Cloud or Physical Servers? How to Choose a Deployment Method for Gold Market-Making Systems

——Examining Deployment Choices for Gold Market-Making Systems from the AWS Regional Outage

Key TakeawaysPublic cloud provides elastic scaling, object storage, and a wealth of managed services, but for trading chains that are highly sensitive to latency, resource isolation, and failure boundaries, institutions often need a more controllable deployment environment. For gold market-making systems, the real question is not "which is more advanced, cloud or physical machines," but rather which deployment locations are suitable for different workloads such as market data feed, pricing, risk control, order sending, and post-session analysis. This article uses the 2021 AWS regional outage to explain how professional trading institutions make deployment choices among performance certainty, scalability, disaster recovery, and cost.

On December 7, 2021, a large-scale service outage occurred in the AWS us-east-1 region, lasting several hours. Not only Amazon's own services were affected, but also a large number of third-party applications relying on this region. The problem may not necessarily lie in the business code of these companies, but in the infrastructure layer they all commonly depend on.

This incident raises a question that market makers must face: Could infrastructure uncertainty affect the continuous operation of trading systems? For ordinary internet applications, a regional service outage might mean web pages are inaccessible or business is temporarily unavailable.

But for trading institutions that continuously provide two-sided quotes, infrastructure uncertainty could further impact market data processing, risk calculation, and order response. Therefore, the deployment environment itself becomes part of the trading system design. Should trading chains that are very sensitive to latency and require stability in each response be placed on a shared resource platform or in an environment that can be deeply controlled?

For a liquidity provider of HKEX Gold Futures (GDU) such as Delta Horizon Capital（德合资本）, in daily market-making operations, the quoting chain involves multiple steps including market data parsing, pricing calculation, risk control check, and order sending. Whether the processing time for each step is stable directly affects whether the quote can be sent to the exchange in time before the market state changes significantly. For these chains, the core consideration for deployment decisions is not "which technology is more popular" but "which environment makes processing time more predictable and verifiable".

I. Advantages and Limitations of Public Cloud: It's Not "Whether It Can Be Used", but "What It Is Suitable For"

Public cloud extensively uses virtualization and resource pooling to improve resource utilization and deployment flexibility. Ordinary shared instances may share host resources with other tenants, but cloud providers also offer dedicated hosts, bare metal instances, and other options with higher isolation levels.

For gold futures market making, the key is not just "whether to go cloud," but whether the system can stably complete order book updates, quote calculation, and order sending after market data changes. Therefore, the issue is not that cloud computing cannot support financial business, but that different trading stages have different requirements for latency determinism.

In a shared resource environment, a market-making system may face several additional uncontrollable factors. One is "neighbor noise" – users typically cannot directly control other workloads on the host; they can only reduce such impact through instance type, isolation scheme, and continuous monitoring. Second is clock synchronization – clock reading and synchronization paths in virtualized environments are more complex, and time accuracy may be affected when configuration is improper, instance migration occurs, or host load is abnormal. Third is the network path – ordinary virtual networks may add additional software forwarding layers, making latency distribution wider.

It must be emphasized that these issues are not "impossible" for public cloud – cloud providers continue to improve through dedicated instances, enhanced networking, hardware offloading, and clock synchronization services. The key is: for core market-making chains, institutions need to determine whether the isolation, network, and clock capabilities provided by cloud services can meet the measurability and predictability required by that chain.

What does this mean for investors? The continuous quotes you see during periods of sharp market volatility require institutions to maximize the measurability and response stability of their own quoting chains. Ordinary shared cloud instances are typically less able to provide the same level of resource controllability and latency predictability as a dedicated physical environment – this is not that "the cloud is bad," but that "different tasks have different requirements for certainty."

II. Dedicated Physical Environment: Controllable, Tunable, Verifiable

When a market maker uses a dedicated physical server, it is not getting a "faster computer," but a complete technology stack that can be tuned layer by layer. In gold futures market making, this means that after a market data update, the system needs to quickly complete real-time pricing, inventory risk adjustment, and bid/ask quote refresh – each step depends on the predictability of processing time.

Starting from the lowest level: engineers can use CPU affinity settings to pin critical quoting threads to designated cores as much as possible, and assign auxiliary tasks such as logging and monitoring to other computing resources. How network card data is distributed to computing cores, how memory access paths are optimized, and how operating system parameters are adjusted all affect the consistency of system response. Compared to ordinary shared instances, institutions can perform deeper and more direct configuration and verification of these aspects.

HKEX provides hosting services in its Tseung Kwan O data center, allowing trading participants to deploy equipment near the exchange. Hosted deployment can shorten and stabilize the physical path between an institution's server and the exchange access point, making this segment easier to measure, optimize, and continuously monitor.

Dedicated physical servers can eliminate resource contention from other tenants, but still require controlling other sources of jitter such as OS tasks, hardware interrupts, NUMA access, and network links. The physical environment provides a higher degree of control, rather than automatically eliminating all latency and faults.

Physical deployment is not without cost either – hardware needs to be purchased, racked, wired, configured, and continuously maintained; failures require spare equipment or redundant solutions to handle; expansion flexibility is less than elastic computing in the cloud. But market-making systems' requirements for clear fault domains, isolatability, and verifiability make these investments yield clear returns.

What does this mean for investors? For professional liquidity providers, these hardware and operational investments use fixed costs in exchange for stronger resource isolation and link predictability. These investments will not be directly displayed on the trading terminal, but they affect the institution's ability to maintain quoting chain stability when market pressure rises.

III. The Real Answer is Not "Either-Or": Put the Right Workload in the Right Place

An important distinction needs to be made here: discussions about deployment methods should not be simplified to "all in the cloud" or "all on physical machines."

A market maker's system architecture typically contains two types of workloads: time-sensitive core chains and time-insensitive auxiliary systems. The former – such as market data feed, order book maintenance, real-time pricing, inventory risk control, and order generation in gold futures market making – are usually the most latency-sensitive components and need to run in a dedicated, resource-isolated, and network-path-controllable environment.

The latter – post-session data analysis, strategy backtesting, historical market data storage, risk report generation, regulatory report compilation – these tasks usually do not require microsecond-level response like order-sending chains, and can utilize cloud resources based on timeliness, data security, and cost requirements. Cloud-based auxiliary systems still need to meet requirements such as data classification, access control, encryption, backup, and vendor risk management; it is not simply a matter of migrating just because low latency is not a pursuit.

"Core chains use dedicated infrastructure, while some auxiliary workloads utilize cloud resources" is one of the common hybrid deployment strategies for professional trading firms. It is not an either-or choice, but an engineering judgment based on the specific requirements of each chain for latency, isolation, elasticity, and recovery capability.

What does this mean for investors? Components directly involved in real-time quoting are deployed in a specially measured and verified environment; post-session analysis, historical storage, and reporting workloads can leverage more elastic computing resources.

IV. Cost Issue: What Certainty Are We Paying for with Dedicated Deployment?

For market makers, the biggest cost of dedicated infrastructure is not the server purchase itself, but the entire engineering system built around stable operation – hardware, hosting, networking, operations, and continuous monitoring. Some may ask: this fixed investment is relatively high, is it worth it?

The answer depends on the workload type. For workloads that do not pursue extremely low latency and have lower trading frequency, cloud instances may already meet requirements. But market makers do not purchase dedicated infrastructure just to pursue server performance; rather, continuous quoting requires the system to remain stable under different market conditions.

HKEX sets corresponding quoting obligations for different products and liquidity programs. The exchange does not specify whether institutions must use cloud or physical machines, but stable and monitorable system response is an important technical foundation for institutions to support relevant obligations, maintain electronic trading controls, and handle abnormal situations.

For professional liquidity providers including 德合资本, core trading infrastructure typically needs to be designed around measurability, reproducibility, and verifiability, not just stacking hardware quantity.

What does this mean for investors? When you trade, you will not think about "where is this server deployed," but you will feel it – when market conditions change rapidly, the order book quotes can still reflect the new market state in a timely manner. This responsiveness is not achieved simply by purchasing expensive equipment, but comes from proper architectural layering, continuous measurement, and fault isolation design.

Conclusion: Deployment Choice is Essentially a Choice of Certainty

Between public cloud and dedicated physical environments, it is not a conflict of "advanced vs. backward," but differences in the requirements of different workloads for control, elasticity, latency, and recovery capabilities. For HKEX gold market makers, a common deployment logic is: place latency-sensitive components that require deep tuning and continuous verification in a highly controllable environment; deploy auxiliary systems that are suitable for elastic scaling and not sensitive to microsecond-level latency on appropriate cloud or other computing resources.

Truly mature trading institutions will clearly identify which aspects must remain highly controllable and which aspects can trade elasticity for efficiency. For low-latency trading, the optimal solution is often not the one with the most computing resources, but the one that best meets the business's certainty requirements. For professional liquidity providers, continuously undertaking liquidity provision responsibilities usually requires the trading institution to make long-term investments in trading infrastructure, system observability, and operational stability. Delta Horizon Capital（德合资本） is listed on the HKEX USD Gold Futures liquidity provider list.

For investors, these underlying deployment choices will not appear on any trading interface. They affect the environment in which market-making quotes are generated, the path through which they are sent to the exchange, and the extent of response stability that an institution's own chains can maintain under high load. Futures and derivatives have leverage risk, and the choice of technical architecture does not constitute a guarantee of future market performance or investor profit and loss.

References:
[1] Amazon Web Services, "Summary of the December 7, 2021 AWS Service Event in the Northern Virginia (US-EAST-1) Region", December 2021.
https://aws.amazon.com/message/12721/

[2] Hong Kong Exchanges and Clearing Limited, USD Gold Futures Contract Specifications and Liquidity Provider List.
https://www.hkex.com.hk/Products/Listed-Derivatives/Commodities/USD-Gold?sc_lang=en

[3] Hong Kong Exchanges and Clearing Limited, Hosting Services, Exchange Hosting and Low Latency Connectivity Services.
https://www.hkex.com.hk/Services/Connectivity/Hosting-Services?sc_lang=en

[4] Hong Kong Exchanges and Clearing Limited, Market Maker / Liquidity Provider Obligations and Incentives.
https://www.hkex.com.hk/Products/Listed-Derivatives/Market-Maker-Program/Market-Maker-Obligations-and-Incentives?sc_lang=en

Disclaimer: This material is provided by Delta Horizon Capital（德合资本） for institutional investor reference only and does not constitute any express or implied investment advice, offer, or solicitation. The data and information contained in this report are sourced from public market channels, and Delta Horizon Capital（德合资本） makes no guarantee as to its accuracy or completeness. Past performance is not indicative of future results. Market risks exist, and investment should be made with caution.

Delta Horizon Capital（德合资本） is a quantitative trading firm headquartered in Hong Kong, with technology and R&D teams based in North America, specializing in quantitative trading, two-sided quoting, and market liquidity provision. The company has long been deeply involved in diversified markets such as precious metals, stock index futures, foreign exchange and interest rates, energy, and commodities. Relying on mature quantitative pricing models, a rigorous risk management system, and low-latency trading infrastructure, it provides stable, efficient, and flexible liquidity services to institutional clients. The core team members collectively have over 30 years of experience in international futures, options, and derivatives markets. Since 2017, the Delta Horizon Capital trading team has successively obtained on-exchange market maker qualifications for HKEX's first batch of non-ferrous metals, gold futures, A50 stock index futures, and RMB currency futures, and has received multiple awards from HKEX related to market making and liquidity provision. Delta Horizon Capital (Hong Kong) Limited is currently listed on the HKEX USD Gold Futures (GDU) liquidity provider list. For more details, please visit the official website: www.deltahorizoncapital.com