diff --git a/_posts/2023-12-21-One-Bite-At-A-Time.md b/_posts/2023-12-21-One-Bite-At-A-Time.md
new file mode 100644
index 0000000..debacd4
--- /dev/null
+++ b/_posts/2023-12-21-One-Bite-At-A-Time.md
@@ -0,0 +1,43 @@
+---
+layout: post
+title:  One Bite At A Time
+header: header-eng
+---
+
+As part of our regular Gradle build scan investigation we looked at an AndroidX
+build that should have been largely a 100% remote cache hit. We found that we
+were in fact hitting the cache as expected (yay!), but the build timeline had
+a long task that hit the cache and yet was taking way longer than any other task
+(nay!).
+
+![Screenshot of Gradle timeline showing one task taking a long time](/assets/2023-12-21-timeline.png){:width="600px"}
+([timeline](https://ge.androidx.dev/s/4zzw27ril4hw6/timeline))
+
+Taking a deeper look, we noticed that this task has a fairly large 60.7MB output,
+but it is downloading it at a measly 750KBps.
+
+![Screenshot of Gradle task detail view](/assets/2023-12-21-task-details.png){:width="600px"}
+([task details](https://ge.androidx.dev/s/4zzw27ril4hw6/timeline?details=tbyyxnqlyzy3c&expanded=WyI3IiwiMSJd&page=126))
+
+Luckily it was easily reproducible locally. Trying to download the exact same
+cache entry using `gsutil` directly took <2 seconds, so we clearly had an issue.
+Taking a cursory look at a network activity on the machine showed that even
+Gradle build cache download takes little time, but the cache entry is not
+returned for over a minute after.
+
+![Screenshot of network manager showing one big burst](/assets/2023-12-21-network.png){:width="600px"}
+
+Then following profiling that I've blogged about before, it was clear that we
+were spending all of our time in `FileInputStream.read`.
+
+![Screenshot of a flamegraph for fetching the cache entyr](/assets/2023-12-21-flamegraph.png){:width="600px"}
+
+And then it downed on us, we were not doing any sort of buffering for this!!!
+[A small PR later](https://github.com/androidx/gcp-gradle-build-cache/pull/40)
+we went from [82s (750KBps)](https://ge.androidx.dev/s/q36vklz4xyytw/performance/build-cache?anchor=eyJpZCI6InJlbW90ZS1oaXQifQ&cacheDetails=remote-hit)
+to [1.7s (36.8MBps)](https://ge.androidx.dev/s/qn2ueapsujhjq/performance/build-cache?anchor=eyJpZCI6InJlbW90ZS1oaXQifQ&cacheDetails=remote-hit)
+to download the same remote cache entry.
+
+Moral of the story? Always be buffering your input streams! How did I not learn
+this from [I/O profiling](({% post_url 2023-03-02-Profiling-Your-IO %})) earlier
+this year?
diff --git a/assets/2023-12-21-flamegraph.png b/assets/2023-12-21-flamegraph.png
new file mode 100644
index 0000000..f912cd5
Binary files /dev/null and b/assets/2023-12-21-flamegraph.png differ
diff --git a/assets/2023-12-21-network.png b/assets/2023-12-21-network.png
new file mode 100644
index 0000000..880ce25
Binary files /dev/null and b/assets/2023-12-21-network.png differ
diff --git a/assets/2023-12-21-task-details.png b/assets/2023-12-21-task-details.png
new file mode 100644
index 0000000..8ca8a64
Binary files /dev/null and b/assets/2023-12-21-task-details.png differ
diff --git a/assets/2023-12-21-timeline.png b/assets/2023-12-21-timeline.png
new file mode 100644
index 0000000..57fec73
Binary files /dev/null and b/assets/2023-12-21-timeline.png differ