Planning for hardware capacity entails focusing on everything from server rack to network card, everything from making an accurate estimate of how much physical space you need to logical space and the connection infrastructure, all over a period of time when you will need to revisit it again. It is geared toward optimization rather than under-utilization. The cost benefit side of this consideration lies in effective planning, efficient management and governance, and, finally, in ensuring that, when capacity is nearing, resources can be deployed to ameliorate the situation. In doing this, all business concerns have to be considered; measurable forecasts have to be made by envisaging trends and what the future needs will be. Tools are available that can be deployed to model the analysis in obtaining other factors, such as scenarios that seem even far-fetched. In the plan, the need to give attention to scalable, stable, predictable, and supportable products has to be critically considered.

Scope of Planning

The first rule to observe when planning like this is to know that there is really no one size fits all capacity planning. An attempt at that will only end in a customized disaster. The better idea is to create an individually managed plan for each service you plan to have in your IT infrastructure. There are, in fact, hardware requirements for everything from planning network infrastructure, messaging systems, collaboration systems, database systems, web servers, and file systems to building the server rooms or data center. There is a general rule of thumb that, if you have some monster hardware, your problems is over. Well, while that sounds pretty much a good logic, it doesn’t mean your company can afford it. Following a simple basic rule of capacity planning, we can always be on top of the situation.

This diagram (left) represents a typical template. It represents the thought process. First, to know what is on ground, and to determine current demand, and map that to current capacity, and the difference is in the lapses that a recommendation fixes.

There will be scenarios where this can be a startup project, where there exists no previous demand and no current capacity, but a recommendation needs to be made. In this scenario, the bottom line has to be made.

Startup Company Hardware Capacity Plan

To understand how to make a hardware capacity plan for a startup, the following has to be known.

  1. Size and Budget—This implies knowing the human resource size of the company that will consume the services the hardware will provide and the location(s) in terms of: Are they housed in one building or are there other locations that will be a part of the setup? This will determine the type and quality of hardware to purchase. Most hardware has certain capacity limits. The likelihood is that the startup already falls within the acceptable limit of a particular hardware type. I have, however, seen a scenario where an active directory environment got saturated with less than 20 users. The mistake the person made? Running domain services on a single Pentium 4 CPU with 2GB of RAM. Nobody does that, not even the guys in Hollywood. Hardware comes with low-end and high-end capacity. Be sure to obtain the right kind for your environment based on your budgetary allocations. Identify the trade-offs and ensure to take deeper look at all the options.
  2. Business Services Requirements—This refers to the business applications that will run in the startup company. These include but are not limited to domain services, network services, and application infrastructure services (those that require disk reads and writes in heavy volumes in databases). Identifying all these business applications will not be a one-time event, because other needs will emerge that seemed unnecessary in the past only to become an urgency in the present and future. Knowing these will help shape the technical needs.
  3. Technical Services Requirements—This refers to all the nitty-gritty details on a per-application basis. For example, domain services require that all network interface cards (NICs), also known as network cards, function at a maximum capacity of 100mbps/gbps, and that all cable runs are the required category (CAT 5e, CAT 6, etc.) This may necessitate that all routers and switches function at specific levels (Level 3, Level 4). Other considerations will be disk revolution cycles (10,000RPM, etc); it will also decide if disk arrays will be used (IDE, SCSI, RAID 0-5, SSD, etc ), or newer technology. With performance considerations, physical or virtual load balancing, failover, clusters, and such technology have to be considered, along with their supporting hardware. Technically, some things are harder to add later, so it always makes some sense to have performing hardware at the beginning, typically processor and memory (DDRs). The location hardware considerations are not excluded: cooling systems, racks, power systems, and backups, among others.
  4. Utilization and Optimization Plan—Utilization planning involves using calculative mechanisms to determine storage consumption over time and to determine storage capacity. This is done by estimating the average size of documents. This is where the usage of across-the-board templates for content management comes into play. By using templates for all document types it is easier to determine the average size of file creation. The backup strategy of critical business data and related contents will help estimate the growth rate for facilities used for onsite and offsite storage.
  5. Assumptions and Constrains – This is one area that can ruin a hardware capacity plan. It is common in nearly every project. It is always assumed that everything will go well as planned. It is assumed that vendors will deliver on the dates they promised. It is assumed that funds will be available. It is assumed that the personnel will be available to implement. It is assumed that business processes and services will run smoothly while the implementation is ongoing. Truth be told, this is not always the case. A constraint, on the other hand, is the fact that all the assumptions are not going to hold true, and will therefore affect this plan.

Growing Concern Hardware Capacity Plan

A growing concern is a company with existing infrastructure, which need to continually review the current hardware state and reappraise it in the light of new demands on the existing capacity, and make recommendations for repair, replace, or upgrade. This is especially necessary if the previous design was not future-proofed, where that means allowing for future possibilities of anything changing, either good or bad. For the greater part, a growing concern will not embark on a complete overhaul, but on systematic replacement based on the diverse needs of old services and new services. It is important to also state that even future-proofed designs still need to be revisited.

By using historical data collected over the lifespan of the existing infrastructure, it is easy to make an estimate for future usage if trends continue. This is also where governance and archiving methodologies come to the fore. A typical outcome of this situation can be documented in a way similar to this:

Item

Current Capacity

Capacity
Requirements

Increase Needed

Capacity
Thresholds

Threshold Response Strategy
(What to be done when limit reached)

Storage

20TB

50TB

30TB

15TB

Archiving, destruction, etc

Server Racks

7

14

7

Upgrade, order

Memory (Server A, B, C, D, etc)

4GB

16GB

12GB

3GB

Upgrade

<Item>

***

***

***

***

***

<Item>

***

***

***

***

***

 

Servers in a rack Battery Bank

Analyze Demand against Current Capacity

Typically, there is no hardware plan that can span a decade without requiring a reappraisal. Somewhere along the line, the server racks have to be expanded. The servers have to be upgraded. The server room cooling system has to be upgraded, and the electrical supplies have to be revisited for efficiency. Of a certainty, something is bound to change.

Item

Disk Usage

Bandwidth Usage

Latency

Host

(Physical/Virtual)

Comment

Application 1

***

***

***

***

***

Application 2

***

***

***

***

***

<Item>

***

***

***

***

***

<Item>

***

***

***

***

***

<Item>

***

***

***

***

***

A pie chart, histogram or Gantt chart, when properly done, can give visual feasibility or a comparative view of the situational history and projected goals. The same can be applied to hardware performance metrics.

The Power Factor in Hardware Planning

It is amazing that a processor runs on 12 volts of power while 240 volts (smaller in other countries, as low as 120 volts) has to be supplied. Over the years, processors and other components have been designed to operate with energy efficiency, to power down considerably when there are few tasks to be done. Theoretically, that will hold true for the end-user’s computer. Data centers and server rooms, on the other hand, have proven to be otherwise. In fact, statistically they consume more than any home or office combined up to an estimated maximum of 80 times, the lowest being 40 times! What is the culprit? It is called high availability. It is demanded and required from the Internet to the intranet, from cloud services to in-house services. Nobody ever wants to hear that the server is down. Service failure is one thing; other functions can continue to operate. Power failure is the worst, because it grinds everything to a halt, data can be corrupted, transactional committals may bounce, and it can really get worse from there. As the years have gone by, technology has become more condensed; the volume of power consumption has been rising considerably, according to statistics from the ASHRAE (American Society of Heating, Refrigerating and Air Conditioning Engineers).

An electrical fire is a terrible phenomenon. I have very recently seen a complete 16-story building engulfed due to electrical faults from one floor in the mid-section. Imagine if you had a data center on the upper floors! That is a crisis you only wish to dream about. You don’t want to build a data center or server room to support racks running at an expected 3,000 watts per rack while you are procuring blade servers running individually at 20,000 watts. That is a potential fire hazard. The IDC (International Data Corporation) estimated that the average customer over a period of a decade upped servers about six times, and storage about 69 times. Within half of a decade, energy usage doubled. In another independent estimate, it was observed that the greatest percentage of hardware cost goes to lighting, cooling, storage tanks, fuelling (third-world and developing countries), power banks (uninterruptible power supplies, inverters) and, while your plan may dictate the use of a single floor, and that may account for, say, 15% of your office space, expect its energy consumption to be at a shocking 45%, if not worse. Finally, do not consider using water-based fire protection devices, it is just a totally bad idea. And that last piece of fire-fighting hardware called an extinguisher is not meant for putting out the fires of your servers; if they ever catch fire, it is for you to use to fight your way out of any fire standing between you and your escape from the data center.