At Dell Technologies World this week, Michael Dell announced the availability of new storage products for the public cloud. Are these really new solutions or just renamed and relaunched products that already existed before the event? We dig into the details to try and find out.
First of all, it’s worth re-emphasising that (as we’ve highlighted before) Dell Technologies has dropped most of the independent influencers and analysts, so we have no pre-briefing information or insights other than what’s been publicly announced. As a result, there may be some inaccuracies or incorrect assumptions, but we will link to sources wherever possible.[side note: you may notice a theme evolving here….]
Dell first announced Project Alpine back in January 2022. In the press release quoted, the aim was to bring Dell’s “flagship block and file storage platforms” to leading public clouds. We’ve specifically picked out this section of the press release to highlight the word “flagship”, which to us means PowerScale, PowerMax and PowerStore. Remember that PowerScale is the Isilon heritage product, PowerMax is the Symmetrix heritage product, and PowerStore is the Clariion/VNX/Unity heritage platform.
Following on from some early news at DTW 2022, we wrote this piece (registration required) which highlighted the potential challenges of porting PowerStore and PowerMax to the public cloud.
If we just concentrate on PowerStore, the architectural hurdles are significant. The PowerStore platform uses the legacy architecture developed originally in Clariion and updated through VNX and Unity. This introductory PDF provides details of the hardware setup. Some points to consider:
- PowerStore uses NVRAM with battery backup to preserve data in the event of a power loss. There’s no easy way to emulate this in the public cloud, as the loss of a virtual instance can mean the loss of any and all data it used.
- PowerStore uses shared NVMe NVRAM and NVMe SSDs in the base chassis. Again, there’s no easy way to share performance volumes between virtual instances in the public cloud and to guarantee that a new instance can be brought up (automatically) to use them – this scenario that would represent a controller failure and/or replacement.
- A significant majority of PowerStore users will connect via Fibre Channel. There is no equivalent in the public cloud (although NVMe/TCP support is possible). This means customers need to rethink architectural designs to build iSCSI or NVMe/TCP-capable solutions.
- Finding a suitable virtual instance type could be a challenge. PowerStore needs a minimum of six drives, with an additional two NVRAM drives in most models and at least 384GB of DRAM. Using AWS as an example, local instance store would not work for virtual PowerStore as the volumes are ethereal. This implies an instance type like i4i.32xlarge, which is limited to eight NVMe drives (not shareable) or falling back to using EBS storage. Two i4i.x32large instances would cost just under $16,000 a month, or around $190,000 per annum, excluding data transfer costs, EBS storage or PowerStore licensing charges.
So, we are not saying PowerStore couldn’t be ported to the public cloud, but the architectural challenges are significant and would require a lot of redesign and rewrites. We know, for example, that NetApp invested a lot of work getting FSx for NetApp ONTAP working, with experience running in the cloud from before 2017. Remember that ONTAP is as close to SDS as you can imagine with mainstream storage software, so none of this is easy.
PowerMax has the same hardware challenges to overcome. Don’t forget that both PowerMax and PowerStore date back to architectures developed 30 years ago, when resiliency and availability features were delivered in hardware.
Of course, there are other products in the Dell storage stable. No, we’re not talking about XtremIO (which seems to have withered and died), but PowerFlex, the renamed ScaleIO. You may remember that EMC acquired ScaleIO back in 2013. EMC (or Dell) never really pushed the technology, possibly because the implementation required a client-side kernel driver to expose storage LUNs to a host (here’s the VMware implementation). That restriction was removed in PowerFlex 4.0 with the inclusion of a new component called SDT that talks to the NVMe/TCP driver (Linux).
It may come as no surprise that due to its software-defined nature, PowerFlex was ported to the AWS Marketplace in November 2022 (check out this blog by Dell employee Itzikr Reich). Follow the link in the blog that points to PowerFlex on AWS, and voila, you reach the same point as the newly announced “Dell APEX Block Storage for AWS”. So Dell has simply renamed the solution that was pushed to the cloud last November.
We could have already worked out that this new offering is PowerFlex, as the marketplace entry mentions the product everywhere. To be fair, PowerFlex makes complete sense as a cloud storage solution because it is inherently an SDS platform. PowerFlex is much more adaptable to variable instance sizes and capacities, so could be grown and expanded across many different configurations over time using its scale-out capabilities.
Reality vs Expectation
Unfortunately, many Dell customers may have been expecting cloud versions of PowerStore and PowerMax, leading to disappointment when they discover the first block storage implementation will be PowerFlex. This is a shame because the hype of Project Alpine hasn’t been met with reality. Customers will need to learn a new platform, as the majority won’t be running PowerFlex today. Additionally, the ability to move data between on-premises and the cloud will be challenging, as we’re not aware of any native replication solutions that connect the legacy platforms to PowerFlex.
What about file storage? It appears that Dell APEX File Storage for AWS is OneFS, aka PowerScale, aka Isilon. Remember that Dell first released Isilon on Google Cloud Platform back in 2018. Why has it taken five years to port to AWS? We don’t know.
The Architect’s View®
There are two interesting threads throughout this discussion that are worth following. First that PowerFlex/ScaleIO seems to have a new lease of life, with additional features, including file services. What always appeared to be one of the least exploited storage solutions within Dell could now be getting more focus due to the software-defined benefits it offers. This is definitely a good strategy, in our view.
Second, porting PowerStore and PowerMax to the public cloud may take much longer than planned or maybe never happen at all (which is what we think will definitely happen with PowerMax). These two solutions are so entrenched in legacy on-premises that the work involved may be misspent. Instead, some capability to replicate natively between PowerStore/PowerMax and PowerFlex in the cloud would be more appropriate.
Finally, we hope that Dell will provide more details on both block and file solutions, including:
- Pricing – the AWS marketplace pricing entries are placeholders and don’t reflect the PowerFlex and PowerScale licence costs. Customers can’t get an idea of costs without contacting sales.
- Reference architectures – in both instances, customers will want and need example reference solutions that pick the best and optimised instance types for both products.
- Operational tools – everything from cluster resizing to automated deployments.
It still feels like there’s a gap between the ease of use of the public cloud and the Dell implementations of these new solutions. As a comparison, NetApp has had FSx for ONTAP as a native solution within AWS for nearly two years, with all the ease of use that entails.
It’s also worth noting NetApp has had cloud ONTAP for at least seven years, while Pure Storage first released Cloud Block Store to general availability nearly four years ago. Dell Technologies is still way behind the market leaders, and that’s without mentioning new solutions like VAST Data, WEKA, Volumez, and Lightbits, all of which are already software-defined and in the public cloud or cloud-capable.
Copyright (c) 2007-2023 – Post #2e45 – Brookend Ltd, first published on https://www.architecting.it/blog, do not reproduce without permission.