HPE Primera – things that all should (allowed to) know

It has been announced and many time presented by HPE, even there are videos uploaded to Youtube, there is a new storage they have and it is called Primera. This week I had the honor to have my hands on one and get myself familiar with their new Tier 0 storage system.

I stated “new” because they always had Tier 0 storage system in their offering, the XP7. They will keep this and also 3PAR.

I hear you whispering that advanced 3PAR models like 9450, 20840 and 20850 are surely Tier 0 aren’t they? Partially, but when 3PAR was built log ago to use the same codebase for all models, regardless if they are all flash or hybrid. So itself it was designed in an era when flash was not that widely adopted. So in its DNA it is not a true Tier 0 architecture and don’t forget it must cover the requirements from bottom Tier 1 to almost Tier 0.

If you came just to her me saying this is a 3PAR or not, let me tell a story. If you ever had some experience or read about the EMC – now DellEMC – Symmetrix you surely remember it has been renamed three times. It bacame DMX and a while after Vmax, not called PowerMax. So this is a direct relationship, what more the same DNA but all with little upgrades by changing some parts to faster/larger components. So faster Xeon CPU, more RAM and different HBAs.

Can we Primera a relabeled 3PAR? No.

According to HPE, nope. After the last week and the knowledge I gained with it while was checking what is under the hood, I need to agree. 3PAR is and was always a good brand and HPE made a quite good deal when bought it. They have added many features and made nice sales numbers based on 3PAR, so this itself justifies it’s existence and the same time proves it’s value. Primera was build on this 3PAR inheritage. It has all features kept which made 3PAR distinguished from other vendors. Not widely used and useless features like adaptive optimization have been removed and it will never return.

Where will it fit?

Tier 0. Period. It is a good option for clients who are aiming for the highest availability, to have a system that is there for sure, so 100% availability. Clients tend to have this resiliency and availability with higher priority, while efficiency is listed only somewhere after the ease of use. This does not mean there is no hybrid model called Primera, but this is a true all flash platform and I suggest that you should go for AFA as it gives long term investment protection and since AFA devices deliver DECO, that is a must.

Primera combines 3PAR’s best features (further modernizing them) with Nimble’s ease of use and administration, incorporating the now widely used Infosight’s in depth analysis and AI/ML capabilities.

What will happen with the 3PAR line?

It will stay and will deliver the best of class Tier 1 as it has done it in the last decade.


By now you might know me, so this will not be new. Nimble is the best option in midrange area and I love it. It can be consumed easily, no need to tweak the RAID set size as it does that for you. Only thing you need is to USE the array. It’s performance elevates some Nimble models up to the 3PAR models and believe me, if a Nimble device is stated to deliver 25k IOPS, it will deliver 25k IOPS. No joke.

How the hell could I choose between Nimble/3PAR/Primera?

Important to understand the top most important thing, which is the controller setup. Nimble is active/standby while 3PAR and Primera are active/active. In a Nimble array there are two and only two controllers and one of them is active. This active controller handles all the front and back end operations. If it fails the earlier standby controller will take over the active role and this can take up 10-25 seconds. This is still below the 30 second SCSI timeout, but while this time is passed all IO will stop. This means that workload will take a loooong breathe before doing work again. If this is not acceptable, 3PAR and Primera are the options. 3PAR can have 2/4/8 controllers while Primera scales 2 to 4 controllers.

How is Primera different?

Simply put, it is different because it has been designed by HPE based on their findings in their 3PAR estate. More specifically majority of storage issues were caused by misconfiguration and incorrect use of features. They shaped Primera to use these best practices at all times and not to leave options to change these major settings. Primera uses RAID6. Not possible to change it, not even through advanced settings. It will set the RAID set size based on the installed disk count, so storage admin can no longer use RAID6 6+2 if they have 10 drives. Infosight integration is more fundamental and it will inform and warn users if there is a new important hotfix or firmware available. Deployment of these updates are required if customer wants to qualify for the 100% availability guarantee. Critical patches must be installed within 10 days and all other new versions within 30 days.

In order to ease up the deployment of these they have rewritten the code and moved it out from the kernel to user space. This itself made it available to use services and if needed to restart them individually. So if remote copy codebase needs an update, it can be restarted and not the whole controller will reboot. This helps to deliver extreme availability.

3PAR had a proprietary ASIC which had quite some tasks to do freeing up the controller’s Xeon CPU. Primera has four of this reworked ASIC – 630 modell has 1 per controller only. 3PAR had it’s control cache in the Xeon attached memory modules, data cache in the ASIC attached memory modules. This made it difficult to upgrade newer Xeon processors since keeping the inter-operation between Xeon memory controller and the ASIC’s memory controller required extensive work. Primera solved this by using Xeon attached memory for both.

Above on the right a Primera 650/670 controller can be seen – without it’s second CPU. ASIC can help the Xeon processor if needed by taking over XOR/SHA256 operations and to handle node to node communication.

If you read all above it is surely clear that this is really in depth redesign of the operations and the logic, so this could hardly be called a new 3PAR. This is justifies by the NVMe adoption as well. That is not yet there, but all vendors must change their beloved storage systems with having NMVe in their mind. A controller today can be fully utilized only by 4-8 SAS SSD. If talking about NVMe 2-3 drives are more than enough to kill the controller back-end performance. Also important to state that there is no end to end NVMe solution out there. There is no vendor who can deliver host-fabric-storage controller-backend all using NVMe only. The standard for NVMe fabric is not there. So NVMe can be consumed in storage systems but will become widely adopted once standard is clear.

If someone is loyal to one vendor, I can imagine the last storage he/she will ever buy is a Primera. With the timeless storage option in every third year customer can have a new controller set (faster and more advanced) or can trade that in to have extension. The effective capacity of Primera is also amazing, just imagine that 4U can deliver 500 terabyte, 8U can scale above 1 petabyte (2:1 DECO). Storage systems can be freely upgraded so if someone buys a 630 (in 650 4N chassis), it can be upgraded to 650 or 670 without downtime.

A Hewlett Packard Enterprise factory – Kutna Hora

Once in a lifetime experience to visit a factory like this. Since Gen 8 Proliant, all servers – Moonshot/Apollo – and all 3PAR storage systems were produced here.

After getting dressed into a white jacked and putting on the ESD on our shoes we found ourselves in the factory floor where everything was clean since rules dictate air must be really clean and without and dust or small particles. I could see how a Synergy 660 server was built from it’s metal case to the end packaging. All systems produced by this factory is thoroughly tested, actually heavily stressed for many many hours to ensure quality. If something is not perfect it has no chance to get into a box and shipped to client.