You spin me round baby...
...and so it happened that the Great Unwashed Masses were thrusted onto the new sparkly "digital highway". Where the Great Thinking Minds had decided that Everything Could Be Done. You wanna buy groceries? You can do it from the comfort of your house, on your computer connected to the network, you don't need to walk 200 mts. to the supermarket at the corner of the street. You wanna buy books that you'll never ever read? That too can be done from your couch. Wanna get a pizza so you don't need to walk to the kitchen and nuke the food you bought from the supermarket in the microwave? Yep. Use your computer. Wanna get a new couch to replace the old one that is covered with the books that you'll never read and the rest of the pizza? Yeah, go ahead, just order a new one. Wanna go to take a dump... Eh... No, you have to get up for that. Unless you can time it with the delivery of the new couch.
And all this was apparently fantastic! Until the Great Unwashed Masses were relatively small.
But the problem is that all that shit has to be done by a computer somewhere and when too much of the "masses" try to pass through the digital funnel, the computer start behaving badly. And to complicate the problem all the corporation out there had listened to the Great Thinking Mind but only to the bits that they liked to listen to. Basically they've heard a lot of "money money Money MOneY MONEYMONEYMONEYMONEYMONEYMONEYMONEYMONEY"... and several... heummm... "uncomfortable details" have not been understood.
Like the fact that all that stuff required server to work and servers required maintenance and management.
Then another revo-volution! Somebody invented Virtual Machine. That are real machine that pretend to be other machines.
And when all is said and done, at the end of the day somebody still need to maintain and manage all that stuff. No, it doesn't work like magic. Even if the Great Thinking Minds were trying to pass it like it.
One of the (supposed) advantage of Virtual Machines is that you can create a large number when you need them and then simply delete them when you don't need the anymore. Not like the real thing that you have to pay for and then pay again when you want the old carcasses disposed of.
That is nice... when it works.
Sure, you can have a wonderful "load balancer" that is there and listen and distribute the "load" on a number of virtual machines behind it, and when you need more machines you can do a Gandalf and "magic" some more, but... You also need to remember that the machines are using, somehow, stuff that is real and whoever is providing you the support will want money for that. Ah and you also need somebody that install, maintains and manage the whole shebang.
And all this doesn't guarantee that everything works as you want because... physics.
And after this historical introduction, we talk about $wearentmagicianyouknow, a company that was doing... different things.
One of their problems was that the extent of their IT experience was the same as my cat: computers are warm and is good to sleep on them. The second problem was that they had listened to some conslutant that enticed them into the "savings" that could be had and then presented his invoice.
After being caught in the trap, all the ULs turned to us to get the whole thing fixed in a way to produce all those immaginary profits. So far the only real things were the expenses.
A nice half-april morning (I think) DB came to ask why wasn't I hurrying to the meeting room for the meeting with UL1 and UL2 of $wearentmagician.
Me - Maybe because I wasn't invited?
DB - What?
(he check the applebook)
DB - Oh, I forgot to send the mail...
Me - And anyway, what do they want? I don't know anything about them..
Anyhow, we get in and UL1 start talking about "electronic marketing strategy" and "technique to retain the audience" and "scalable infrastructure" and "e-commerce dynamics" and "elastic redistribution of the load" and such.
DB - (looking at me) What do you thinl?
Me - (looking at my paper) ...I miss "blockchain" and "neural network" and then I have jackpot.
UL1 - What?
Me - Nothing... But seriously, what has this to do with me?
UL2 - What? Didn't you see our dynamic-self-sizing architecture plan...
Me - No I didn't because probably was in the mail that I didn't got.
DB - Hemmm...
UL1 - Anyway, what we want is to get the maximum efficiency in the customer's request management maintining the lowest structural footprint.
Me - "footprint"?
UL1 - Yes, to get the best ratio of utilized resources compared with the functionality of the information processed by...
Me - So to spend as little as possible.
Yes, even if you try to run around it for a couple of hours tossing buzzwords like there is no tomorrow, in the end is what you're talking.
After some attempt at faking technology knowledge that nobody believes and nobody cares about, after a while we started talking about serious stuff. And a more "standard" thing couldn't be.
These people had a system that managed a bunch of different web site and web applications. In an attempt at paying as litt... I mean "optimize the resource distribution to improve the position in the efficiency curve for the maximal return of investment", they ended up with a thing that worked... sometimes... until it doesn't.
Then some conslutant began to talk about "dynamically sizing the structure" and they swallowed the bait, the hook, the sinker, the rod and the fisher at once.
UL2 - So with this "dynamic resizing" what we want to get is to optimize...
Me - Yeah, you want that when the load on the server increases more services are started to split the load and then those are discarded when not needed to spend as little as possible.
UL2 - Hmmm.... we prefer the term "optimize the investment"...
Me - Whatever float your boat. Doesn't work.
....silence....
UL1 - What do you mean?
Me - From what I understood of your system (looking at DB's applebook), how is it configured and what is the load and activity... What you want to do make no sense and is probably going to get the opposite effect: make you spend MORE money instead of less.
UL2 - But... the efficiency... dynamic... adapting...
Me - Nice words but they don't mean what you think they mean.
DB - Maybe if you could explain...
Me - Ok, let's see...
I get up and go to the whiteboard, with great shock of UL1 and UL2 that were expecting a presentation on the big screen. I grab a marker.
Me - So, what I understood of all this mess of buzzwords and dancing around with words that sounds great but means nothing, is that you got is a system tha usually works fine, but then it doesn't. You have peaks of activity that send the entire thing in a coma.
UL1 - Exactly! That's why we want to dynam-
Me - What you want to do is to add more servers when you get the peak and zap them when the peak is gone. That is a nice idea, but won't work on your system. And that's why.
Scribble something like this:
Me - This is the ideal situation for every system. The load fluctuate between two limits that are kinda close together. At this point you can size the sytem for the average load, so you use most of the time the maximum of your resources and when it goes beyond that is "loaded" but not that much.
Another scribble.
Me - This is another normal situation, the load is relatively low but for some times it gets higher. This situation is ideal for what you want to do, when the "plateau" approaches you can spin more machines and when it is gone you can spin them down.
UL2 - Right! That is what...
Me - BUT! Your situation aint' this. What you got is like this:
Me - In this situation, the system is most of the time doing very little because is sized for an higher load, but then it is subjected to very heavy spikes of activity that bypass a lot its capacity. In this situation what you want to do, doesn't make any sense.
UL1 - But...
Me - IT DOESN'T MAKE ANY SENSE because the spikes are too short to be managed this way.
(pointing to one of the spkies)
Me - When the activity start rising, the system decide that it needs more capacity, but the capacity cannot be added instantaneously, the only way to do so is to have it already available. The moment the capacity is added, the spike is gone and the new capacity is left unused, until it gets removed and then a new spike came along and it is back to square 0.
The end result is that you don't get any advantage but you still have to pay for the extra resources that are not used.... I mean "needlessy exceed the optimal capacity without improving the resource usage".
UL1 and UL2 seems to ponder the thing a bit.
UL1 - But if we look at the first graph... that looks almost like the last one except... the spikes are higher...
Me - Yes, that's the problem.
UL1 - And if we treat the thing the same way...
Me - In this case you need to size the system for about 80% of the maximum load, that result in a system that is waaay oversized for the most of the time and is still undersized for the top load. You'd spend more and save nothing.
UL2 - So...
UL1 - The problem is that this thing is too slow as it is. I mean, it works fine most of the time and then nothing works anynore.
Me - It works fine when you're in the situation of low-load, then you hit a spike and everything slow down
UL2 - So... how do we fix it?
Me - (scrolling my shoulder) The best option is to analyze your system and identify WHAT is causing those spikes, is some marketing activity that attract a lot of visitors? Some other process? If those things can be split out on a separate system then it is possible to move to one of the other two scenario. But first you need to identify the problem.
UL1 and UL2 are shocked and horrified by the idea that they have to do something that sounds a lot like work.
UL1 - But...
Me - What?
UL1 - If we do this thing of the dynamic loading...
Me - (looking DB) I can go rigth? We all knows how is gonna end...
Davide
11/07/2019 10:27
