By Simon Stevens, PowerFlex Engineering Technologist, EMEA May 2024.
I am currently writing this blog during the month of May. If Christmas Carols were allowed to be heard on the airwaves at this time of the year (Spoiler Alert: anything Christmas-related should remain ‘hidden from view’ until we have got Halloween done and dusted, but you try telling that to the people who run supermarkets in the UK!) then it might explain why the festive ditty “It’s the most wonderful time of the year” is currently stuck in seemingly never-ending permanent loop playing within my head! But thankfully, Dear Reader, there are a couple of good reasons to explain this strange occurrence, both of which thankfully mean that I am not, in fact, going ever-so-slightly-round-the-bend. Reason number 1 – if it is the month of May, then it must be Dell Technologies World! And Reason number 2 – if it is Dell Technologies World, then that often means (amongst hundreds of exciting announcements) … that a new version of PowerFlex gets announced during DTW!
As it happens, this was indeed the case this year – Dell PowerFlex 4.6 will be RTS on 30th May 2024. This follows hot on the heels of DTW 2024, where, amongst the AI-mania, PowerFlex took centre stage during a number of Technical Breakout sessions. This included a couple that focused on building the multicloud with software-defined infrastructure, as well as sessions that presented best practices for delivering IaaS using infrastructure as code. Anything to do with AI was a central theme this year, this included demonstrations and sessions around AI/ML use cases with PowerFlex, as well as a demo that showcased a unique and robust solution for high-frequency trading and market analysis applications using PowerFlex, AMD compute nodes and NVIDIA GPUs at the KIOXIA booth.
Full Lifecycle Management for 16G Nodes
With DTW2024 firmly behind us, all thoughts can now turn to the release of PowerFlex 4.6, which brings with it lots of new capabilities that our customers have been waiting for. The most important of these (in my opinion) is the addition of full PowerFlex Manager Life Cycle Management support for the next generation of PowerFlex Nodes. This means that customers can now take advantage of the latest Dell PowerFlex nodes, underpinned by various Dell 16G PowerEdge node options. This will include support for 1U & 2U 16G Nodes with AMD & Intel CPU options for both rack size. Figure 1 below summarises the 4 new 16G PowerFlex node options that will be supported by PowerFlex 4.6, along with the key features of each option.
The major changes from the 15G node options are highlighted in the figure above in red font. We all know that each newer CPU generation brings faster CPUs with more cores, as well as larger total RAM counts. However, one of the major changes with the 16G Intel nodes is the introduction of SDPM – Software Defined Persistent Memory. This is a totally new hardware solution that has been introduced specifically at the request of the PowerFlex team, who needed a hardware solution to replace the NVDIMM-N hardware that provided a Persistent Memory solution for the earlier 14G & 15G Intel storage nodes.
PowerFlex uses Persistent Memory for Fine Granularity Storage Pools, which are typically used by customers who wish to benefit from enabling compression within their PowerFlex storage pools. With the 16G nodes, Dell have introduced Software Defined Persistent Memory with Intel nodes as the mechanism to provide persistent memory. This technology emulates NVDIMMs, with the result that no changes to PowerFlex software is required.
SDPM is a solution that that combines standard DDR5 memory modules with a BBU (Battery Backup Unit) and a VOSS drive (Vault-Optimized Storage Solution). The VOSS is a 960GB m.2 NVMe drive. SDPM therefore allows the user to allocate a portion of DRAM memory to be used as Persistent Memory. The amount of DRAM that needs to be allocated as persistent memory is dependent on the storage capacity within the node. While SDPM itself can be configured from 32-128GB, users will only need to configure between 32 – 96GB per PowerFlex storage (or hyperconverged) node. SDPM is configured via a BIOS setting when the node is initially configured.
The battery in the BBU system has been designed to power the system while data is copied from memory over to the VOSS. It has also been designed to support two back-to-back power loss events. The persistent memory data gets copied back from the VOSS into memory upon a power restore, ensuring that all of the persistent state information used by any PowerFlex Fine Granularity Storage Pools is ready to go & in the same state that it was before the power event occurred.
One of the major benefits seen when using SDPM is that it is approximately 70% better for writes than NVDIMMs – meaning that with PowerFlex 4.6 running on 16G nodes, our customers will benefit from seeing a huge increase in write performance on compression-enabled storage pools. It goes without saying that 16G nodes also deliver noticeable read performance gains vs the previous 15G nodes as well. I expect to see 16G PowerFlex Nodes flying off the shelves over the coming weeks and months!
PowerFlex 4.6 Management & Orchestration Enhancements
The PowerFlex Engineering Team have been super-busy adding improvements to the PowerFlex Management Platform (PFMP) with PowerFlex 4.6. Customers who have been using PFMP since its introduction two years ago will be delighted to see improvements made to the HA & resiliency capabilities of several of the major PFMP services. Users should now see a 5-10 second failover introduced for the following PFMP services: Alerts, Event History & UI, the PowerFlex Block Front-End, Gateways, KubeOps, PowerFlex Status and SSO. Likewise, users will be happy to see improved messages in the UI during any periods of service recovery. Finally, the introduction of parallel, multiple simultaneous upgrades of compute nodes will reduce the time it takes to perform upgrades on compute-only services.
PowerFlex 4.6 also sees the addition of full automation for deployment and Life-Cycle Management for VMware ESXi 8.0. This means that PowerFlex Manager is now able to co-ordinate the upgrade from ESXi 7.x to 8.x. Other LCM improvements include the introduction of dynamic enabling and disabling of the SSH for LCM operations.
PowerFlex 4.6 Security Enhancements
Security is quite a wide-ranging area, so I will group a range of new capabilities under the “umbrella” that is this section that covers security. First off, let me start off by saying that “CloudLink is Dead! Long Live PEEKS!”. What this means is that CloudLink has recently been renamed to “PowerFlex Enterprise Encryption & KeyStore Services” – aka ‘PEEKS’ for short. CloudLink went End-Of-Sale to non-PowerFlex customers at the end of March 2023, hence the new name reflects this. Starting with PowerFlex 4.6, there will be automation for PEEKS/CloudLink 8.1 for users who have the correct licensing.
Security-conscious customers will be interested to hear that we have also introduced Secure Component Validation (SCV)/TPM enablement on all PowerFlex nodes. SCV ensures that the PowerFlex node hardware has not been charged between leaving the Dell factory and arriving at the customers site. SCV is an optional feature that is supported for both 15th & 16th Generation PowerFlex appliance nodes. If customers choose this when ordering their nodes, the TPM is enabled and an SCV license is loaded into the iDRAC at the factory. Once the hardware has been delivered, a customer can verify the SCV status of every node by running the SCV tool that can be obtained form the Dell Support site (https://support.dell.com).
One interesting little fact that I stumbled over whilst reading up on SCV is that it makes use of something called the Security Protocol and Data Model (SPDM). SPDM is a protocol that is used for establishing security capabilities and authenticity between hardware components. SPDM allows message exchange between iDRAC and end devices such as storage controllers and NIC controllers. Note the similarity between this acronym and the Software-Defined Persistent Memory capability that we have introduced with 16G hardware – SDPM. I await with baited breath to see if this similarity in acronym names ends up causing any confusion for both customers and Dell support teams… ;o)
Back to the new things included with PowerFlex 4.6. Audit logs for PowerFlex File and PEEKS have been added, matching the audit logs that are already included with PowerFlex itself. PEEKS supports adding up to 4 syslog servers and now includes Include events, security events, audit logs, and authentication logs across multiple servers, enhancing monitoring capabilities beyond previous versions
These logs can be transmitted either directly to the Dell ESE Remote Server or collected via PowerFlex Manager. If using PowerFlex Manager, the admin user will need to configure and enable a Notification Policy that contains the source and destination applicable for the audit log collections to occur.
PowerFlex 4.6 – APEX AIOps (formerly CloudIQ) Enhancements
Yes, I know – yest another product name change! APEX AIOps is the new name for CloudIQ, but rest assured, the APEX AIOps Team have not just been spending their time choosing a new name for themselves – they have been busy adding in yet more functionality that benefits PowerFlex customers!
Performance and Capacity Reporting for PowerFlex systems has been expanded to cover up to 17 possible metrics. Up to 2 years of historical performance reporting can be obtained on any system and 1 to many systems can be shown for each metric. PowerFlex Storage Pools can now also include capacity forecast, as shown in Figure 3 below:
If you have never seen APEX AIOps at work, you can look at the following simulator here https://cloudiq.dell.com/simulator/index.html#/overview to see its capabilities (please note that the URL still refers to the CloudIQ branding – watch this space!).
PowerFlex 4.6 – On-Demand SDC Compilation
I have possibly saved the best piece of news to last. In fact, this functionality has been available to selected customers for a few weeks now, but the PowerFlex 4.6 release sees this capability being officially announced to the world. This is one of those asks that many PowerFlex customers have dreamed about, and finally, those dreams have come true!
If you have read this far, you are very likely already aware that most PowerFlex customers need to install the SDC into the compute hosts that wish to access and use PowerFlex block storage resources. The SDC is a kernel driver and of course, Linux distributions release kernel updates on a regular basis. Some kernel driver updates can “break” the SDC – for example, a driver that was compiled against an older kernel can become incompatible when the customer upgrades their Linux kernel to a newer version, which means that we, as the driver vendor, have to recompile the driver against the new kernel.
In today’s security-conscious world, Linux kernels are being updated far more frequently than ever before, and there appears to be no sign yet of this changing anytime soon. At the same time, PowerFlex is being integrated into a growing number of cloud platforms, meaning that we need to support more container hosts that run specialised kernels (for example CBL-Mariner, Red Hat Core OS, SUSE Harvester, etc).
The way that we deal with this today is a solution that has existed for several years. As well as the precompiled SDC drivers that come with the PowerFlex software bundles, there is an FTP repository (ftp.emc.com) that contains a driver repository, that customers can access by setting up a local repository that points to this site. When the SDC driver service running on a host starts, it attempts to load an SDC that best fits the running kernel and if that fails then the service will reach out and download the correct SDC from the FTP repository. Likewise, for Container platforms, the PowerFlex “SDC init” container includes a handful of compiled SDC drivers for various kernels but can also be configured to connect to the FTP repository if needed.
While the above model has worked well in the past, todays modern IT challenges require better solutions to fix todays issues. As an example, Linux distributions are releasing ABI-breaking kernel updates more frequently, while at the same time, the “SDC init” container is becoming bloated by having multiple drivers to support all possible container hosts. Plus, a lot of customers disable access to external FTP repositories. Hence a better solution was needed that works better, not only for today but also for going forwards.
The solution is that we have introduced is “SDC On-Demand Compilation”, which pretty much does what it says on the tin. The SDC gets compiled on the customers host machine, which ensures a perfect match of the driver to the kernel whenever the kernel gets updated. In short, the SDC package has been modified to contain a tar archive that encapsulates some pre-compiled kernel agnostic object files, kernel API wrappers in source code form and a script – build_driver.sh – that initiates the compilation as & when needed.
This solution approach works well and has been used successfully by other software companies in the past. The advantages of doing this are clear – for a start, it means that there is no need to build and store thousands of SDC binaries within our repository! However, more importantly, for our customers, this will also mean that critical ABI-breaking kernel releases (for example, urgent security patches) can now be easily applied by customers without any delay, as the SDC will auto build on the fly during reboot.
Clearly, we still need to provide a repository to store the latest tarballs for the latest kernel APIs, and we will need to handle any customers asking for as SDC for any new OSs via Product Management, but overall, this new process for handling updated SDC drivers will be both faster and more secure for our customers going forwards.
I am going to end this latest blog here, even though there are still lots of other goodies left to discover with the PowerFlex 4.6 release. Clearly, full details can be found in the product documentation, but you can also reach out to your local Dell PowerFlex Specialists should you wish to hear more. I have covered most of the “headline grabbers” in the pages above. One never knows – perhaps one of topics I discussed above may have even made you think that Christmas had indeed come early this year!
