6 ms·

I just want to make a comment about optimizing applications even though the article is about optimizing organizations: The way to arrive at the optimal system

I just want to make a comment about optimizing applications even though the article is about optimizing organizations: The way to arrive at the optimal system is to continually optimize all individual components as the system develops. You have to walk the razor’s edge between “premature optimization is the root of all evil” and not optimizing anything until it’s too late and a bunch of bad design decisions have been baked in. If you want to write the fastest program, profile its performance from the start and try different ways of doing things at each step of the process. There’s usually a way of doing things that maximizes simplicity and performance at the same time because maximum performance == doing as little work as possible to complete a task. Chances are that ‘littlest amount of work’ will be elegant.

llamaimperative 2y ago
Ehhh this isn’t quite the right takeaway, or at least it’s contrary to Deming’s approach. The key insight from Deming’s work is that at any given moment there is only and exactly ONE thing that should be optimized. Once you optimize that, there will be a new SINGLE thing that is now slowing down the entire system, and must be optimized. The goal of an engineer of a system (or manager of an org) is to develop a method to repeatedly identify and improve each new bottleneck as it appears, and to ignore other components even if they’re apparently not performing as well as they could.
- Jensson 2y ago
  That is what everyone currently and we know the results, that only takes you from horribly slow to very slow. In order to have a chance to get anywhere close to fast you need each component to already be very fast, and then you can build a fast system on top of those. When you work with slow components you wont use the right architecture for the system, instead you start working around the slowness of each component and you end up with a very slow mess. Example: A function call is slow, it calls a thousand different things so you speed this up by putting a cache in front, great! But now this cache slows down the system, instead you could have sped up the function calls execution time by optimizing/trimming down those thousand things it calls. Adding that cache there made you get further away from a fast system than you were before, not closer. One cache is negligible, but imagine every function creating a new cache, that would be very slow and basically impossible to fix without rewriting the whole thing.
  - rudasn 2y ago
    OK so the cache is the one thing, and the next most expensive "little thing" is the one thing to focus on next?
    - Jensson 2y ago
      But removing the cache makes the system slower, it didn't optimize anything. So no, fixing this isn't possible with that approach, you need to take the holistic approach.
      - llamaimperative 2y ago
        There's nothing about identifying and solving one problem at a time that prevents one from taking a holistic approach. In fact the entire point is to interpret the behavior of the entire system in order to find the right singular intervention. To find the true globally optimal point of intervention, one must look at the entirety of the system.
        Jensson 2y ago
        If you are saying that your method includes all methods, then sure being open to consider all methods of making something faster will include the best way to do it. But I thought were were talking about useful advice here, to be useful you have to narrow it down a bit.
  - llamaimperative 2y ago
    "Everyone knows this" yet you don't actually see organizations work like this almost ever in practice. Especially not software organizations. Instead you see big splashy initiatives to "overhaul performance" or whatever. There is definitionally no more efficient way to improve the performance of a system than sequentially targeting each new bottleneck in its performance and nothing else. Your example is just a case of picking a non-ideal method to improve a bottleneck. It's a lot easier to get this right when you're focusing on one important problem instead of generally "optimizing everything," which produces a clear incentive to take easy, immediately available, probably-not-ideal solutions all over the place.
- RaftPeople 2y ago
  > The key insight from Deming’s work is that at any given moment there is only and exactly ONE thing that should be optimized. Once you optimize that, there will be a new SINGLE thing that is now slowing down the entire system, and must be optimized. Are you sure this is what Deming thought? A complex system has many different things that can be locally optimized independently, here's an example: 1-DC picking optimization 2-Store inventory placement to align with demand to both increase sales and decrease costs of liquidating old inventory Both of these have very significant impact on costs or revenue or both, and are largely independent. The process of picking in the DC is independent of the specific selection and inventory levels of skus in the stores. Why would you not do both at the same time to get the benefits as quickly as possible?
PaulKeeble 2y ago
Usually "optimisation" is something that happens once a slow part of the system has been identified. So you develop data and expose the problem with a profiler and fix a few things that seem to be holding it up and then find the next stage of bottlenecks. You keep doing this until you meet your performance goal or you can't improve it any further and determine a completely different approach is required. Personally I prefer to recognise those components ahead of time and think about performance and do experiments to start from an architecture that has a better chance of succeeding in its performance goals from the outset. So I tend to agree with the article that its much more effective to optimise at the architecture than the individual components but its also much harder to do that once the thing is built and working already and the same is likely true for a lot of organisations as well. Once organisational culture has been set its hard to fix it.
- drewcoo 2y ago
  > Usually "optimisation" is something that happens once a slow part of the system has been identified I wonder if that's because speed is easy to measure. It's certainly not the only thing that can be optimized.
- ZoomerCretin 2y ago
  Finding and fixing the one and only source of poor performance is a minority of optimizations. The more common case is a lot of suboptimal code creating poor performance.

llamaimperative 2y ago

Ehhh this isn’t quite the right takeaway, or at least it’s contrary to Deming’s approach. The key insight from Deming’s work is that at any given moment there is only and exactly ONE thing that should be optimized. Once you optimize that, there will be a new SINGLE thing that is now slowing down the entire system, and must be optimized. The goal of an engineer of a system (or manager of an org) is to develop a method to repeatedly identify and improve each new bottleneck as it appears, and to ignore other components even if they’re apparently not performing as well as they could.

Jensson 2y ago
That is what everyone currently and we know the results, that only takes you from horribly slow to very slow. In order to have a chance to get anywhere close to fast you need each component to already be very fast, and then you can build a fast system on top of those. When you work with slow components you wont use the right architecture for the system, instead you start working around the slowness of each component and you end up with a very slow mess. Example: A function call is slow, it calls a thousand different things so you speed this up by putting a cache in front, great! But now this cache slows down the system, instead you could have sped up the function calls execution time by optimizing/trimming down those thousand things it calls. Adding that cache there made you get further away from a fast system than you were before, not closer. One cache is negligible, but imagine every function creating a new cache, that would be very slow and basically impossible to fix without rewriting the whole thing.
- rudasn 2y ago
  OK so the cache is the one thing, and the next most expensive "little thing" is the one thing to focus on next?
  - Jensson 2y ago
    But removing the cache makes the system slower, it didn't optimize anything. So no, fixing this isn't possible with that approach, you need to take the holistic approach.
    - llamaimperative 2y ago
      There's nothing about identifying and solving one problem at a time that prevents one from taking a holistic approach. In fact the entire point is to interpret the behavior of the entire system in order to find the right singular intervention. To find the true globally optimal point of intervention, one must look at the entirety of the system.
      - Jensson 2y ago
        If you are saying that your method includes all methods, then sure being open to consider all methods of making something faster will include the best way to do it. But I thought were were talking about useful advice here, to be useful you have to narrow it down a bit.
- llamaimperative 2y ago
  "Everyone knows this" yet you don't actually see organizations work like this almost ever in practice. Especially not software organizations. Instead you see big splashy initiatives to "overhaul performance" or whatever. There is definitionally no more efficient way to improve the performance of a system than sequentially targeting each new bottleneck in its performance and nothing else. Your example is just a case of picking a non-ideal method to improve a bottleneck. It's a lot easier to get this right when you're focusing on one important problem instead of generally "optimizing everything," which produces a clear incentive to take easy, immediately available, probably-not-ideal solutions all over the place.
RaftPeople 2y ago
> The key insight from Deming’s work is that at any given moment there is only and exactly ONE thing that should be optimized. Once you optimize that, there will be a new SINGLE thing that is now slowing down the entire system, and must be optimized. Are you sure this is what Deming thought? A complex system has many different things that can be locally optimized independently, here's an example: 1-DC picking optimization 2-Store inventory placement to align with demand to both increase sales and decrease costs of liquidating old inventory Both of these have very significant impact on costs or revenue or both, and are largely independent. The process of picking in the DC is independent of the specific selection and inventory levels of skus in the stores. Why would you not do both at the same time to get the benefits as quickly as possible?

PaulKeeble 2y ago

Usually "optimisation" is something that happens once a slow part of the system has been identified. So you develop data and expose the problem with a profiler and fix a few things that seem to be holding it up and then find the next stage of bottlenecks. You keep doing this until you meet your performance goal or you can't improve it any further and determine a completely different approach is required. Personally I prefer to recognise those components ahead of time and think about performance and do experiments to start from an architecture that has a better chance of succeeding in its performance goals from the outset. So I tend to agree with the article that its much more effective to optimise at the architecture than the individual components but its also much harder to do that once the thing is built and working already and the same is likely true for a lot of organisations as well. Once organisational culture has been set its hard to fix it.

drewcoo 2y ago
> Usually "optimisation" is something that happens once a slow part of the system has been identified I wonder if that's because speed is easy to measure. It's certainly not the only thing that can be optimized.
ZoomerCretin 2y ago
Finding and fixing the one and only source of poor performance is a minority of optimizations. The more common case is a lot of suboptimal code creating poor performance.