Realize better value and performance migrating from Azure Database for PostgreSQL – Single Server to Flexible Server with AMD EPYC

Realize better value and performance migrating
from Azure Database for PostgreSQL – Single Server
to Flexible Server with AMD EPYC
The upcoming retirement of Azure Database for PostgreSQL –
Single Server, which is scheduled for March 28, 2025,1
highlights
the pressing need for organizations to shift to newer instances
to continue running their databases. Through a simple migration
progress, transitioning from Azure Database for PostgreSQL –
Single Server to Azure Database for PostgreSQL – Flexible Server
with AMD EPYC™
processors can help organizations maintain
business continuity while simultaneously offering considerable
performance advantages.
At Principled Technologies, we completed hands-on testing of
both Flexible Server and Single Server, explored the performance
and cost differences between the two solutions, and migrated
a database to Flexible Server using the Azure offline migration
tool. Using the HammerDB benchmark, we found that Azure
Database for PostgreSQL – Flexible Server with AMD EPYC
handled 4.71 times the NOPM of the Single Server instance while
offering 2.85 times the IOPS and 3.88 times the performance/
dollar. Completing the migration using the Azure migration tool
made moving from Single Server to Flexible Server and reaping
these performance benefits a simple task. These significant
performance gains show that moving to Azure Database for
PostgreSQL – Flexible Server makes business sense, as the newer
solution offers the opportunity to boost performance to improve
customer experiences.
Stronger OLTP
performance for
PostgreSQL databases
4.71x the new orders per
minute (NOPM)*
Better database
throughput
2.85x the input/output operations
per second (IOPS)*
Improved value
3.88x the performance per dollar*
*Compared to the Azure Database for PostgreSQL - Single Server instance we tested
Realize better value and performance migrating from Azure Database
for PostgreSQL – Single Server to Flexible Server with AMD EPYC
February 2024
A Principled Technologies report: Hands-on testing. Real-world results.

A look into the new features that Azure Database for
PostgreSQL – Flexible Server provides
Azure Database for PostgreSQL – Flexible Server is a full managed, production-ready
database service that stands out with its high-availability (HA) features, supporting single and
multiple availability zones.2
Its architecture separates compute and storage: the database
engine runs in a Linux VM, with data files stored on Azure storage, maintaining three
synchronous copies for data durability. For enhanced reliability, the service offers a zone-
redundant setup, employing a standby server in a different availability zone for synchronous
data replication and immediate activation during failovers.3
Automated maintenance is a key aspect of this service, including patching of the underlying
hardware, the operating system, and the PostgreSQL database engine. The service also
offers security and software updates and minor PostgreSQL version upgrades. Additionally,
the service provides data safety through automatic backups stored in zone-redundant
storage (ZRS), enabling server restoration during the backup retention period.
Flexible Server offers three compute tiers: Burstable, General Purpose, and Memory
Optimized.4
The Burstable tier caters to low-cost, low-concurrency workloads. The General
Purpose and Memory Optimized tiers specifically target high-concurrency, scalable
production workloads. The service enhances cost-efficiency through its on-demand
server stop-start feature, which immediately stops compute tier billing when the server
becomes inactive.
New features for Azure Database for PostgreSQL – Flexible Server
What does Azure Database PostgreSQL – Flexible Server have that Single Server doesn’t?
According to our research, plenty:
• Enhanced high-availability zone options like zone-redundant HA and same-zone HA,
allowing for more robust data resilience and continuity; support for Geo-replication for
regional redundancy
• A burstable compute tier and an automatic IOPS scaling feature called autoscale IOPS
• Support for Premium SSDv 2 storage, which is “designed for IO-intense enterprise
workloads that require sub-millisecond disk latencies and high IOPS and throughput
at a low cost”5
• Improved data security for data at rest using the Federal Information Processing
Standards (FIPS) 140-2 validated cryptographic module, ensuring data encryption6
• Comprehensive, built-in performance monitoring and alerting, with metrics updated
every minute and a 30-day history available. Users can configure slow-running query
logs, which can be instrumental in pinpointing and resolving performance bottlenecks
in database operations.7
• Granular backup features. The service automatically creates server backups and allows
users to store them locally in redundant or geo-redundant storage, safeguarding
against the loss of data in a variety of situations.8
For more information about Azure Database for PostgreSQL – Flexible Server, visit
https://azure.microsoft.com/en-us/pricing/details/postgresql/flexible-server/.
February 2024 | 2

How we tested
Using the HammerDB benchmark, we compared the OLTP performance of two solutions:
• a Flexible Server instance (Standard_D16ads_v5) with AMD EPYC processors and a 1TB Premium SSD
version 2 volume configured with 51,200 IOPS and throughput of 768 MBps. Dadsv5-series instances
utilize 3rd
Generation AMD EPYC processor in a multi-threaded configuration.
• a Single Server instance (GP_Gen5_16) with a 6.5TB volume, which allowed us to utilize the instance’s
maximum disk IOPS of 20,000.
After our performance testing, we used the offline Azure migration tool to complete an example migration of a
525GB database from Single Server to Flexible Server. For more information about the configurations we used,
as well as step-by-step performance and migration details, read the science behind the report.
About HammerDB
To test OLTP performance, we used the TPROC-C workload from HammerDB—an open-source tool that
tests the performance of many leading databases including PostgreSQL. While HammerDB developers
derived their TPROC-C workload from the TPC-C standard, it is not a full implementation of TPC-C
specifications. As such, the results we cite in this paper are not directly comparable to those officially
published by TPC. HammerDB reports results in both transactions per minute (TPM) and new orders per
minute that a system can handle. For more information about HammerDB, visit www.hammerdb.com.
Flexible Server offered stronger database performance:
HammerDB testing results
One reason you may have postponed your organization’s migration from Azure Database for PostgreSQL –
Single Server to Azure Database for PostgreSQL - Flexible Server is uncertainty about how performance levels
might change. We compared the database performance of both solutions using the HammerDB TPROC-C
benchmark using a 10-minute runtime and 64 virtual users. We tested with the HammerDB option of “All
warehouses,” which ensures the workload exercises nearly all the data instead of a portion. We did this to ensure
data was not cached in memory, which caused more IO. We ran each test three times and report the median of
three runs. Table 1 shows a summary of our test results, which we expand on below. Note: These results reflect
the actual IOPS the instances achieved during the tests. For the Flexible Server instance, we captured IOPS
directly from the Metrics section of Azure portal. For the Single Server instance, we calculated IOPS achieved by
1) taking the IO percent from the Metrics section and 2) multiplying that IO percent by the number of configured
IOPS, which was 20K.
Table 1: HammerDB results for both 16vCPU Azure Database for PostgreSQL solutions. Higher numbers are better.
Source: Principled Technologies
HammerDB testing –
16vCPU configurations
Flexible Server Single Server
Improvement with
Flexible Server
NOPM 30,605 6,490 4.71x
CPU utilization 52.4% 22.1% -
IOPS 37.7K 13.2K 2.85x
February 2024 | 3

Figure 1 shows the NOPM that the two solutions achieved during HammerDB testing. The performance of the
Azure Database for PostgreSQL – Flexible Server instance with AMD EPYC processors dwarfed that of the Single
Server solution, delivering 4.71 times the database performance.
Selecting an instance that handles more NOPM means that you can have more customers on your ecommerce
site searching, making purchases, and more—which could only be good for business.
Figure 1: Database performance, in NOPM, that the Azure Database for PostgreSQL solutions achieved in our HammerDB testing.
Higher is better. Source: Principled Technologies.
Database performance | NOPM, higher is better
16vCPU Flexible Server
16vCPU Single Server
30,605
6,490
Figure 2 shows the IOPS that the instances reported during testing. Again, Flexible Server outperformed Single
Server by a wide margin, delivering 2.86 times the IOPS. The IOPS metric measures the number of reads and
writes that storage can handle. Every application is different, but typically higher IOPS numbers indicate that
users can perform more database actions at once without degrading user experience.
Looking at processor utilization while running a database workload can give insight into how efficiently
your instance is using its available resources. Processor utilization that’s too low might indicate performance
bottlenecks elsewhere (often the storage) and means that you aren’t reaping the performance benefits of the
processors powering your workloads.
As Figure 3 shows, the Azure Database for PostgreSQL – Flexible Server instance made better use of its available
processing resources than did the Single Server solution.
Figure 2: IOPS that the Azure Database for PostgreSQL solutions achieved in our HammerDB testing.
Higher is better. Source: Principled Technologies.
Database throughput | IOPS, higher is better
37,700
13,200
Figure 3: CPU utilization for both Azure Database for PostgreSQL solutions during our HammerDB testing. Higher is better.
Source: Principled Technologies.
Processor utilization | Utilization percentage, higher is better
52.40%
22.10%
February 2024 | 4

About the AMD EPYC 7763 processor
In our tests, the Flexible Server instance used the 3rd
Gen AMD EPYC 7763v processor—a version of the
7763 model. The AMD EPYC 7763 features 64 cores, 128 threads, and 256MB cache. Its base speed is
2.45 GHz with a maximum boost clock speed of up to 3.5 GHz.9
The AMD EPYC 7763v processor powers both General Purpose and Memory Optimized Flexible Server
instances for Azure Database for PostgreSQL. Both instance types offer a range of 2 to 96 vCores, and
up to 900 MB/s of I/O bandwidth, while the General Purpose instances have up to 384 GB of memory,
and Memory Optimized have up to 672 GB of memory.
To learn more about the AMD EPYC 7763 processor, visit https://www.amd.com/en/products/cpu/
amd-epyc-7763.
See increased performance/dollar with Flexible Server
Azure Database for PostgreSQL – Flexible Server outperforms the Single Server option, but how much does a
similarly configured Flexible Server instance cost? Our research shows that Flexible Server offers this significant
database performance jump without incurring a large cost increase. The total monthly cost (using the maximum
monthly hours of 730 hours) for Flexible Server was $2,172.72, while Single Server cost $1788.60. To learn details
about where we sourced this information and how we arrived at our performance per dollar calculations, read
the science behind the report.
As Figure 4 shows, Flexible Server offers an impressive 3.88 times the performance per dollar that a Single Server
instance can provide. This data shows that organizations moving to Flexible Server could expect better value
than you’re currently getting by continuing to use Single Server instances.
Figure 4: Performance per dollar comparison of Azure Database for PostgreSQL – Flexible Server vs. the Single Server offering.
Higher numbers are better. Source: Principled Technologies.
Performance per dollar | NOPM per dollar, higher is better
16vCPU Flexible Server (1TB storage)
Single Server (6.5TB storage)
14.09
3.63
February 2024 | 5

Energy usage with AMD EPYC processors
Recent tests have shown that two-socket AMD EPYC 7763
processor-powered servers demonstrated up to 85% better
energy efficiency than the most comparable non-AMD-
based servers.10
AMD studies show organizations were
able to consolidate multiple servers onto fewer servers,
reducing energy and green house gas emissions and
thereby reducing operating expenditures including power
usage, rack footprint, and more.11
To learn about AMD energy efficiency initiatives,
visit https://www.amd.com/en/campaigns/epyc-
energy-efficiency.
Moving to Flexible Server is easy using
the migration tool in the Azure UI
Thoughts of migration woes may be holding your IT team back
from making the move to Flexible Server. We found that using the
offline migration tool in the Azure UI makes it easy to move your
databases onto the newer solution.
In our tests, we were able to easily migrate a 525GB database
from a 16vCPU Single Server instance to a 16vCPU Flexible Server
instance based on AMD EPYC. While we used the offline migration
option in our tests, Azure offers both offline and online options,
and users can monitor the entire process using the Azure portal.
A benefit to this migration tool is its ability to migrate the entire
server, including users; roles and privileges; schema; and data, in a
secure and reliable way.
To see how we completed our migration, read our step-by-step
guide in the science behind the report.
To learn more about how to use the Azure migration tool visit
https://learn.microsoft.com/en-us/azure/postgresql/migrate/
concepts-single-to-flexible.
February 2024 | 6

Azure Database for PostgreSQL – Flexible Server offers performance
and cost benefits
It’s time to move your databases from Azure Database for PostgreSQL – Single Server instances now, before
March 2025 when Microsoft Azure retires the service. Our hands-on testing shows that moving to Azure
Database for PostgreSQL – Flexible Server is a simple process that can offer significantly better database
performance and value. With as much as 4.71 times the NOPM on HammerDB and up to 3.88 times the
performance per dollar compared to Single Server, moving to Azure Database for PostgreSQL – Flexible Sever
with AMD EPYC processors can give you higher performing databases for lower relative cost.
1. Sunil Agarwal, “Retiring Azure Database for PostgreSQL Single Server in 2025,” accessed February 3, 2024,
https://techcommunity.microsoft.com/t5/azure-database-for-postgresql/retiring-azure-database-for-postgresql-single-serv-
er-in-2025/ba-p/3783783.
2. Microsoft, “Azure Database for PostgreSQL,” accessed February 4, 2024,
https://azure.microsoft.com/en-us/products/postgresql/#features.
3. Microsoft, “Azure Database for PostgreSQL – Flexible Server,” accessed February 4, 2024,
https://learn.microsoft.com/en-us/azure/postgresql/flexible-server/overview.
4. Microsoft, “Azure Database for PostgreSQL – Flexible Server Pricing: Frequently Asked Questions,” accessed February 4,
2024, https://azure.microsoft.com/en-us/pricing/details/postgresql/flexible-server/#faq.
5. Microsoft, “Deploy a Premium SSD v2,” accessed February 15, 2024,
https://learn.microsoft.com/en-us/azure/virtual-machines/disks-deploy-premium-v2?tabs=azure-cli.
6. Microsoft, “Azure Database for PostgreSQL – Flexible Server: Enterprise Grade Security” accessed February 4, 2024,
https://learn.microsoft.com/en-us/azure/postgresql/flexible-server/overview#enterprise-grade-security.
7. Microsoft, “Azure Database for PostgreSQL – Flexible Server: Monitor and Alerting,” accessed March 4, 2024,
https://learn.microsoft.com/en-us/azure/postgresql/flexible-server/overview#monitor-and-alerting.
8. Microsoft, “Azure Database for PostgreSQL - Flexible Server: Automatic Backups,” accessed March 4, 2024,
https://learn.microsoft.com/en-us/azure/postgresql/flexible-server/overview#automatic-backups.
9. AMD, “AMD EPYC 7763,” accessed February 2, 2024, https://www.amd.com/en/products/cpu/amd-epyc-7763.
10. AMD, “AMD EPYC 7003 Series Processors,” accessed February 20, 2024,
https://www.amd.com/en/processors/epyc-7003-series.
11. AMD, “Data Center Sustainability,” accessed February 1, 2024,
https://www.amd.com/en/corporate/corporate-responsibility/data-center-sustainability.html#:~:text=In%20addition%20
to%20the%20environmental,and%20may%20require%20fewer%20servers.
Principled Technologies is a registered trademark of Principled Technologies, Inc.
All other product names are the trademarks of their respective owners.
For additional information, review the science behind this report.
Principled
Technologies®
Facts matter.®
Principled
Technologies®
Facts matter.®
This project was commissioned by Microsoft.
Read the science behind this report at https://facts.pt/AK7am5k
February 2024 | 7

Realize better value and performance migrating from Azure Database for PostgreSQL – Single Server to Flexible Server with AMD EPYC

More Related Content

Similar to Realize better value and performance migrating from Azure Database for PostgreSQL – Single Server to Flexible Server with AMD EPYC (20)

More from Principled Technologies (20)

Recently uploaded (20)

Realize better value and performance migrating from Azure Database for PostgreSQL – Single Server to Flexible Server with AMD EPYC