Every stream, checkout, and click passes through the global content delivery and edge cloud networks that sit between your application and your users. When someone hits play on a video, completes a purchase, or loads a homepage, that request is routed through these networks first.
When the edge is fast, users stay engaged and transactions complete quickly. When it’s not, the impact is immediate: slow pages, failed checkouts, and frustrated customers.
Keeping those experiences fast depends on how quickly teams can see what’s happening at the edge while traffic is still flowing. Too often, that visibility comes too late. Data is delayed or locked behind batch workflows, and by the time it’s usable, the incident is already over.
That gap is why teams need hot data at the edge. This means having access to fresh logs and performance signals from edge providers in a form that’s immediately searchable and usable, so decisions can be made in minutes instead of waiting on batch jobs or ticket queues.
Edward Tsinovoi has spent his career building systems at this boundary. As co-founder and CEO of IO River, and formerly a leader on Akamai’s core platform team, he’s seen that the problem isn’t a lack of data. It’s what happens when data gets spread across providers, tools, and time.
This is his guide for keeping edge data hot where it matters most.
1. Align on the Performance Question That Actually Matters
“Data today is the basis to make decisions, whether it’s automatic decisions taken by the platform or by the team members,” says Tsinovoi.
Most teams agree with that in principle. Where things break down is in how those decisions are framed. Instead of anchoring on the question that should guide how the edge is operating, teams often default to vendor dashboards and surface-level metrics.
Tsinovoi insists the focus should be much narrower: Which edge is actually performing best for this customer right now?
Answering that question means correlating at least three classes of data in something close to real-time:
- Logs from each edge provider
- Synthetic testing data from in-house and third-party tools
- Real user monitoring (RUM) data from live traffic
“The question of how well the edge is performing is a very complicated question,” Tsinovoi explains. “There are many data points from many different sources that try to measure the performance that the customer is getting.”
When those sources are poorly correlated or delayed, the impact shows up quickly. Users see buffering video, slow pages, or broken sessions. Ecommerce teams see abandoned carts. Media companies watch edge costs eat into already thin margins. It’s brand, revenue, and margin all tied directly to performance decisions.
Best Practices
Make the question of how the edge is performing your north star and design the data model around it.
Hot data at the edge means:
- Logs, synthetic tests, and RUM can be pulled together in seconds.
- Teams can pivot between those views during an incident without waiting on exports.
- “Best performing for this customer right now” becomes a live view, not a weekly report.
Hydrolix fits in as the engine that keeps those log and event streams hot, so routing logic and teams aren’t making decisions on stale insight.
Questions to Ask Your Team
- Can you answer “which edge is best right now” with data, not gut feel?
- How many systems do you need to touch to build that picture today?
- Which of those data sources are truly hot versus delayed or sampled?
2. Define Hot Data as What Is Actually Available
Tsinovoi’s definition of hot and cold data is simple: “Hot is available, it is here. Just click and you get the data. Cold means theoretically we have something, but we should invest significant effort to bring back the specific information that you need.”
Hot data is not a storage tier. It’s data that is ready the moment you need it. Most teams equate “stored somewhere” with “usable.” In practice, cold data often means:
- You need to request a restore or spin up a new cluster.
- You need to wait for a job to finish before you can refine your question.
- By the time you can query it, the incident is over and the damage is done.
Hot data, in contrast, is log and event data that ready for querying with sub-second response times, even at terabyte-per-day scale. With Hydrolix, you can:
- Slice it by any dimension, including attributes you did not know you would care about.
- Pivot on a new question in the middle of an incident.
- Drill from a chart to raw records without waiting for a batch job.
So why do teams fall back to using cold data? Tsinovoi sees the same two pressures everywhere:
- Cost pressure: “There is much more potential cold data than hot data,” he notes. Teams default to cheaper tiers or aggressive rollups because holding raw, fully indexed data looks expensive on paper.
- Speed tradeoffs: The more you dump into your hot tier without a strategy, the slower it becomes. “If you keep a small amount of hot data, it is faster to find the answer there,” he says. When the hot tier includes everything, but the platform isn’t designed to keep all your data hot, queries bog down. As a result, teams shrink the hot window.
The Hot Data Approach
Redefine “we have the data” to mean “we can get it in seconds, without extra work.” Then:
- Use a platform designed to keep large log volumes hot and compressed.
- Treat cold storage as archival, not a source you can quickly restore.
- Make “click and get the data” your bar for anything tied to user experience or revenue.
Questions to Ask Your Team
- When was the last time you said “we have the data” but could not get it during an incident?
- How many of your current “hot” datasets actually take minutes or hours to query?
- Do your edge teams trust that the data they see reflects what is happening right now?
3. Stop Aggregating Away Your Historical Edge
If Tsinovoi could fix one thing with a magic wand, it would be the ability to “search everything in seconds” without losing detail over time. What he wants most is to find historical twins for whatever pattern is happening right now. But that is nearly impossible in most environments.
“We’re storing a huge amount of real-time data for a short period of time,” Tsinovoi explains. “It’s impossible to store it and keep it for years in the same granularity. As a result, all the industry is aggregating the data. When you aggregate [while throwing away the underlying data], you are losing the ability to look at the data in lower granularity and find similar behaviors that happened in the past.”
To control cost and performance, teams will often:
- Keep raw logs for a short time window.
- Roll everything else into minute- or hour-level aggregates.
- Lose the exact shapes and sequences that defined past incidents.
Then a familiar story plays out:
- An edge node in one region starts showing intermittent latency spikes.
- You see the same story in user complaints and synthetic checks.
- You suspect this has happened before, but the raw data is gone and the aggregates are too smooth to prove it.
The Hot Data Approach
Treat long term, granular history as a competitive edge, not a luxury. Tsinovoi’s target state is simple:
“When something happens now, I want to identify similar use cases in the past,” Tsinovoi says. “To do it, I should be able to retrieve the data in the same granularity that I have now.”
To make that real, you need an architecture that:
- Keeps long retention windows hot instead of pushing them to slow tiers.
- Uses compression and columnar storage so raw detail stays affordable.
- Lets you run the same investigative queries against last year as you run against last hour.
Questions to Ask Your Team
- How far back can you go while keeping full log granularity?
- Can you run the same query over 30 days that you run over 30 minutes?
- When was the last time you wished you could “see this exact pattern from last year” and could not?
4. Measure Success by Speed to New Answers
When Edward describes what a successful hot data system looks like, he doesn’t start with cost per gigabyte. He starts with how fast you can move from a new question to an answer.
“Beside cost, it will be how fast I can retrieve the data in different slices without me defining the exact query ahead of time,” Tsinovoi says. “I would like to slice and dice and filter and aggregate all this data with different conditions and still be able to get the results instantly. This is the magic I would be glad to get, in addition to cost efficiency.”
Teams often design their data stack around fixed questions using structures like:
- Predefined dashboards.
- Pre-modeled cubes.
- Queries tuned for known reports.
The moment someone asks a new question in the middle of an incident, everything slows down. A ticket gets filed. A batch job gets queued. The incident window closes before the answer lands.
The Hot Data Approach
Measure your edge data stack by its “speed to new questions”:
- How quickly can an engineer or GTM operator try a new slice?
- How many iterations can they run during a single incident?
- How often do they have to give up because the query is “too big”?
Platforms like Hydrolix are built for that exploratory mode. For log heavy workloads, hot data means:
- Sub-second queries across both real-time and historical logs.
- Freedom to explore new dimensions mid-investigation.
- Long retention that keeps those explorations economically possible.
For performance-sensitive businesses in streaming, media, and ecommerce, that speed becomes a GTM advantage. You can respond to incidents faster, validate product and edge strategy changes with real behavior, and tell a more credible reliability story to the market.
Questions to Ask Your Team
- When something breaks, how many new questions can you realistically ask before people give up?
- Do your teams feel free to explore, or are they afraid of expensive queries?
- If you added a new edge provider tomorrow, how hard would it be to compare performance across all of them?
5. Treat Hot Data as a Survival Requirement in the Age of AI
Everyone in performance engineering knows that faster experiences tend to drive more revenue. Tsinovoi is careful not to overpromise on precision.
“Performance world is a challenging area,” Tsinovoi says. “Performance impacts the revenue of businesses. The challenge is that there are a lot of different factors which might impact performance, and it is very difficult to correlate a single change to revenue.”
The Hot Data Approach
GTM leaders need to be realistic. You are unlikely to get a perfect A/B test that shows keeping twelve months of edge logs hot increased bookings by a precise percentage. What you can do is track the leading indicators that sit between your data layer and outcomes:
- Time to detect and resolve edge incidents
- Number of customer visible incidents per quarter
- Query latency for the dashboards teams rely on in the moment
- Frequency of support escalations tied to “we could not see what was happening”
Questions to Ask Your Team
- Are oyur AI and routing systems learning from full-fidelity edge data, or just high-level aggregates?
- When an AI-driven decision looks wrong, how fast can you trace it back to the exact logs and signals behind it?
- Do security, SRE, and GTM all share one hot view of edge performance, or are they working from different partial datasets?
Final Takeaway
Tsinovoi’s final point zooms out from day-to-day operations to the requirements data systems need to meet as AI becomes more deeply embedded in decision-making.
“In a world where AI is changing the way that people are consuming data, technological leaders should think about reliability, performance, security of the data, and accessibility of the data,” Tsinovoi says. “These will be critical aspects for any technological solution to survive the AI wave.”
Here are some takeaways worth remembering:
- Hot data is defined by how quickly you can use it, not where it is stored.
- Aggregation can help control cost, but it should not erase the patterns you need to recognize.
- The ability to ask new questions quickly matters more than perfectly modeled dashboards.
- AI and automation are only as effective as the data they can learn from.
That’s the problem space Hydrolix is built for. By keeping large volumes of log and event data compressed, searchable, and available over long retention periods, Hydrolix gives teams a way to work through incidents and edge behavior using full-fidelity data, not summaries created after the fact.
That capability is what makes fast, confident decisions possible when systems are under pressure.
Next Steps
Interested in getting real-time and historical visibility from the origin to the edge?
- Check out CDN Insights, which gives you visibility across all your CDN providers.
- Watch a short demo of CDN Insights in action.
- Or try a guided demo to learn how CDN Insights can specifically help with your use case.
