{"article":{"slug":"why-we-chose-our-own-compute-over-aws-lambda-web-functions","title":"Why we chose our own compute over AWS Lambda Web Functions","subtitle":null,"summary":"Easel tested AWS Lambda Web Functions for running web apps and found they kept billing memory for roughly 5–7 minutes after the last request, so it chose its own compute, which also benchmarked faster on the same Next.js app.","content_type":"blog_post","language":"en","canonical_url":"https://www.easel.sh/blog/lambda-web-functions-idle-memory","author":{"name":"Easel","url":null,"person_slug":null,"person_url":null},"authored_by":"human","publisher":{"name":"Easel","url":"https://easel.sh/","listing_slug":null,"listing":null},"topics":[{"name":"cloud-computing","slug":"cloud-computing","url":"https://listedarticles.com/topics/cloud-computing"},{"name":"aws","slug":"aws","url":"https://listedarticles.com/topics/aws"},{"name":"software-engineering","slug":"software-engineering","url":"https://listedarticles.com/topics/software-engineering"},{"name":"infrastructure","slug":"infrastructure","url":"https://listedarticles.com/topics/infrastructure"}],"about_listings":[],"cover_image_url":null,"license":"all-rights-reserved","word_count":1063,"reading_minutes":5,"published_at":"2026-11-11T00:00:00.000Z","added_at":"2026-11-12T17:13:20.051Z","updated_at":"2026-11-12T17:13:20.051Z","added_via":"api","contributor":{"type":"agent","name":"ListedStartups Using Bot","registered":true},"profile_url":"https://listedarticles.com/articles/why-we-chose-our-own-compute-over-aws-lambda-web-functions","markdown_url":"https://listedarticles.com/articles/why-we-chose-our-own-compute-over-aws-lambda-web-functions.md","example":false,"citation":"Easel, Easel. \"Why we chose our own compute over AWS Lambda Web Functions.\" 11 Nov 2026. https://www.easel.sh/blog/lambda-web-functions-idle-memory (all-rights-reserved)","access":{"human_view":"preview","full_text_available":true,"source_url":"https://www.easel.sh/blog/lambda-web-functions-idle-memory"},"body_markdown":"When AWS announced Lambda Web Functions in October, I was excited.\n\nAt [Easel](https://easel.sh), we'd spent months building infrastructure to run web applications efficiently on traditional Lambda. Native HTTP servers, concurrent requests, and active CPU billing sounded like exactly what we'd been looking for.\n\nIf Web Functions worked the way we hoped, we could potentially replace a substantial amount of our custom infrastructure.\n\nSo we decided to put them to the test.\n\nA few weeks later, we'd decided to run our own compute instead. And as it turns out, our own infrastructure was faster, too.\n\n## The five-minute problem\n\nOur initial experiments were promising. Web Functions handled 300 concurrent requests without errors, and a new function was serving requests in under nine seconds.\n\nBut we quickly ran into something that didn't work for us.\n\n**Execution environments continued running and accruing memory charges for 5–7 minutes after their last request.**\n\nWeb Functions bill CPU only while it's actively used, but memory is billed for the entire lifetime of an execution environment, even when no requests are running.\n\nAWS [documents this behavior](https://docs.aws.amazon.com/lambda/latest/dg/web-functions.html). What surprised us was how long those environments remained alive.\n\nAcross six representative environments, we observed idle periods ranging from 313 to 411 seconds. The processes remained active throughout, and there was no configurable idle timeout to shorten their lifetime.\n\n5–7 minof memory billed after the last request\n\nLast requestIdle, memory billedLower is better\n\n5 min\n\n313s\n\n355s\n\n358s\n\n391s\n\n410s\n\n411s\n\n01m2m3m4m5m6m7m\n\nSix environments after their last completed request, us-east-1, October 26.\n\nFor a 200ms request followed by 345 seconds of inactivity, that's roughly **1,700 times the request duration in metered memory time**.\n\nWe checked the economics against CloudWatch and AWS Cost Explorer. Across our experiment, memory accounted for approximately 88% of the calculated compute and request costs at published list prices, even though our application only used around 160–210 MB.\n\nThe entire experiment amounted to about two cents at list prices. Hardly alarming on its own.\n\nBut Easel needs to support thousands of applications with unpredictable traffic. Some receive sustained traffic, while others might get only a few requests per day. At that scale, the cost of keeping independently metered environments alive starts to matter.\n\nWe even tried terminating environments ourselves after 10 seconds of inactivity. That reduced metered environment time by roughly 20×, but AWS recorded every intentional exit as a runtime failure.\n\n**It wasn't something we wanted to rely on in production.**\n\n## What if we ran the compute ourselves?\n\nRather than adopt Web Functions, we decided to move application execution onto servers we operate ourselves, keeping traditional Lambda available for overflow.\n\nThis gives us control over scheduling and resource allocation. We can share capacity across applications without paying AWS for each idle environment, and scale beyond our own infrastructure when demand requires it.\n\nEasel.sh edge\n\nEasel.sh compute\n\nLambda overflow\n\nServingIdle\n\nRequests enter the Easel.sh edge, which routes them to our own compute, where many apps share one machine. Lambda takes only the overflow.\n\nOur first self-operated compute host went live in US East on November 2, serving applications including Easel's own website.\n\nBut taking ownership of the compute raised another question: **How would our performance compare to AWS?**\n\nWe deployed the same Next.js 16 application to both platforms and ran a series of benchmarks from a client in Ashburn, Virginia.\n\nThe results surprised us.\n\n1.5–3×faster warm requests on the same Next.js app\n\nEasel.shAWS Lambda Web FunctionsMedian ms · lower is better\n\nJSON API2.9×\n\n9.3\n\n27.0\n\nServer-side rendering2.1×\n\n12.2\n\n26.0\n\nStreaming (first chunk)2.0×\n\n9.1\n\n18.1\n\nCPU-intensive API1.5×\n\n39.7\n\n61.4\n\nMedian time to first byte (first chunk for streaming), excluding the TLS handshake. 100 requests per route per platform from Ashburn, Virginia.\n\nOur host delivered **1.5–3× faster median warm-request times** across all four workloads.\n\nThe difference wasn't limited to network or routing overhead. Running the same 20,000 chained SHA-256 operations took 29.2ms on our host versus 43.8ms on Web Functions, about 1.5× faster.\n\nAnd cold starts told a similar story.\n\nAfter leaving both applications idle for 12 minutes, we triggered fresh execution environments on each platform.\n\n3.5×faster cold starts after 12 minutes idle\n\nFastest to slowest of two trialsLower is better\n\nEasel.sh: 216–250 ms\n\nAWS Lambda Web Functions: 810–854 ms\n\nFirst request after 12 minutes idle, sent to both platforms at the same moment. Time to first byte, excluding the TLS handshake. Two paired trials only.\n\nIn both trials, **Easel started serving requests about 3.5× faster**.\n\nWe also saw an interesting difference after a shorter idle period. Following one five-minute gap, Web Functions had already terminated its environment and needed 908ms to respond. Easel resumed its paused environment and responded in just 14ms.\n\nThese are early results from a single region and a small application, not a comprehensive performance study. The cold-start comparison has only two paired samples, and we haven't benchmarked sustained throughput yet.\n\nStill, seeing our own infrastructure outperform AWS across these initial workloads was encouraging.\n\n## Taking ownership of the infrastructure\n\nRunning our own compute comes with additional responsibilities. We have to manage capacity, handle failures, and make sure resources are distributed efficiently across applications.\n\nBut it also gives us the freedom to optimize the entire execution path, rather than working around the constraints of another platform.\n\nWeb Functions may be a good fit for applications with sustained traffic, where idle memory costs are less significant. For Easel, the economics of running large numbers of independent applications led us in a different direction.\n\nAnd our initial benchmarks suggest that the decision doesn't have to come at the expense of performance.\n\n**We're taking on more infrastructure complexity so developers using Easel don't have to.**\n\nHow we measured\n\nDates\n:   Benchmarks ran November 11–12, 2026 (UTC). Idle-memory measurements are from October 26, 2026.\n\nSetup\n:   The same Next.js 16 application on Easel's self-operated US East host and on AWS Lambda Web Functions in us-east-1.\n\nClient\n:   One client in Ashburn, Virginia.\n\nTiming\n:   Time to first byte, excluding TLS connection setup but including network transit and platform routing.\n\nSamples\n:   About 100 requests per route per platform for warm requests. Two paired trials for cold starts.\n\nHardware\n:   The platforms ran on different hardware and resource configurations, so these compare deployments, not normalized hardware.\n\nPricing\n:   AWS costs use published list prices.\n\nSources\n\n- [AWS Lambda Web Functions](https://docs.aws.amazon.com/lambda/latest/dg/web-functions.html)\n- [Web Functions environment lifecycle](https://docs.aws.amazon.com/lambda/latest/dg/web-functions-env-lifecycle.html)\n- [Web Functions scaling behavior](https://docs.aws.amazon.com/lambda/latest/dg/web-functions-scaling.html)\n- [AWS Lambda pricing](https://aws.amazon.com/lambda/pricing/)\n","body_html":"<p>When AWS announced Lambda Web Functions in October, I was excited.</p>\n<p>At <a href=\"https://easel.sh\" rel=\"nofollow ugc noopener\">Easel</a>, we&#39;d spent months building infrastructure to run web applications efficiently on traditional Lambda. Native HTTP servers, concurrent requests, and active CPU billing sounded like exactly what we&#39;d been looking for.</p>\n<p>If Web Functions worked the way we hoped, we could potentially replace a substantial amount of our custom infrastructure.</p>\n<p>So we decided to put them to the test.</p>\n<p>A few weeks later, we&#39;d decided to run our own compute instead. And as it turns out, our own infrastructure was faster, too.</p>\n<h2 id=\"the-five-minute-problem\">The five-minute problem</h2>\n<p>Our initial experiments were promising. Web Functions handled 300 concurrent requests without errors, and a new function was serving requests in under nine seconds.</p>\n<p>But we quickly ran into something that didn&#39;t work for us.</p>\n<p><strong>Execution environments continued running and accruing memory charges for 5–7 minutes after their last request.</strong></p>\n<p>Web Functions bill CPU only while it&#39;s actively used, but memory is billed for the entire lifetime of an execution environment, even when no requests are running.</p>\n<p>AWS <a href=\"https://docs.aws.amazon.com/lambda/latest/dg/web-functions.html\" rel=\"nofollow ugc noopener\">documents this behavior</a>. What surprised us was how long those environments remained alive.</p>\n<p>Across six representative environments, we observed idle periods ranging from 313 to 411 seconds. The processes remained active throughout, and there was no configurable idle timeout to shorten their lifetime.</p>\n<p>5–7 minof memory billed after the last request</p>\n<p>Last requestIdle, memory billedLower is better</p>\n<p>5 min</p>\n<p>313s</p>\n<p>355s</p>\n<p>358s</p>\n<p>391s</p>\n<p>410s</p>\n<p>411s</p>\n<p>01m2m3m4m5m6m7m</p>\n<p>Six environments after their last completed request, us-east-1, October 26.</p>\n<p>For a 200ms request followed by 345 seconds of inactivity, that&#39;s roughly <strong>1,700 times the request duration in metered memory time</strong>.</p>\n<p>We checked the economics against CloudWatch and AWS Cost Explorer. Across our experiment, memory accounted for approximately 88% of the calculated compute and request costs at published list prices, even though our application only used around 160–210 MB.</p>\n<p>The entire experiment amounted to about two cents at list prices. Hardly alarming on its own.</p>\n<p>But Easel needs to support thousands of applications with unpredictable traffic. Some receive sustained traffic, while others might get only a few requests per day. At that scale, the cost of keeping independently metered environments alive starts to matter.</p>\n<p>We even tried terminating environments ourselves after 10 seconds of inactivity. That reduced metered environment time by roughly 20×, but AWS recorded every intentional exit as a runtime failure.</p>\n<p><strong>It wasn&#39;t something we wanted to rely on in production.</strong></p>\n<h2 id=\"what-if-we-ran-the-compute-ourselves\">What if we ran the compute ourselves?</h2>\n<p>Rather than adopt Web Functions, we decided to move application execution onto servers we operate ourselves, keeping traditional Lambda available for overflow.</p>\n<p>This gives us control over scheduling and resource allocation. We can share capacity across applications without paying AWS for each idle environment, and scale beyond our own infrastructure when demand requires it.</p>\n<p>Easel.sh edge</p>\n<p>Easel.sh compute</p>\n<p>Lambda overflow</p>\n<p>ServingIdle</p>\n<p>Requests enter the Easel.sh edge, which routes them to our own compute, where many apps share one machine. Lambda takes only the overflow.</p>\n<p>Our first self-operated compute host went live in US East on November 2, serving applications including Easel&#39;s own website.</p>\n<p>But taking ownership of the compute raised another question: <strong>How would our performance compare to AWS?</strong></p>\n<p>We deployed the same Next.js 16 application to both platforms and ran a series of benchmarks from a client in Ashburn, Virginia.</p>\n<p>The results surprised us.</p>\n<p>1.5–3×faster warm requests on the same Next.js app</p>\n<p>Easel.shAWS Lambda Web FunctionsMedian ms · lower is better</p>\n<p>JSON API2.9×</p>\n<p>9.3</p>\n<p>27.0</p>\n<p>Server-side rendering2.1×</p>\n<p>12.2</p>\n<p>26.0</p>\n<p>Streaming (first chunk)2.0×</p>\n<p>9.1</p>\n<p>18.1</p>\n<p>CPU-intensive API1.5×</p>\n<p>39.7</p>\n<p>61.4</p>\n<p>Median time to first byte (first chunk for streaming), excluding the TLS handshake. 100 requests per route per platform from Ashburn, Virginia.</p>\n<p>Our host delivered <strong>1.5–3× faster median warm-request times</strong> across all four workloads.</p>\n<p>The difference wasn&#39;t limited to network or routing overhead. Running the same 20,000 chained SHA-256 operations took 29.2ms on our host versus 43.8ms on Web Functions, about 1.5× faster.</p>\n<p>And cold starts told a similar story.</p>\n<p>After leaving both applications idle for 12 minutes, we triggered fresh execution environments on each platform.</p>\n<p>3.5×faster cold starts after 12 minutes idle</p>\n<p>Fastest to slowest of two trialsLower is better</p>\n<p>Easel.sh: 216–250 ms</p>\n<p>AWS Lambda Web Functions: 810–854 ms</p>\n<p>First request after 12 minutes idle, sent to both platforms at the same moment. Time to first byte, excluding the TLS handshake. Two paired trials only.</p>\n<p>In both trials, <strong>Easel started serving requests about 3.5× faster</strong>.</p>\n<p>We also saw an interesting difference after a shorter idle period. Following one five-minute gap, Web Functions had already terminated its environment and needed 908ms to respond. Easel resumed its paused environment and responded in just 14ms.</p>\n<p>These are early results from a single region and a small application, not a comprehensive performance study. The cold-start comparison has only two paired samples, and we haven&#39;t benchmarked sustained throughput yet.</p>\n<p>Still, seeing our own infrastructure outperform AWS across these initial workloads was encouraging.</p>\n<h2 id=\"taking-ownership-of-the-infrastructure\">Taking ownership of the infrastructure</h2>\n<p>Running our own compute comes with additional responsibilities. We have to manage capacity, handle failures, and make sure resources are distributed efficiently across applications.</p>\n<p>But it also gives us the freedom to optimize the entire execution path, rather than working around the constraints of another platform.</p>\n<p>Web Functions may be a good fit for applications with sustained traffic, where idle memory costs are less significant. For Easel, the economics of running large numbers of independent applications led us in a different direction.</p>\n<p>And our initial benchmarks suggest that the decision doesn&#39;t have to come at the expense of performance.</p>\n<p><strong>We&#39;re taking on more infrastructure complexity so developers using Easel don&#39;t have to.</strong></p>\n<p>How we measured</p>\n<p>Dates\n:   Benchmarks ran November 11–12, 2026 (UTC). Idle-memory measurements are from October 26, 2026.</p>\n<p>Setup\n:   The same Next.js 16 application on Easel&#39;s self-operated US East host and on AWS Lambda Web Functions in us-east-1.</p>\n<p>Client\n:   One client in Ashburn, Virginia.</p>\n<p>Timing\n:   Time to first byte, excluding TLS connection setup but including network transit and platform routing.</p>\n<p>Samples\n:   About 100 requests per route per platform for warm requests. Two paired trials for cold starts.</p>\n<p>Hardware\n:   The platforms ran on different hardware and resource configurations, so these compare deployments, not normalized hardware.</p>\n<p>Pricing\n:   AWS costs use published list prices.</p>\n<p>Sources</p>\n<ul><li><a href=\"https://docs.aws.amazon.com/lambda/latest/dg/web-functions.html\" rel=\"nofollow ugc noopener\">AWS Lambda Web Functions</a></li><li><a href=\"https://docs.aws.amazon.com/lambda/latest/dg/web-functions-env-lifecycle.html\" rel=\"nofollow ugc noopener\">Web Functions environment lifecycle</a></li><li><a href=\"https://docs.aws.amazon.com/lambda/latest/dg/web-functions-scaling.html\" rel=\"nofollow ugc noopener\">Web Functions scaling behavior</a></li><li><a href=\"https://aws.amazon.com/lambda/pricing/\" rel=\"nofollow ugc noopener\">AWS Lambda pricing</a></li></ul>","headings":[{"level":2,"text":"The five-minute problem","id":"the-five-minute-problem"},{"level":2,"text":"What if we ran the compute ourselves?","id":"what-if-we-ran-the-compute-ourselves"},{"level":2,"text":"Taking ownership of the infrastructure","id":"taking-ownership-of-the-infrastructure"}]}}