — where would Apple and Microsoft be today without Xerox? And how would Xerox have fared if HP hadn't come out with a home printer? Without its introduction, Xerox might not have fallen from grace.
Control Data Corporation (CDC) - This company decided that no one needed a computer faster than the CDC-7600 and then let one of its founders and the inventor (Seymour Cray) of all of its computers leave — two moves that basically put the company out of business in the late 1980s.
These are just two of the companies that missed the mark in the 1970s and 1980s. CDC is effectively gone and is completely out of the computer business today, and Xerox, a giant of those days, is still struggling (although less than they were 18 months ago) as a much smaller company.
My point is that this is a continual cycle. We've seen the cycle occur in the 1970s, 1980s, and during the dotcom bubble of the late 1990s, and we will continue to see it over and over again. This cycle is part of progress and part of our human evolution.
HBAs – A Case Study
In September 1997 I got to test one of the first Fibre Channel HBAs attached to a FC RAID that was available with a working driver for Solaris. The company that produced the product was called Jaycor at the time and is now known as JNI. They quickly came to dominate the HBA market, but today they command less than 20% of the market, with Emulex and Qlogic controlling over 80% of the market according to the Gartner group.
That raises the critical question of how do you go from complete market domination, which JNI had in the late 1990s (and was still the market leader even as recently as 2000), to a company struggling to maintain a much smaller market share?
There are a number of points that are significant to note in this and similar cases, as such a fall from grace is common throughout our industry today and has existed in the past and will exist in the future.
- JNI depended on other companies for its HBA parts. Specifically, JNI used the HP Tachyon Fibre Channel chipset, which meant the following:
- HP possessed the ability to produce its own HBAs and compete with JNI (and did)
- At the time, JNI depended on HP to move forward with full fabric and other features. This, of course, meant that if HP was late with a feature addition, JNI was as well.
- Any issues within the HP fibre chipset also became JNI issues. For example, you could not make I/O requests over 8 MB, as the chipset could not DMA more than that amount
- OEM relationships were beginning to be developed by other vendors with major server vendors. SGI teamed with Prisa (and then re-teamed with Qlogic when the company acquired Prisa), Emulex teamed with IBM, and Qlogic teamed with Sun. JNI did get an agreement with EMC and other storage manufacturers, but not with any server vendors. Did this happen because:
- As the first vendor to market, JNI had a virtual monopoly, and they figured they did not have to adjust their prices for OEM agreements while newcomers did?
- Did JNI believe the best market strategy was to have a direct sales force?
- Who was going to use the technology, and with what components? Back in the early days of Fibre Channel, many things did not work together, and certification of RAIDs, Fibre Channel hubs, and early switches was a requirement. Figuring out which storage vendors to work with, what worked, and who needed that storage on what operating system were big issues. If I only have X number of good device driver software engineers, I need to choose where to spend my time given the current and anticipated market directions.
- How can you compete with better technology if the standard limits you? From what I have seen, once something becomes a standard, it becomes, by default, a commodity part. Once Fibre Channel standards were released and were working, as a seller the question then became what is our value add compared with the competition? You cannot go faster than the standard, you cannot change the standard, and you cannot be different than the standard. It therefore becomes a race to see who can manufacture the standard commodity product at the lowest cost and keep up with the standards as they are released.
Given all of the industry happenings, what happened to JNI is not at all uncommon in this industry. Qlogic developed its own chipset, which became used in RAID controllers, Fibre Channel switches, and other HBAs as an alternative to HP Tachyon. JNI now develops its own Fibre Channel chipsets as well, but I do not see them in the broad switch and/or HBA market selling their chipsets. HP is now Agilent and is getting out of the HBA business.
What will happen to Emulex and Qlogic, as they are now the two dominant players? I will leave it up to you to do your on analysis. Will we have only one dominate vendor, will we continue to have two, or will we see something different emerge?
Analyzing vendors and technologies is a difficult and rather tricky science. Of course, you have industry pundits, stock analysts, and others (like myself) making predictions all of the time, but how many of these predictions will be correct six months or one, two, three, even 5 years out? And no matter how much analysis and planning goes into a prediction, a bit of chaos theory will always come into play when you are dealing with people. Being in the right place at the right time is another important factor that's difficult to account for.
Still, the real issue comes down to the “haves” and the “have nots.” Of course, just because a company is, for now, one of the “haves,” an industry leader, and at the forefront of technology does not mean that it cannot become a “have not” extremely quickly. Sure you have WorldCom/Enron-like criminal behavior that can take a company down very quickly, but the same result can also happen purely from a technology fall from grace.
Often, though, as you have seen in the above example, the problem is quite different. A common case is the company forced to change from a technology leader blazing the path for others to what is now a reseller of commodity products. Some companies understand this evolution better than others and are able to seamlessly make the transition from leader to commodity reseller. This happens for a variety of reasons — sometimes it's due to marketing, sales, or engineering, or perhaps the ever-popular corporate leadership causes it. Whatever the cause, whatever the reason, the result is that vendors come and vendors go.
What can you do with all of this information? The good news is that for the most part you need not have to do anything. The market is self regulating, and you can often completely recapitalize equipment long before companies come and go. So for the most part, dying companies are not an issue, as hardware and software is generally recapitalized long before companies fail. The key term here, though, is generally, as this is not always the case. Here are a few examples of services that are difficult to transition:
- A multiple Petabyte HSM archive – This is really hard to just switch over every few years given the time for migration
- Tape backups that might need to be saved for years for legal reasons – Finding tape drives and backup software might be difficult if you are constantly switching
- Large proprietary databases – Moving data around databases is expensive and time consuming
All and all, much of this market churning happens over long enough periods of time that the writing can be seen on the wall. The key is to be aware that it happens and will continue to happen; it's all part of human nature.