Skip to content
+90 850 885 27 43 info@argeict.com

Cloud and infrastructure

E-commerce server infrastructure and campaign scaling

ARGE ICT builds the server, database, cache, security, and monitoring infrastructure for e-commerce sites on Magento, WooCommerce, PrestaShop, or custom software, on ARGE Cloud or in your own environment, and sizes it for campaign load.

According to the Ministry of Trade's "E-Commerce Outlook in Türkiye 2025" report, Türkiye's e-commerce volume reached TRY 4.567 trillion in 2025. In a market that size, a cart page that fails on campaign day hands that day's sales to a competitor. That is why we size the infrastructure for the busiest hour of the year.

Who this page is for

If you run your store on a hosted (SaaS) platform such as Shopify, ikas, IdeaSoft, Ticimax, or T-Soft, you do not need to rent servers. The platform provider handles hosting and scaling, and the security of the payment page is also largely the responsibility of the platform and the payment provider. Check in your provider's contract what remains with you. You are not expected to build the architecture described here. If you only need an integration server that connects the store to your ERP, warehouse, or marketplaces, ARGE ICT can build that one server.

This page is for businesses that have to manage their own infrastructure or have chosen to:

  • You run Magento (Adobe Commerce), WooCommerce, or PrestaShop on your own servers.
  • You built custom software, and the store is a separate front end (headless) with several services behind it.
  • The store's database, stock, and prices are continuously synchronized with your ERP.
  • Every November the site slows down or the cart page times out during campaign days.
  • You want to decide where customer data and order records are kept and who controls them.

What breaks on campaign days

When load rises, the narrowest point clogs first. In e-commerce that is usually the database: stock deduction, cart updates, and order records are writes and cannot be cached. Product lists and images come from cache and scale easily; an order writes to the database every time.

The failures we see most often in campaigns are exhausted database connections, a full session and cart cache, a slow search service, timeouts on requests to the payment provider, and a growing backlog of jobs that send orders to the ERP. Bots also scan stock and price pages at the same hours and consume the same capacity as real customers.

Architecture: splitting the load into layers

Traffic first reaches the load balancer and the web application firewall (WAF). The load balancer spreads requests across several application servers; adding servers is horizontal scaling. For this to work, the application must keep sessions in a shared cache. If the session stays local, a customer who lands on a different server loses the cart.

Caching takes load in two places. A page cache (Varnish in Magento) serves product and category pages without querying the database. A Redis-compatible service (Valkey or Redis) holds sessions and objects. Serving static files through a content delivery network (CDN) reduces the requests that reach your servers further; you choose the CDN provider and we configure it with you.

In the database, one primary server takes writes and one or more read replicas serve product queries and reports. A read replica splits the read load; it does not reduce the write load. If order writes are slow, the fix is query and index tuning, a correctly sized connection pool, and, where it fits, moving heavy jobs to a queue. For database operations, see database management.

A queue (RabbitMQ in Magento) takes the non-critical steps of an order out of the request: the confirmation email, the ERP transfer, stock synchronization, and the shipping notice. The customer sees the order saved, and the cart page does not wait even if the ERP transfer is a few seconds late. Adobe Commerce requires OpenSearch for search, so the search service is sized as its own layer. We choose components from the platform's official system requirements and confirm versions in the vendor's current documentation before installation.

What autoscaling can and cannot do

When you run the application on Kubernetes, the Horizontal Pod Autoscaler (HPA) raises or lowers the replica count based on a metric such as average CPU use. According to the Kubernetes documentation, HPA checks every 15 seconds by default, calculates CPU as a percentage of the pod's resource request, and waits a default 5-minute stabilization window before scaling down. A pod with no resource request cannot be scaled on CPU.

HPA makes its decision within seconds; however, opening a new pod and getting it traffic, adding a new node if needed, and pulling the image can take minutes. If a campaign opens on the hour, the load arrives before the autoscaler catches up. So we reserve capacity for campaign days in advance and use autoscaling on top of it as a margin for unexpected growth. Where containers are not used, the server count is raised by hand or by a scheduled script according to the campaign calendar.

Autoscaling does not scale the database. If the application layer grows while the database stays the same, the additional application servers open more database connections and the database saturates sooner. For the scope of our Kubernetes service, see Kubernetes and containers.

Capacity plan and load test before a campaign

We do not give an unsourced "how many times more" multiplier for campaign traffic; there is no reliable, current general figure. The right measure is your own past campaigns. The plan works backward from the campaign date and has these steps:

  1. We extract the peak of previous campaigns and its ratio to a normal day from analytics and server logs: orders per minute, concurrent sessions, peak hour.
  2. We add the growth you expect for this campaign and set the target load together.
  3. We run the cart, payment, and search flows against a production-like environment with a load test. The test pushes the flows that write to the database as well as the home page.
  4. We fix the bottlenecks it reveals: queries, indexes, connection pool, cache settings, server count.
  5. We set a change freeze before the campaign; during it no new release or configuration change goes in.
  6. For campaign day we record in a written run plan who watches which alert and who does what when something fails. Working hours and response scope are set in the contract.

Bots, DDoS, and WAF

Not all campaign-day traffic is customers. According to one security vendor's 2025 report, more than half of web traffic in 2024 (51%) was automated. The figure comes from a single vendor and does not show your site's share, but it explains why scraping, card testing, and cart stuffing need a plan. Bots that hold stock keep real customers from reaching the product.

We build this layer in two parts. The WAF filters malicious application-level requests, and rate limits stop sources that send too many requests. For volumetric attacks, DDoS protection applies within the scope described on its page. No control stops every attack; we test the rules against real traffic before the campaign to keep false alarms low.

Payment page and PCI DSS

Card data falls under PCI DSS, the card industry's data security standard. Many merchants never take the card number on their own servers: they send the customer to the payment provider's page, or embed the provider's payment form in an iframe. Either choice narrows the number of systems in PCI scope considerably. A setup that processes card data on your own servers needs far broader assessment.

Redirecting or using an iframe does not end your responsibility. The integrity of the pages that lead to the payment page, control of the scripts on them, patching, and the WAF remain on your side. According to a 2025 clarification from the PCI Security Standards Council (PCI SSC), merchants that use an iframe are also expected to confirm that their site is not susceptible to script attacks. Vendors read the exact scope of that criterion differently, so confirm your scope with your payment provider or a Qualified Security Assessor (QSA). ARGE ICT does not provide PCI DSS compliance or certification services; we separate the infrastructure, segment the network, and keep logs in a way that supports that scope.

Turkish regulations and record keeping

Law No. 6563 on the Regulation of Electronic Commerce makes service providers responsible for storing and securing the personal data they obtain through transactions. On the personal data side, Law No. 6698 on the Protection of Personal Data (KVKK) also applies; for technical measures, see our KVKK technical measures checklist.

The record-keeping duty sits in two separate regulations, and they do not cover the same records. The Regulation on Electronic Commerce Intermediary Service Providers and Electronic Commerce Service Providers requires those providers to keep transaction records and to submit them to the Ministry on request. The Distance Contracts Regulation governs the seller's records of pre-contractual information, withdrawal, and delivery. The retention periods and the parties bound differ between the two texts. Your legal counsel should therefore decide how long each record is kept, and you should check the current texts with them. We do not give legal opinions.

On the infrastructure side, our job is to make storage match the period you set: records kept with redundancy, access logged, and backups held in the isolated backup zone. You tell us the retention period and deletion rule in writing.

Monitoring and alerting

Monitoring that checks only whether the server is up is not enough for an e-commerce site. The payment success rate can fall while the server is up. We watch two kinds of measures.

  • Business measures: orders per minute, payment success rate, add-to-cart errors, zero-result search rate.
  • Technical measures: 95th percentile response time (p95), database connection count, queue length, cache hit rate, disk and memory.

We set alert thresholds from the results of the pre-campaign load test. Who receives which alert and the monitoring hours are set in the contract. For details, see infrastructure and application monitoring.

Automation: infrastructure as code and release pipeline

Configuring servers and networks by hand slows down building a second environment on campaign day. When we define infrastructure as code (IaC), we can build the same environment again with the same settings, and the pre-campaign test environment comes up with the same structure as production. Application releases go through a continuous integration and delivery (CI/CD) pipeline, and a rollback step is ready for a faulty release. Scaling rules are also written as code, so it is documented at which threshold servers are added, and how many. We carry out this work under our DevOps and CI/CD pipeline setup.

On ARGE Cloud or in your own environment

We can build the e-commerce infrastructure on ARGE Cloud, our own infrastructure in a data center in Istanbul, in a VPC dedicated to you. Administrator access is limited through ARGE ZTNA or SSL VPN. We can also build the same structure in your own server room or data center; in a hybrid setup, the web and cache layers can run in the cloud while the ERP connection and the warehouse system stay on site.

Offices and warehouses connect to this setup as well. Warehouse terminals and ERP users in the office see orders and stock from the same system. The head office and branches can connect to each other and to ARGE Cloud over your carrier's MPLS line, a private line that does not go through the internet. Without MPLS, we build encrypted tunnels over your existing internet lines with SD-WAN-capable routers and firewalls. We describe the options in how head office, branches, and the cloud connect. Your carrier provides the line; we design, build, and monitor the network on top of it.

Which model fits you

Comparison of three models
Hosted SaaS platformYour own serversManaged infrastructure on ARGE Cloud
Who runs the infrastructureThe platform providerYour IT teamARGE ICT, within the scope written in the contract
Software freedomAs far as the platform allowsFull; code and data are yoursFull; code and data are yours
Campaign scalingDepends on the provider's planYou buy and plan the capacityCapacity plan and load test done together
Payment security scopeLargely the provider and the payment providerNarrowed by redirecting; everything that remains is yoursNarrowed by redirecting; network separation and logs are ours, page content and scripts are yours
When it fitsStandard store, small IT teamExperienced in-house team, special requirementsMagento, WooCommerce, or custom software; limited in-house team

Out of scope and limits

We do not sell store software licenses; the license for Adobe Commerce and paid extensions is your own contract, and updating open-source software also stays with you. We do not develop themes, plugins, or custom modules. The contract with the payment provider is between you and the provider, and we are not a party to it. PCI DSS certification and legal advice are not part of our scope either. We do not guarantee that the site will carry a particular campaign load; the capacity plan rests on measurement, and you see the result in the load test.

For the general IT needs of retail and e-commerce businesses, see our retail page.

Frequently asked questions

I use Shopify, ikas, or IdeaSoft. Do I need a separate server? No. On these platforms the provider handles hosting and scaling. Infrastructure may be needed only if you run your own integration server that connects the store to your ERP, warehouse, or marketplaces.

If I have autoscaling, do I still need to prepare before a campaign? Yes. With autoscaling, new capacity can take minutes to start serving traffic, and the database's write capacity does not grow. We reserve capacity before the campaign and use autoscaling as a margin for unexpected growth.

How much capacity do I need for a campaign? We do not give a general multiplier. We extract the peak and its ratio to a normal day from your previous campaigns, add the growth you expect, and confirm the result with a load test.

I use the payment page in an iframe. Am I in PCI scope? Your scope is considerably narrower than for a setup that processes card data on its own servers, but it is not zero. The integrity of the pages that lead to the payment page stays with you. Confirm the exact scope with your payment provider or a Qualified Security Assessor.

Why is the database the most common bottleneck, and what does a read replica solve? Orders and stock changes write to the database and cannot be cached. A read replica takes product queries and reports off the primary server; it does not reduce the write load. For writes, we tune queries, indexes, and the connection pool.

How long should I keep order and pre-contractual information records? The period depends on the type of record and which regulation applies to you; the electronic commerce regulation and the Distance Contracts Regulation cover different records. Check the current texts with your legal counsel. We build the redundant storage that matches the period you set.

How do you tell bot traffic from real customers? With WAF rules, rate limits, and request behavior. We test the rules against real traffic before the campaign so that real customers are less likely to be blocked. No rule set catches every bot.

What if moving from my current hosting falls in the campaign period? We do not recommend a migration during the change freeze before a campaign. If the date is close, we move the migration to after the campaign, run the campaign on the current environment, and add capacity only to remove the bottleneck.

Talk to us about e-commerce infrastructure

Share your store software and campaign calendar, and we will reply within 1 business day.

Request an infrastructure call