Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article3 min read

Cloudflare Freed 100 TB of Memory: When a Server Needs Optimization, Not a New Plan

A recent Cloudflare analysis reveals how a single data structure can bloat memory usage. I examine the metrics site owners need before upgrading server capacity.

Memory modules and neatly stacked metal blocks on a dark surface
In this article

On September 18, 2026, Cloudflare announced it had reduced its operational memory consumption by more than 100 TB across its infrastructure. The culprit was a data structure in its Pingora-based load balancing service. Fixing the algorithm produced an effect that, across thousands of servers, translated into a massive figure.

For a small online store, the scale is different, but the question remains the same: why is memory insufficient? I would start any conversation about a new plan by answering that. Additional gigabytes are useful when actual workload increases. However, if an application hoards data uncontrollably, raising the limit only delays the next outage.

Used memory does not necessarily indicate a problem

A graph showing high consumption explains little on its own. Part of the memory may be occupied by useful cache. It is more concerning when response times rise simultaneously, the system starts heavily using the disk instead of RAM, or processes are terminated due to resource shortages.

Ask the administrator to show resource consumption by service: database, site handlers, search, and background jobs. You need data from both normal operation and peak traffic periods. A snapshot taken on a quiet Sunday evening does not answer what happens during a promotional campaign.

Three scenarios with different solutions

First scenario: memory runs out only when interacting with the accounting system. In this case, examine the product package size, upload concurrency, and scheduling. If imports, backups, and catalog recalculation start simultaneously, the issue may lie in how the tasks are organized.

Second scenario: memory usage gradually increases even with steady traffic, then drops sharply after a restart. This is a sign to look for retained objects, unfinished tasks, and unbounded caches. Regular restarts can temporarily sustain operations, but they do not explain the root cause.

Third scenario: each process behaves predictably, but the number of buyers has grown and all resources are now fully utilized for productive work. Here, increasing server capacity or separating services may be justified. It is unwise to save on server costs at the expense of lost orders.

What should be included in an optimization task

A good task includes baseline measurements, the problematic scenario, and acceptance criteria. For example: when the catalog loads and orders are placed simultaneously, the site maintains consistent response times, and handlers do not terminate abnormally. Merely reducing memory usage does not prove an improvement.

After making changes, compare identical conditions: visitor count, catalog size, and enabled integrations. Otherwise, it is easy to attribute results to optimization when they actually occurred simply because a sale ended.

When upgrading the plan is immediately necessary

If the site is already losing orders, temporarily increasing resources may be the fastest way to restore operations. It is important to document which issue this resolves and when the team will return to diagnostics. For long-term changes, a test copy and the ability to roll back are essential.

The Cloudflare figure is the result of specific engineering work, not a promise of similar savings for any VPS. A more practical takeaway for a site owner is this: before committing to permanent cost increases, determine what the additional gigabytes are actually buying—business growth or a delay in fixing a bug.

Discussion0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.