The Overloaded Allocator: A Confession About Wealth, Incentives, and Economic Deadlock

What happens when a system rewards the accumulation of resources long after the marginal benefit of accumulation has fallen below the cost imposed on everyone else?

There is a mistake that an economist ought to be able to recognize, even when the mistake is profitable.

It is the assumption that because a person has acquired resources, giving that person still more resources must be a good use of them.

The assumption is tempting. Wealth can be evidence of productive achievement. Capital finances machinery, research, infrastructure, and businesses that employ people. The prospect of a return encourages investment, and the possibility of failure disciplines bad decisions. A society that destroys these incentives in pursuit of an abstract equality can impoverish everyone.

These are not trivial considerations. They are among the strongest arguments for a market economy.

But they do not establish that every additional dollar allocated to someone who already has a billion dollars is necessarily more productive than a dollar made available to someone who has almost nothing.

The distinction is important. And it becomes difficult to ignore when we stop discussing money and start discussing computers.

The machine with too little memory

Imagine a computer with 24 gigabytes of physical memory. It runs a database, an administrative interface, a monitoring service, and a thousand smaller processes that perform useful work.

The database administrator has configured the database with a 48-gigabyte InnoDB buffer pool.

There is nothing inherently improper about configuring a buffer pool larger than physical RAM. Depending on the workload and actual memory residency, the database might perform acceptably. But suppose the database actively uses most of its allocation, the other applications need memory too, and the combined working sets exceed what the machine can sustain.

The operating system begins reclaiming memory and moving pages to swap. The administrative interface becomes sluggish. Saving a configuration change takes several seconds. Monitoring falls behind. The machine spends more and more time moving data between memory and storage instead of doing useful work.

Eventually, the system may thrash, suffer allocation failures, or terminate processes.

The database was given an enormous allocation because its operators expected a large cache to improve performance. Yet beyond some point, the additional allocation can reduce the performance of the entire machine.

The important question is not whether the database deserves memory. It plainly needs memory to function.

The question is whether the marginal benefit of giving it more memory exceeds the cost imposed on the rest of the system.

That is an economic question as much as an engineering question.

When success becomes its own justification

Now substitute wealth for memory.

A person with substantial wealth can invest in productive businesses, finance innovations, absorb risk, and support ventures that would otherwise never be attempted. The prospect of accumulating wealth can encourage saving, entrepreneurship, and long-term investment.

But wealth also gives its owner the capacity to acquire more wealth.

Assets generate returns. Returns finance additional assets. Ownership provides collateral for borrowing. Inherited wealth can give the next generation access to opportunities that others must finance from current earnings. Wealth can also buy access to professional advice, political advocacy, and the ability to withstand setbacks.

None of these mechanisms requires wrongdoing. Some are indispensable features of a functioning economy.

Yet they create a feedback loop: resources help determine access to the opportunities through which additional resources are acquired.

The result is that today's distribution of wealth influences tomorrow's distribution of wealth.

At that point, the fact that someone possesses a great deal of capital tells us something about their financial position. It does not, by itself, tell us how productively the next dollar would be used, whether the existing allocation is socially beneficial, or whether the rules that produced it should remain unchanged.

A balance sheet is not a complete measure of social value.

The market signal that never arrives

Consider a hypothetical investor with a trillion dollars in wealth.

The investor is not obliged to spend money merely because other people need goods and services. The investor is entitled to consider risk, expected return, liquidity, and the opportunity cost of each investment.

Suppose, however, that the investor has become so selective that no available project meets the required threshold. The money remains invested in existing assets or liquid financial instruments while the investor waits for a sufficiently attractive opportunity.

That decision might be entirely rational from the investor's individual perspective.

But now consider a small business that could employ twenty people if it obtained financing. Consider a household that needs a modest loan to repair its home. Consider a useful local service whose prospective customers cannot pay enough to make it attractive to conventional investors.

There may be genuine economic value in these activities. Yet their existence does not guarantee that the investor will finance them.

Markets respond to effective demand: not simply to what people need, but to what they can pay for under prevailing conditions. A need that lacks purchasing power may produce a weak commercial signal even when satisfying it would improve people's lives.

This does not mean that every unfunded project deserves investment. Many projects are unproductive, and capital wasted on them cannot be used elsewhere. Nor does it mean that money sitting in a bank account is necessarily idle; financial institutions can channel deposits into loans and investments.

The narrower point is more defensible.

The ability to pay and the capacity to create value are related, but they are not identical.

A market can be extraordinarily effective at responding to expressed purchasing power without guaranteeing that every important human need generates an adequate signal.

And an investor can make a perfectly rational decision to wait while opportunities that would benefit others remain unfunded.

The thousand one-megabyte processes

Return to the computer.

Imagine a thousand small processes, each with a modest memory requirement. Individually, each seems insignificant beside a database that could use another gigabyte. Collectively, however, those processes perform authentication, record transactions, maintain services, support users, and monitor the machine.

Some of their pages may already be swapped out, so reclaiming those pages will not magically produce a gigabyte of additional physical memory. But if the operating system kills the processes, imposes damaging limits, or deprives them of the resources required to make progress, their useful work can disappear.

The database's demand is easy to see. Its performance can be measured. Its administrators can produce charts showing the benefits of a larger cache.

The dispersed costs imposed on a thousand small processes may be harder to measure. Each process experiences a little more latency, a little less availability, or a little less capacity to complete its work. No individual failure may appear important enough to change the allocation policy.

Yet the aggregate effect can be substantial.

This is a familiar problem in economic reasoning. Concentrated benefits can be easy to identify and defend, while dispersed costs are individually small and politically difficult to organize around.

A subsidy, tax preference, regulatory exception, or publicly financed benefit may be defended by its beneficiaries as essential to investment or competitiveness. Its costs may be distributed among millions of taxpayers or consumers. The fact that each person's share is small does not establish that the total cost is small, or that the arrangement is the most productive use of public resources.

The correct question is not whether the beneficiary can explain why the benefit is useful to them.

It is whether the benefit produces enough value to justify its full cost, including costs borne by people who have little influence over the decision.

The allocator is not neutral merely because its rules are consistent

Suppose we configure the computer so that any process with a large existing allocation receives priority for future allocations.

The rule is clear. It can be applied consistently. No administrator needs to make a fresh judgment about every process each time memory is requested.

But the rule creates a peculiar result: the processes with the most memory are best positioned to obtain still more, while the smallest processes become increasingly vulnerable.

The allocator has turned an existing advantage into a claim on future resources.

Now imagine that the privileged process can influence the rules governing its own allocation. It can persuade the administrator that its workload is uniquely important, that its growth must not be constrained, and that the smaller processes should be optimized to accommodate it.

The system may continue operating for a while. The database may even achieve impressive performance benchmarks under favorable conditions.

But if the total demand becomes unsustainable, the allocator has a problem that no amount of local optimization can solve. It cannot manufacture physical memory. It can only determine who gets access to it and who bears the consequences of scarcity.

The same distinction applies to institutions.

Rules can be formally consistent while producing cumulative advantages for those who already possess wealth. The people who benefit may have legitimate reasons to defend the arrangements. They may also have more resources with which to make their case, finance research, hire specialists, and influence future rules.

This does not establish that every policy benefiting wealthy people is improper. Investment incentives can create widespread benefits. Property rights can support long-term planning. Stable institutions can encourage productive risk-taking.

But the defense of these arrangements should rest on evidence of their effects, not on the presumption that whatever benefits existing owners must be beneficial to society.

A rule is not justified merely because the people who prosper under it can afford to explain why it should remain in place.

The mistake of confusing accumulation with productivity

There is a legitimate economic argument for rewarding the creation of value. There is a separate argument for allowing people to retain wealth they have accumulated. And there is yet another argument for permitting that wealth to be transferred to their descendants.

These arguments overlap, but they are not interchangeable.

An entrepreneur may create a business that produces goods people value. An investor may finance a technology that improves productivity. A family may save over many years to provide security for its children.

But an inheritance can also confer an advantage on someone who did not create the original wealth. An asset can rise in price because of scarcity rather than because its owner has improved it. A tax preference can increase returns without increasing the underlying productivity of the investment. Political influence can protect an incumbent from competition.

The relevant question is not whether wealth is good or bad. It is which mechanisms create productive value, which merely transfer claims over existing value, and which allow the benefits of earlier accumulation to determine access to future opportunities.

Economists have long recognized the distinction between productive activity and rent-seeking: obtaining economic benefits through control, privilege, or political arrangements rather than by creating equivalent new value. Rent-seeking can consume resources that would otherwise support production and investment.

A system that rewards both activities indiscriminately risks confusing the ability to secure a larger share of the economy with the ability to make the economy larger.

That is a mistake even if one has no objection to unequal outcomes.

Why taking from the smallest is not automatically efficient

If the database needs more memory, an administrator should examine its working set, identify unnecessary allocations, reduce waste, and determine whether the hardware should be expanded. The administrator should not automatically begin by terminating the smallest processes.

The small processes may be essential. Their combined requirements may be modest. Their work may prevent failures elsewhere. The apparent savings may be outweighed by the cost of losing their functions.

Equally, an administrator should not assume that the database must receive every resource it requests merely because it is important.

A sensible system uses budgets, priorities, isolation, monitoring, and explicit limits. It distinguishes a workload's genuine requirements from its maximum possible appetite. It measures the consequences of allocation decisions across the machine rather than optimizing a single process in isolation.

Economic policy faces a related challenge.

Some public expenditures are wasteful. Some benefits discourage productive activity. Some regulations protect incumbents. Some redistribution policies create unintended incentives. Those problems deserve scrutiny.

But it does not follow that reducing assistance to people with the least resources is automatically the most efficient response to fiscal pressure. Nor does a subsidy to an established business become efficient simply because the business employs many people or has a compelling argument for receiving it.

The relevant comparison is between alternatives and their consequences.

Would a change increase total productive capacity? Would it reduce a bottleneck? Would it improve access to work, education, housing, or credit? Would it create new costs elsewhere? Who would bear the risks if the policy failed?

These are empirical questions. They should be asked of small beneficiaries and large ones alike.

A confession about the objective function

There is a temptation to believe that a sufficiently well-designed system can solve these problems by optimizing a single measure.

A computer administrator might maximize database throughput. A company might maximize shareholder returns. A government might seek to maximize economic growth. An investor might maximize risk-adjusted returns.

Each objective can be useful. None, by itself, captures every condition required for a healthy system.

The database with the highest possible cache allocation may perform poorly when the rest of the machine runs out of working memory. A company can increase short-term returns while damaging its long-term productive capacity. An economy can expand its aggregate output while leaving some people unable to obtain necessities or participate meaningfully in its opportunities.

This is not an argument against measurement. It is an argument for measuring the right things, recognizing trade-offs, and making those trade-offs visible.

A good administrator does not ask only how much memory the database can consume. The administrator asks how much memory it should consume given the needs of the whole machine.

A good economic analysis should similarly ask not only how much wealth can be accumulated, but what the rules of accumulation do to investment, competition, opportunity, resilience, and the productive capacity of the wider society.

And a good political system must confront an additional problem: those who benefit from an allocation policy may have strong incentives to resist changing it, even when evidence suggests that the policy is producing diminishing returns or imposing growing costs elsewhere.

That is why transparent accounting, independent evaluation, competition, and the ability to challenge established rules matter. They help distinguish a policy that creates broad value from one that persists because its beneficiaries have the strongest influence over its continuation.

The purpose of wealth is not simply to become more wealth

Capital is valuable because it can support production, innovation, security, and future consumption. Saving can finance investment. Investment can increase the amount a society is capable of producing. Accumulation can therefore serve a purpose beyond accumulation itself.

But the relationship is not automatic.

At some point, the relevant question becomes whether an additional allocation is being directed toward a productive use or whether the system is merely reinforcing the position of someone who already has considerable resources.

That question should not be answered by assuming that wealth is undeserved. It should not be answered by assuming that every redistribution improves outcomes. It should be answered by examining incentives, opportunity costs, productive returns, and the consequences for the system as a whole.

The market's ability to transmit information through prices is one of its great strengths. But prices reflect purchasing power as well as preferences. The return to an investor is not a comprehensive measure of the social value of an activity, and the absence of a profitable market signal does not prove that a human need is unimportant.

The challenge is to preserve the mechanisms that encourage productive investment while ensuring that the rules governing them remain open to examination.

We should reward the creation of value without assuming that every existing advantage represents its creation. We should protect the incentives to invest without treating every demand for additional resources as an entitlement. And we should evaluate public assistance, private privilege, and inherited advantage by consistent standards.

The objective is not to make every process identical. It is to keep the machine capable of doing useful work.

A system that allocates resources well should be able to explain why an allocation improves the whole, identify who bears its costs, and revise the decision when the evidence changes.

Otherwise, we risk building an allocator that becomes increasingly successful at concentrating resources and increasingly unsuccessful at sustaining the system that gives those resources their value.

The most important question is not whether the wealthiest participant can use another dollar.

It is whether giving that dollar to the wealthiest participant is the best available use of it—and whether the rules allow anyone to ask that question.

Efficiency is not the art of giving the largest process everything it can consume. It is the art of allocating scarce resources so that the whole system can continue to produce useful results.