Vendor Fair Application: Fill & Download for Free

There's a lot of different facets to consider, and which ones are really the most important is fairly application specific. That said, here are some thoughts:1) Query latency: This is the most obvious, but how long do you wait between asking the question and getting the answers? Test a variety of types of queries, because some queries may run super fast, while others may be slow as a dog.2) Concurrency: How many users are going to be using the system at any given time? If the answer is one, pretty much any system will qualify. If the answer is hundreds, some may falter.3) Compression: How much space on disk will your dataset require? You should probably consider redundancy for high availability, any indexing structures necessary, etc. More disks means more money shelled out, although disk is pretty cheap these days, and probably isn't the limiting factor. However, compression can also play into performance, since databases have historically been bottlenecked on disk access.4) Cluster size: How big does the cluster need to be? If a 5 node cluster for one system performs as well as a 20 node cluster for another, that can mean a big difference in total hardware cost, at scale. How well does the system scale? If you're evaluating an MPP system, you should absolutely test clusters of different sizes, to get a sense of whether that system you're looking at really scales linearly like they say it does.5) Ease of administration: How easy is it to keep a cluster going? Will it fall apart under heavy load? What kinds of administration tools are offered for you to manage the system. Be careful in POCs, especially if the vendor is hesitant to let you play with the cluster yourself. Beware of operators behind the curtain.6) Adaptability to changing workloads: Nobody's requirements stay the same forever. How easy is it to reoptimize the database if you have a drastically different query that comes into the picture?7) Elasticity: Can you add nodes easily? Remove nodes? Replace nodes? How hard is it to redistribute data across an MPP system?8) Load speed: Some applications do nightly bulk loads, some applications do trickle loads throughout the day. If you're doing a nightly bulk, load speed may not be as important, but in a system where you trickle data in, it's useful to know how loads will affect running query performance.9) High availability/Disaster recovery: What happens if you lose a node? What happens if you lose a data center? How hard is it to recover, and how fault tolerant is the system?I'd have to think a bit to come up with some more, but this is a probably a decent start for building POC criteria. Some of these ides bleed into other ones I've listed, so it may not be comprehensive, but you get the idea.

As long as the Federal Government of the United States was paying for it, any use of the Internet and its predecessor outside the strictures laid down in the relevant government contracts was questionable at best, and either of "misuse of federal funds," or "unfair competition" to private companies offering the same/similar services, at worst.First came the ARPANET in 1968. Your university had to have a (Department of Defense) Advanced Research Projects Agency (ARPA) research grant (contract) to be connected to the ARPANET. It was quite expensive, and the fastest thing the monopoly telephone companies will sell anyone for long haul bandwidth are 56Kb/s leased lines. They refuse to part with T1's (1.5Mb/s). This predates by more than a decade the 1984 breakup (and subsequent reformation) of the AT&T (company) "Bell System" monopoly.Odds were high that the first thing you (Professor Principal Investigator) had to do was get your graduate students to write the operating system software to implement NCP and BBN-1822 (not to mention the other protocols) to make use of the network because the vast majority of computer manufacturers have no clue whatever (well, OK, Digital Equipment Corporation (DEC) had something of a clue early on). In fact, you might even have to hie yourself over to the Electrical Engineering department and get them to design/build a BBN-1822 interface for your computer so it can actually connect to the Interface Message Processor (IMP) that ARPA plonked down in your machine room ...Naturally, you would never do anything with your ARPANET connection that might bring the ire of U.S. Senator William Proxmire (D-WI) down upon you & your university for misuse of federal government funds. This is the age of the "Golden Fleece Award," after all.However, there's a problem: the ARPANET is a new communication medium. What do humans do with new communication mediums? The same things they do with old communication mediums: they exchange all kinds of information. Technical debates. Jokes. Amusements. Serious scholarly work. Flirtations.The trick, of course, is to keep the "fun, but not wholly appropriate" kind of ARPANET use "on the down low" so that the people who are paying for it all (the federal government) don't notice, or don't get too upset when they do notice. You don't want to be entirely humorless, do you? What sort of work environment for creative, intelligent people does that foster?Lots of blind eyes are turned towards things that should not be seen.A text parody of a 60 Minutes investigation into the ARPANET from the period might be illustrative: http://www.textfiles.com/humor/COMPUTER/60_minut.arp [I first saw that in the HUMOR directory on MIT-AI, long ago; references therein indicate that it was written in the late 1970s]Then ARPA handed operational management of the ARPANET over to the Defense Communications Agency (DCA). Those guys are serious: they have guns and nuclear weapons. You don't muck about with them. The people in DCA are also preternaturally aware of the medium that they're managing; an apocryphal quote from an officer: "I write all my E-mails as if they will show up on the front page of The New York Times tomorrow." (No, I don't think that was Oliver North).During this period, it is noticed that the ARPANET is going to outstrip the IMP address space (8 bits IMP, 8 bits of host address per IMP) quickly, and plus there are these LAN technologies like Ethernet showing up everywhere for these ... "workstation" computers. Why, it seems pretty clear that we're going to have one on every desk!We need new network & transport protocols.Enter TCP/IP. It's the replacement for NCP, and it has lots of new capabilities, including the ability to "gateway" network "packets" between different physical networks! ("gateway" is the old tongue for "router").Wow, with these new protocols, a computer being "on the net" isn't restricted to being physically connected to an IMP! That Ethernet workstation on my desk can be a full fledged "host" of its own on a network of networks, all engaging in inter-networking! Hmm, there must be a shorter word we can use to describe this ... maybe "Internet"?Except that now our slaves – [cough] I mean, "graduate students" – have a whole lot more software to write, again, to implement these new protocols for all the computers and operating systems out there ...(quick aside; this is also the time of Berkeley Software Distribution (BSD) (BSD) UNIX development at University of California, Berkeley - ARPA funded it; see Open Sources: Voices from the Open Source Revolution)DCA set a "flag" conversion date: January 1, 1983. Everyone is supposed to have their TCP/IP implementations done (tested & deployed) by then, and on that date, the IMPs will stop passing NCP packets around. Yup, they're centrally controlled enough to make that happen.Oh, wait, you're serious? We're not ready! Please give us more time!OK, two months grace, but on March 1, 1983, NCP will be turned off, for sure. Remember, we have guns. Or maybe this just means your university isn't going to be on the network anymore.This is the birth of today's Internet. The 30th birthday was New Year's Eve, 2013.This is also an important learning experience: never set a "flag day" conversion on an international holiday, especially not one during which the responsible people will likely be nursing hangovers ...While we (DCA) are at it, we're going to split the military sites on ARPANET into their own separate network, the MILNET. After all, those long-haired hippies at the universities are a security risk. In fact, after October 1983, the "gateways" between the ARPANET and the new, split-off MILNET will only pass Simple Mail Transfer Protocol (SMTP) traffic which is why we're calling them "mail bridges" (this can be viewed as the first network "firewall").Wait, what do you mean we can't do that, and it cripples your ability to do your jobs? You mean we have to allow all the IP packets through the mail bridges? [sigh] Oh, alright. They'll still be proper IP network gateways, passing all packets.Enter the National Science Foundation (NSF). They take over operational management of the ARPANET from DCA; DCA keeps MILNET, which becomes the Defense Data Network (DDN). The lucky Department of Defense is now off the hook for whatever happens on the ARPANET. Whew!It's becoming clear that we're going to need a new Internet backbone - those aging ARPANET IMPs can't handle it any more, and besides, they're only connected to each other with those thin 56Kb/s leased lines, and we've got 10Mb/s Ethernet LANs everywhere. Many orders of magnitude bandwidth mismatch? Oh, yeah.First cut, Dr. Dave Mills (University of Delaware) makes routers out of his "fuzzballs" (DEC PDP-11 minicomputers with his own, home-grown OS, using "delay" as the routing system metric, which requires some quite extensive time & frequency measurement, along with some atomic clocks attached ... care to guess where the Network Time Protocol came from?). That was the first NSFNET Backbone.Unfortunately, while the Fuzzballs are an improvement on the IMPs, they're still not enough, so the NSF turns to IBM who pitches an idea to make a routing cluster out of computers (with IBM token-ring interconnect!) with their brand-new RISC CPUs in 'em, to connect to 1.544Mb/s T1 leased lines: a Network Switching System (NSS). The NSS nodes are a improvement over the 56Kb/s stuff ... once IBM finally makes their NSS clusterf*ck actually perform at T1 line rate.Sometime during this period, the old ARPANET is decommissioned, and the IMPs turned off. IMP #1 (from University of California, Los Angeles) is in the Smithsonian Institution now.Oh, what about "commercial" use of the Internet? Um, well, still officially prohibited during this period, since the NSF is paying for all that long-haul NSFNET bandwidth with federal funds, and the commercial X.25 network service vendors (TELENET, TYMNET) might get ticked off at the federal government for "giving away" a service that they sell quite dearly. Worse, the Internet, to the extent it's priced at all, is flat-rate! No charging by the byte! No asking, "mother, may I?" when you want to deploy a new application or protocol! (this question of bandwidth pricing is at the heart of the arguments over Net Neutrality; see Has net neutrality always been around since the internet was created?).However, if you're a commercial company on the Internet with wares to sell to the research and university community that's "in support of" their research, isn't that justifiable? Does it matter that RFQs, price quotes, purchase orders, and sales negotiations go on in E-mail over the Internet? Can't I FTP a new software release for my computer from the vendor who's now on the Internet? Isn't technology transfer and public/private cooperation a good thing?Of course, the envelope on "non-commercial use in support of research or education" is being pushed by darn near everyone all the time at this point, and the NSF sees it's in an untenable position: the Internet has to be properly privatized ASAP. Heck, it was during this period that I and some friends deployed NNTP to (among other things) give the Usenet a new backbone: the Internet (see How was the "Usenet 8" newsgroup hierarchy standardization developed and decided upon? for a description of what I was trying to do). Push the envelope much? Uh ...Things go bumpily by half-measures during this period. UUNET (a commercial UUCP service network hub that later became AlterNet, which was swallowed by MFS, which was swallowed by MCI, which was swallowed by WorldCom, which renamed itself MCI after scandal) and a few of the NSF "regional" networks (PSInet (formerly NYSERNET), and CERFnet) collaborate to start up the Commercial Internet eXchange (CIX, pronounced "kicks") for Internet Service Providers to connect to and not worry about the "commercial use" restriction through that venue, but ... how do you know your IP traffic is being routed that way if you're interacting with a company still connected to the NSFNET? [Shhhh!]Finally, the NSF, to the howls of the Internet community, more or less hands off the NSFNET backbone to a commercial spinoff of IBM: Advanced Network & Services (ANS). ANS tries to play like a monopoly telecommunications company (with all the haughty arrogance that implies), and competitors spring up rapidly. ANS is competed into well-deserved irrelevancy.Thus is the backbone of the Internet privatized, and the "non-commercial use only" policy dies. After all, it's no longer being paid for by our federal tax dollars now!The NSF still funds some network backbone bandwidth, but only between the National Supercomputer Centers and the various university research grantees/partners, all of whom are also connected to the "commercial" Internet, so it's easy for them to route just the traffic between those institutions over the NSF (federally) funded bandwidth. Effectively a private network; no "transit" traffic.Sometime during this period, Netscape puts Secure Sockets Layer (SSL) into its Web Browser and an implementation is put into the Apache HTTP Server, and, despite National Security Agency (NSA) and Federal Bureau of Investigation (FBI) protestations, encryption begins to be commonly used on the Internet. This gives people the confidence to allow their credit card numbers to be transmitted over the Internet.Thus began E-Commerce as we know it today.That was also the impetus for the 1995-2000 Dot-Com Bubble and Dotcom Crash.

Define ‘justify’ - evidently the second part of your question does that, i.e. retail businesses (distributors or resellers) augment the ability of a factory (vendor) in placing products with end users.They have a local phone number just about everywhere there are customers. They place a few orders for large amounts, which factories find to be a great time-saver. The factory ships large quantities to few addresses, also a very good thing in their eyes.Resellers answer the questions of thousands of individual customers, again relieving the factory of a large burden.My experience in commerce forty years ago is still fairly applicable today, especially with complex specialized products. Apple has in-house distributors, in other words they are both the factory and the local retail outlet. That’s how Hi Tech works.But factories that live one rung lower on the food chain making more generic, interchangeable items, have to find someone to buy in bulk and sell in onesies.Back in the day the factory’s cost per unit was M L B - material (stuff coming in the door that got made into things) labor (people working in the factory to make things) and burden (management, retirement, insurance, keeping the light bill paid, shipping products out into the sales channel.)MLB tended to be about 35% of retail. If it costs the factory $35 to make one widget, you need to sell it for about $100 to pay for all the other pieces - engineering, marketing, legal, corporate, and HUGE numbers of people needed to sell and fix widgets in the field. With MLB of $35, the gross corporate profit usually fell below $10, and the net corporate profit was a lot smaller due to taxes, dividends to shareholders, and many smaller yada’s.When you are the reseller, you may have the privilege of buying a hundred dollar widget for, say, forty or fifty dollars, and you please the factory by surviving to buy more next time by [[ IF YOU ARE GOOD AT RUNNING YOUR BUSINESS ]] keeping five or ten of the hundred dollars you receive as retail. Less markdowns, defective merchandise, returns, and your own set of yada’s.Justify? Fair enough. I prefer “maintain functional relationships that survive competitive pressures.”