PublishedApplicationData

Spectre and Meltdown Published Application Performance

This page collects public information posted by reliably identified sources that describes (to at least a minimal degree) how workload performance has been affected by the Meltdown and Spectre mitigations. We've attempted to identify what fixes were in place at the time of the run, as most reports unfortunately do not indicate those — though given the complexity of the situation, it's not surprising that's unclear.

For the tests which were run against AWS, note Amazon applied a first set of updates on Jan 4th, and a further set on Jan 12th, the second set generally noted as reducing the performance impact. Note also that for full protection the guest microcode and OS also needs to be updated, but not all tests implemented or compared using them.

Workload

Source

Mitigations

Substrate

Regression Measured

Reference

Apache Spark

Databricks

Host: KPTI, retpoline? Guest: ?

AWS r3 and i3

3%-5%

link

Cassandra

The Last Pickle

KPTI only

Mobile Broadwell

20-50% latency increase

link

Elasticsearch

elastic.co

KPTI only

Skylake

0-4%

link

Hadoop (CDH)

Cloudera

?

?

3-12%

link

nginx

George Liu

KPTI Only

Haswell i7 4790K

5.5%

link

Percona

Percona

KPTI, IBRS, IBPB

Haswell

15-25%

link

PostgreSQL

2nd Quadrant

KPTI Only

Skylake

7%

link

ScyllaDB

Scylla

Host: KPTI, STIBP? Guest: KPTI Only

AWS i3.16xlarge

6% (guest KPTI delta only, no baseline)

link

SecurityTeam/KnowledgeBase/SpectreAndMeltdown/PublishedApplicationData (last edited 2018-01-25 02:28:05 by seth-arnold)