UbUbuntu Community
Community homeReleases Service deployment System tuning

Ubuntu Community · Why does Ubuntu use twice the memory of comparable tools?

157
Ubr/ubuntu·posted by li_ming·2 hours agoJobs

Why does Ubuntu use twice the memory of comparable tools?

It took me two weeks of on-and-off digging and plenty of wrong turns. Writing the process down as it happened so the next person spends less time.

One last trap: in container environments remember to adjust the memory-related parameters in step. Otherwise the host limit and the process expectation disagree, and the symptom is intermittent, unreproducible failure.

# Load test: do not jump straight to the max concurrency.
# Ramp it up, otherwise you miss the knee.
for c in 50 100 200 400 800; do
  wrk -t8 -c$c -d60s --latency http://127.0.0.1:8080/api/feed
  sleep 20
done

The first thing was to collapse the variables. We were changing config and upgrading the version at the same time, and afterwards nobody could say which change caused what. We rolled back to moving one variable at a time, re-ran three times, and only then did the curve settle. Tedious, but not skippable.

Worth noting: the official docs do cover this, just in a very inconspicuous spot. I only found it reading the source comments, where the author explains the reasoning — roughly "so that it degrades into predictable behaviour in extreme cases".

82 comments

82 comments

M
Ttang_hao·2 days agoedited

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

485
Rran_bo·2 days ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

134
Oops_wangMod·2 hours ago

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

18
Nnikic·just now

Saved. I am reworking this area this week — this saves a lot of wrong turns.

418
Mmike_xu·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

398
Cchen_dev·2 days ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

364
Lli_ming·5 hours agoedited

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

357
Sslow_query·2 days agoedited

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

337
Ddev_zhou·yesterday

This is not a Ubuntu problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

167
Ttang_hao·1 hour ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

293
Aalice_dev·1 hour ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

288
Llinlin·12 minutes ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

9
Hhuang_keMod·12 minutes agoedited

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

250
Rran_bo·2 hours ago

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

217
Wwinter·just now

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

298
Lli_ming·2 days ago

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

24
Rran_bo·2 days ago

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

216
Rrase·2 days agoLevel 6

Saved. I am reworking this area this week — this saves a lot of wrong turns.

384
Rran_bo·28 minutes ago

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

90
Hhuang_ke·yesterday

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

478
Bbob_chen·12 minutes agoedited

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

1
Zzhu_zong·28 minutes ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

70
Ttang_hao·2 days ago

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

192
Wwinter·2 days ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

143
Kkite·2 days ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

30
Aalice_devOP·2 days ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

22
KkiteOP·2 days ago

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

190
Lli_ming·just now

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

245
Ttang_hao·just now

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

235
Lli_ming·2 days ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

217
Oops_wang·2 days ago

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

204
Wwinter·28 minutes agoedited

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

127
Sswoole_leeOP·5 hours ago

Saved. I am reworking this area this week — this saves a lot of wrong turns.

144
Mmike_xu·2 days ago

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

150
Rran_bo·2 days ago

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

113
Oops_wang·2 days ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

177
Nnikic·2 days agoedited

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

171
Aalice_dev·2 days ago

There is actually a simpler fix that needs no architecture change: move this check up to the gateway and the problem disappears. The cost is one extra lookup at the gateway.

164
Bbob_chen·2 days ago

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

4
Ddev_zhou·2 days ago

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

163
Llinlin·2 days ago

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

175
Kkite·2 days ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

150
Zzhou_yi·2 days ago

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

3
Zzhou_yi·2 days ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

117
Lli_ming·3 minutes ago

Agreeing with the above. One addition: with this option enabled the GC count in your metrics doubles, so adjust the alert threshold at the same time or it will keep firing.

97
Aalice_dev·2 days ago

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

88
Bbob_chenMod·1 hour ago

One counter-example: below Ubuntu 7.4 the semantics of that code are different, so do not copy it verbatim. We got burned in staging and rolled back once.

72
Sslow_query·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

127
Rran_bo·2 days ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

8
Rran_bo·2 days agoedited

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

63
Ttang_hao·just now

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

57
Zzhu_zongMod·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

57
Ddev_zhouMod·28 minutes ago

This is not a Ubuntu problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

32
Mmike_xu·2 days ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

239
Ttang_hao·2 days agoedited

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

209
Ddev_zhouOP·just now

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

16
Wwinter·2 days ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

361
KkiteOP·2 days agoedited

Has anyone run a controlled experiment? I did, reducing it to a single variable, and the difference was 4% — within noise. So I suspect the main cause is something else.

146
Oops_wang·just now

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

71
Cchen_dev·yesterday

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

31
Ddev_zhou·2 days agoedited

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

19
Sslow_query·2 days ago

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

16
Bbob_chen·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

86
Bbob_chen·3 minutes ago

We have run this in production for two years without hitting it. That said, we never reached this scale, so our experience is not really evidence here.

4
Llinlin·3 minutes ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

282
Nnikic·2 days ago

A question: what changes in a container with a 512Mi memory limit? That is how we run it in production.

93
Zzhu_zong·2 days ago

Thanks for sharing real numbers — far more useful than the articles that only cover concepts.

86
Kkite·2 days ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

38
Rran_bo·2 days ago

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

239
Zzhu_zong·2 days ago

This is not a Ubuntu problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

18
Zzhou_yi·28 minutes ago

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

202
Sslow_query·2 days agoedited

I see point 3 differently. The trade-off depends on your read/write ratio: read-heavy with little writing means caching actually widens the inconsistency window.

17
Ddev_zhouOP·5 hours ago

Can you give a minimal reproduction? I ran it locally for ten minutes and could not reproduce on macOS with the latest version.

473
Sswoole_lee·2 days ago

Worth learning from this debugging approach. We went straight at the logs and took a much longer route.

502
Zzhu_zong·yesterday

Saved. I am reworking this area this week — this saves a lot of wrong turns.

1
Sslow_query·2 days ago

This is not a Ubuntu problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

180
Nnikic·2 days ago

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

1
Ttang_hao·2 days agoedited

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

3
Kkernel_panic·2 days ago

This matches what we see in production. We only hit it past 3k QPS; the earlier load tests showed nothing — the test traffic was too clean, with no long-tail requests.

100
Zzhu_zong·2 days ago

I just read the Ubuntu source — the author actually explains the reasoning in a comment, roughly "so that it degrades into predictable behaviour in extreme cases".

3
Zzhu_zong·2 days agoedited

This is not a Ubuntu problem, it is a usage problem. The docs say this API is not thread-safe and you must lock around it yourself.

1
Hhuang_ke·2 days ago

Sharing our numbers, 8 cores 16GB, same scenario:

| Concurrency | P50 | P99 |
|---|---|---|
| 200 | 12ms | 88ms |
| 500 | 31ms | 340ms |

P99 clearly collapses at 500 concurrency, which lines up with your knee point.

1

This is the post detail page /en/c/ubuntu/post/p0. Posts and comments are generated deterministically from a seeded PRNG, so the same post always renders the same content and the link can be shared, reloaded and indexed. In production this page reads MySQL for the post, Redis for hot-post caching, and fetches the whole comment tree in a single query on the path column.

See the database schema →