Servers: The Right Compute
cancel
Showing results for 
Search instead for 
Did you mean: 

Are HPE 8GB NVDIMMs Too Big?

emull

Bret Gibbs at HP Garage.jpg

Guest blog by Bret Gibbs, Persistent Memory Product Manager, HPE  

Yes – you did read that title correctly and now that I have your attention let me explain how I came up with that controversial title!

Since we first announced the HPE 8GB NVDIMM in early 2016, I often get asked:  “When will you have larger persistent memory?”  The leading assumption behind that question is that 8GB of capacity may not be large enough for today’s demanding workloads.  While that may be true if you are thinking about accelerating an entire multi-TB data workload on persistent memory, you may not be aware of how even small capacity NVDIMMs can bring big performance benefits to your existing database applications.  But let’s start with some basics first.

As we rolled out the HPE 8GB NVDIMM on HPE ProLiant servers, we have been working with our software ecosystem partners getting them ready to use persistent storage in memory and tap into the full performance and data resilient potential of HPE Persistent Memory.  On the operating system front, we are the first NVDIMM server vendor in the market with a Windows Server 2012 R2 driver.  Microsoft has also enabled HPE Persistent Memory support in Windows Server 2016.  SUSE recently announced NVDIMM support in SUSE Enterprise Linux 12 Service Pack 2.  Red Hat announced NVDIMM support in Red Hat Enterprise Linux 7.3.  HPE showcased HPE 8GB NVDIMMs on a pre-production version of VMware at VMworld 2016 in Las Vegas in August 2016 (stay tuned to VMware for future announcements on NVDIMM support).  It’s clear the operating system software community understands the HPE’s vision of building a strong operating system ecosystem around this exciting technology.

How about application support for HPE 8GB NVDIMMs?  First, let me make an important point:  Today’s software applications can take advantage of the performance advantages of HPE 8GB NVDIMMs.  These existing software applications communicate with NVDIMMs as block storage devices – just like they communicate with HDDs and SSDs today.  While preserving the storage semantics of data durability, you’ll see immediate performance benefits because you are accelerating data on DRAM on the low-latency memory bus.  NVDIMMs are ideal for hardware storage IO bottlenecks - like database transaction log writes - which typically have longer write latencies due to the storage hardware not keeping pace with the write commands coming from the application.  Every database transaction has to get written to a log and much like rush hour in a major city, those transactions tend to backup waiting for the write to commit to the storage device and creating that bottleneck.  What if instead of writing to traditional storage media first, we could write those transactions to faster and data-resilient NVDIMMs?  In this scenario, NVDIMMs speed up your database performance by reducing the storage bottleneck (transaction log writes moved to faster, lower-latency media) and improving overall database performance.  That is awesome but we can do better!

The real performance potential of HPE Persistent Memory is unleashed when software applications are fundamentally changed to communicate with persistent memory technology as byte-addressable storage – much in the same way these applications communicate with server memory. As an example, database transactions don’t have to wait for previous transactions, memory range is recoverable across application crashes and server resets and power losses and can be easily and efficiently replicated.

Byte-addressable storage capability is a realization because the hardware technology can now keep pace with the software application issuing the commands.  New programming models are necessary because storage devices can now keep pace with CPU’s issuing instructions.  Byte addressable storage capability paired with HPE Persistent Memory technology delivers new levels of performance to your workloads while keeping your data persistent!

Microsoft has been a strong advocate of HPE Persistent Memory technology both on the operating system and software application front.  They not only understand the vision of byte-addressable storage but are delivering it with their new Microsoft SQL Server 2016 Tail of the Log functionality.  Microsoft changed SQL Server 2016 Tail of the Log (portion of the transaction log where the most recent transactions are logged) to communicate with persistent memory as byte-addressable storage.  We’ve been showcasing this functionality at HPE Discover events and most recently at Microsoft Ignite 2016 (actual demo starts at ~23:55) in Atlanta this year.  We are also showing this at HPE Discover London 2016 (see details at the end for sessions and demos).

So what exactly does this Tail of the Log demo deliver?  We have an HPE ProLiant DL380 Gen9 E5-2600v4 server running Microsoft Windows Server 2016 and Microsoft SQL Server 2016.  We have the log files stored on our fastest HPE NVMe SSDs with a single HPE 8GB NVDIMM (782692-B21) with Tail of the Log byte-addressable storage functionality enabled.  The baseline test runs the transaction log on the NVMe SSD only.  We then added the single NVDIMM for the transaction log and enabled Tail of the Log functionality.  The results speak for themselves:  doubled the number of transactions/second and cut the latency in half using a single NVDIMM versus the fastest NVMe SSD on ProLiant servers.  Detailed configuration and results are outlined in the chart below:

SQL Server 2016 Tail of Log.png

Slides are powerful at conveying the story but sometimes it’s just better to see it with your own eyes.  Want to see the demo in action?  Click on the slide below to see what byte-addressable storage can deliver to your business:

Click here to view: SQL 2016 Tail of the Log Demo

SQL2016 Tail of Log.png

 

  • SQL Tail of the Log demo is focused on 3 key areas
    • New levels of performance (2x faster than fastest NVMe SSDs)
      • “Game-changing performance for SQL Server 2016”
      • ~60K vs. 120K transactions per second
    • CPU Utilization (100% CPU utilization with PMEM vs ~63% with NVMe)
      • “You paid for processor so why not get full use of it”
    • Less kernel time (apps run more efficiently with fewer IOs using HPE PMEM)
      • “This is why apps must change to use byte-addressable storage”

So how does this all tie back to my title: “Are 8GB NVDIMMs Too Big?” Did you know that the typical size of a SQL Server Tail of the Log is 20MB?  That’s MB – not GB.  And the good news here is that you can have multiple “Tail of the Log” instances running on a single NVDIMM to get even better utilization of your single HPE 8GB NVDIMM for your deployed scenarios with multiple database instances.  This illustrates an important point about HPE Persistent Memory having the “right-sized capacity” for the job and how you can do great things with small capacity points. 

We will have much larger Persistent Memory coming in the next-generation of HPE ProLiant servers that will address larger workload needs and completely change the game yet again.  As a matter of fact, we will be showcasing one of the next-generation HPE Persistent Memory solutions at the NDA Tech Showcase at HPE Discover London 2016!

For those attending HPE Discover London 2016, come visit us at one of these demos or sessions:

Tuesday, Nov 29 Session

  • 11750 12:30pm - HPE Persistent Memory with Microsoft and VMware

Wednesday, Nov 30 Session

  • 11745 12:30pm - Accelerate database performance with HPE Persistent Memory

Demos:

  • Microsoft SQL Server 2016w/ HPE Persistent Memory
  • Accelerate workloads with HPE Persistent Memory & HPE 3PAR 3D Cache Preview
  • Next Generation HPE ProLiant Rack Servers and HPE Persistent Memory (NDA required)

Bret Gibbs is the Persistent Memory Product Manager for Hewlett Packard Enterprise.  He has worked in various Product Management roles in storage and servers during his 22 year career in the IT Industry.

About the Author

emull

Eric Mullins is an HPE Digital Marketing Manager covering infrastructure news, insights and analysis from multiple segments. Come along for the ride to find out more about the future of data center technology

Events
Jan 30-31, 2018
Online
Expert Days - 2018
Join us online to talk directly with our HPE experts during online Expert Days. Find information here about past, current, and upcoming Expert Days. P...
Read more
28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
View all