Category: Storage

Posted on

20x Faster Time to First Token: The HPE Alletra X10000 Edge for AI

In March 2026, HPE, NVIDIA, Kamiwaza and Signal65 published a paper (check here and here) showing  about a 20x acceleration for both time-to-first-token (TTFT) and token generation rate, using an HPE Alletra X10000 to store the KV cache. The storage system used S3 over RDMA to achieve this (and is in fact the first NVIDIA-certified object storage system).

The test has, as of the date of this writing, the most complete benchmark disclosure of all the KV Cache competitive tests I could find. It also consists of an extremely hard workload with high concurrency.

I will explain why all this is important, why it’s different from competitor numbers, plus provide some insights about what this means regarding overall system efficiency.

Because the goal isn’t just to keep GPUs busy. It’s to keep them busy generating new stuff, not recalculating old stuff.

The benefits with this solution are numerous:

  • Far more workload becomes possible but also…
  • One could approach it as a lot less infrastructure is needed due to far more efficient use of the hardware, which means…
  • Lower power and rackspace requirements, which all leads to…
  • Lower Watt/token and lower $/token.
Continue reading “20x Faster Time to First Token: The HPE Alletra X10000 Edge for AI”
Posted on

HPE X10000 Deep Dive – Differentiation For Unstructured Data

At HPE Discover Barcelona 2024, HPE released the Alletra Storage MP X10000, the latest in our new line of shared hardware platform storage offerings.

It’s an innovative new platform specially made for unstructured data, and a long time in the making. This is HPE tech, not a partnership.

The initial workloads this solution is aimed at are anything requiring fast S3 performance, including AI workloads, data lakes, cloud native app development and high speed restore and backup.

It has several innovations such as RDMA for object, and is highly differentiated – plus, allows this kind of technology in a smaller possible starting capacity instead of only focusing on the huge side of the scale.

As usual, my aim is not to regurgitate basic information but rather to explain the true technical differentiation and get people excited about the possibilities on offer here. 

The summary of the X10000 benefits are:

  1. Disaggregation flexibility for separately expanding compute and/or capacity
  2. Ability to scale down and not need huge capacities to get good performance
  3. Balanced read/write performance and low latency for all workloads
  4. Flexible, fully container-based architecture that opens up tons of possibilities for running customer code inside the storage solution.

Let’s get to it:

Continue reading “HPE X10000 Deep Dive – Differentiation For Unstructured Data”
Posted on

The Architectural Benefits of HPE Alletra MP – Plus R4 Coolness

When we first released the new HPE Alletra MP platforms, I wrote a few articles going over the benefits and how the flexible new hardware platform manifests into different “personalities” for high end block and file solutions.

This time I want to take a deeper dive into the architectural benefits of our approach and how the new R4 software for Alletra MP Block enables certain things no other vendor can come close to – plus give a taste of what may be possible in the future given the amazing flexibility of the underlying architecture (it’s a blog, I can’t provide roadmaps here).

I will cover things like fractional multi-dimensional scaling (that allows things impossible with other vendors like adding a single controller node without needing to add capacity) but also resiliency in the face of simultaneous failures that would cripple all other storage systems I’m aware of. It’s not meant to be a comprehensive coverage of everything, but hopefully enough to give you a taste.

Let’s go!

Continue reading “The Architectural Benefits of HPE Alletra MP – Plus R4 Coolness”
Posted on

Modern RAID Must Protect Against Multiple Temporally Correlated Errors

Modern data protection needs to adapt to protecting modern media. RAID is no exception. In this article I will explain why modern storage consumers need to be asking for certain kinds of protection and not settling for less.

To summarize, don’t bother with storage that can’t provide at least dual parity protection for any given piece of data (whether that’s an array, HCI or the cloud, it doesn’t matter).

Why? Two big reasons:

  1. Because media these days is both larger and fails differently than in the past. Which means Temporally Correlated Errors are far more likely to happen, so you need protection against those. It’s not doom-mongering. It’s based on data.
  2. In the olden days, arrays had small RAID groups that each held a handful of volumes. If something was damaged in a RAID group, at most you’d just lose that handful of volumes. Modern arrays use pools of space, typically made up of multiple RAID groups. This means that you can potentially damage all volumes in an array merely by losing data integrity in a single RAID group in the pool. I’m sure you aren’t exactly looking forward to experiencing that.

I will take you step by step through this, as is my idiom. It is though rather sad that I have to write this kind of thing in 2020…

Continue reading “Modern RAID Must Protect Against Multiple Temporally Correlated Errors”

Posted on

When Terrified Vendors Attack: The Dell PowerStore Edition

Dell is at it again. This time, they paid Principled Technologies to do some tests in order to produce a ridiculous report trying to compare the high-end HPE Primera to the midrange Dell EMC PowerStore.

I’ll expose some of the more egregious errors in their methodology and overall thinking, but first I want to direct readers to an easy way to impartially compare for themselves, without having to read a FUD document sponsored by anyone at all.

Executive Summary: A Primera 670 is multiple times faster than a PowerStore 9000T, has stronger data protection, and much higher uptime.

Continue reading “When Terrified Vendors Attack: The Dell PowerStore Edition”

Posted on

The Harsh Realities of PCIe Lane Shortage in Storage Systems

There are a lot of myths and misinformation, plus more than a modicum of misunderstanding, regarding how storage systems can use available bandwidth, especially with certain newer kinds of media.

I wanted to explain some of the harsh facts of storage system design in the real world, and why one shouldn’t just add up drive speeds to estimate performance.

Continue reading “The Harsh Realities of PCIe Lane Shortage in Storage Systems”

Posted on

HCI Failure Modes and Maintenance

I got the idea for this blog after speaking with multiple customers that were contemplating switching to certain kinds of grid computing/storage (like HCI) without fully understanding the ramifications of doing so.

You see, they were (rightly so) enamored by concepts such as automation, ease of consumption and scaling. But they forgot to ask some very important questions. See here for the dangers of getting too carried away with something new and taking things for granted.

This isn’t a post claiming HCI and grid-type storage constructs are bad. Like any tool, they can be used in various ways, some of them aggressively ill-advised. The point of this post is to help customers ask for the right configuration so they don’t get stuck with a sub-optimal and risky design.

I tried to make this post as short as possible but as someone once said, “Everything should be made as simple as possible, but not simpler”. Which, ironically, is a simplification of what Einstein actually said 🙂

Continue reading “HCI Failure Modes and Maintenance”

Posted on

The Loss of Important Knowledge and Acumen Through Perceived Commoditization

I posit that we now have a whole new class of consumer that is completely oblivious to certain hitherto fundamental concepts – and this can lead to poor business decisions and overall sub-optimal execution and results.

I got the idea after a discussion with an ex colleague (that’s now working for a cloud vendor) where he proudly proclaimed that infrastructure is unimportant and uninteresting.

I’ll start generically and shift to IT. The generic aspect of this problem is very interesting, since it’s lowering quality in all sorts of fields.

And never forget: Just because something is widely and easily available doesn’t mean it’s better. It simply means that more people have access to it.

Continue reading “The Loss of Important Knowledge and Acumen Through Perceived Commoditization”

Posted on

HPE Memory-Driven Architectures Extend to 3PAR and Nimble Storage

HPE has been innovating in the Memory-Driven Compute space for a while now (for example, HPE Labs’ The Machine project and Gen-Z ).

The driver behind this has been to transform application performance, not by increments but by leaps and bounds. Think orders of magnitude in reduction of execution time. For instance, at an organization performing Alzheimer’s cure research, they had a certain key analytics operation that took 22 minutes for each iteration (and they need to do many, many iterations). With a Memory-Driven system from HPE it now takes 13 seconds. This allows the researchers to reach useful results much faster – which, in turn, means the cure could materialize in a much shorter timeframe.

Continue reading “HPE Memory-Driven Architectures Extend to 3PAR and Nimble Storage”

Posted on

Media is Not Created Equal and NVMe is Just a Protocol

In this era of over-marketing and misinformation, it can be refreshing to clarify things for customers.

Allow me to be refreshing regarding NVMe 🙂

NVMe is simply a protocol. Just like SCSI is a protocol. NVMe is most assuredly not a media type. Yet, storage vendors keep talking about “NVMe drives” and customers often think those devices are equal as long as “NVMe” is mentioned.

Alas, that’s not how things work…

Strictly speaking, there’s no such thing as an NVMe drive. Or, at the very least, calling something an “NVMe drive” isn’t enough to describe what that media is, and it’s especially not enough to describe how fast it may be.

Continue reading “Media is Not Created Equal and NVMe is Just a Protocol”

Posted on

When Terrified Vendors Attack: The Dell Edition

It recently came to my attention that Dell is now advertising some kind of benchmark that shows one of their platforms can be faster than Nimble in some very specific test of their own concoction.

While I don’t doubt that’s possible (indeed, we could do it the other way around), it may be worthwhile investigating what’s prompting the attack.

I also wanted to point out the various technically fishy points of the benchmark.

Continue reading “When Terrified Vendors Attack: The Dell Edition”

Posted on

How Nimble Storage Systems do Block Folding

I got the idea for this post after seeing certain vendors claim they were the first and only with certain data reduction technologies (I’m not talking about dedupe and compression). I thought – how come Nimble never made a big deal about this? After all, what those vendors were claiming didn’t seem to be very interesting compared to how Nimble systems efficiently write data… yet those vendors were acting as if they’d cured a particularly virulent disease.

Executive Summary

Nimble systems naturally avoid any wasted space when writing. This is an inherent part of the design and not something that was added later.

Continue reading “How Nimble Storage Systems do Block Folding”

Posted on

Why it is Incorrect to use Average Block Size for Storage Performance Benchmarking

Just a quick post to address something many people either get wrong or just live with due to convenience.

In summary: Please, let’s stop using average I/O sizes to characterize storage system performance. It’s wrong and doesn’t describe how the real world works. Using an average number is as bad as using small block 100% read numbers shown in vanity benchmarks. Neither is representative of real life.

Using a single I/O size for benchmarking became a practice for a vanity benchmark and to provide a level playing field to compare multiple products.

But, ultimately, even though the goal of comparing different systems is desirable, using a single I/O size is fundamentally flawed.

Continue reading “Why it is Incorrect to use Average Block Size for Storage Performance Benchmarking”