Storage Performance Considerations for AI Data Paths

The hype of Deep Learning (DL), Machine Learning (ML) and Artificial Intelligence (AI) has reached an unprecedented frenzy. Every infrastructure vendor from servers, to networking, to storage has a word to say or play about DL/ML/AI. This prompted me to explore this hyped ecosystem from a storage perspective, notably from a storage performance requirement point-of-view.

One question on my mind

There are plenty of questions on my mind. One stood out and that is related to storage performance requirements.

Reading and learning from one storage technology vendor to another, the context of everyone’s play against their competitors seems to be  “They are archaic, they are legacy. Our architecture is built from ground up, modern, NVMe-enabled“. And there are more juxtaposing, but you get the picture – “We are better, no doubt“.

Are the data patterns and behaviours of AI different? How do they affect the storage design as the data moves through the workflow, the data paths and the lifecycle of the AI ecosystem?

Continue reading

Scaling new HPC with Composable Architecture

[Disclosure: I was invited by Dell Technologies as a delegate to their Dell Technologies World 2019 Conference from Apr 29-May 1, 2019 in the Las Vegas USA. Tech Field Day Extra was an included activity as part of the Dell Technologies World. My expenses, travel, accommodation and conference fees were covered by Dell Technologies, the organizer and I was not obligated to blog or promote their technologies presented at this event. The content of this blog is of my own opinions and views]

Deep Learning, Neural Networks, Machine Learning and subsequently Artificial Intelligence (AI) are the new generation of applications and workloads to the commercial HPC systems. Different from the traditional, more scientific and engineering HPC workloads, I have written about the new dawn of supercomputing and the attractive posture of commercial HPC.

Don’t be idle

From the business perspective, the investment of HPC systems is high most of the time, and justifying it to the executives and the investors is not easy. Therefore, it is critical to keep feeding the HPC systems and significantly minimize the idle times for compute, GPUs, network and storage.

However, almost all HPC systems today are inflexible. Once assigned to a project, the resources pretty much stay with the project, even when the workload processing of the project is idle and waiting. Of course, we have to bear in mind that not all resources are fully abstracted, virtualized and software-defined whereby you can carve out pieces of the hardware and deliver a percentage of that resource. Case in point is the CPU, where you cannot assign certain clock cycles of CPU to one project and another half to the other. The technology isn’t there yet. Certain resources like GPU is going down the path of Virtual GPU, and into the realm of resource disaggregation. Eventually, all resources of the HPC systems – CPU, memory, FPGA, GPU, PCIe channels, NVMe paths, IOPS, bandwidth, burst buffers etc – should be disaggregated and pooled for disparate applications and workloads based on demands of usage, time and performance.

Hence we are beginning to see the disaggregated HPC systems resources composed and built up the meet the diverse mix and needs of HPC applications and workloads. This is even more acute when a AI project might grow cold, but the training of AL/ML/DL workloads continues to stay hot

Liqid the early leader in Composable Architecture

Continue reading

WekaIO controls their performance destiny

[Preamble: I have been invited by GestaltIT as a delegate to their Tech Field Day for Storage Field Day 18 from Feb 27-Mar 1, 2019 in the Silicon Valley USA. My expenses, travel and accommodation were covered by GestaltIT, the organizer and I was not obligated to blog or promote their technologies presented at this event. The content of this blog is of my own opinions and views]

I was first introduced to WekaIO back in Storage Field Day 15. I did not blog about them back then, but I have followed their progress quite attentively throughout 2018. 2 Storage Field Days and a year later, they were back for Storage Field Day 18 with a new CTO, Andy Watson, and several performance benchmark records.

Blowout year

2018 was a blowout year for WekaIO. They have experienced over 400% growth, placed #1 in the Virtual Institute IO-500 10-node performance challenge, and also became #1 in the SPEC SFS 2014 performance and latency benchmark. (Note: This record was broken by NetApp a few days later but at a higher cost per client)

The Virtual Institute for I/O IO-500 10-node performance challenge was particularly interesting, because it pitted WekaIO against Oak Ridge National Lab (ORNL) Summit supercomputer, and WekaIO won. Details of the challenge were listed in Blocks and Files and WekaIO Matrix Filesystem became the fastest parallel file system in the world to date.

Control, control and control

I studied WekaIO’s architecture prior to this Field Day. And I spent quite a bit of time digesting and understanding their data paths, I/O paths and control paths, in particular, the diagram below:

Starting from the top right corner of the diagram, applications on the Linux client (running Weka Client software) and it presents to the Linux client as a POSIX-compliant file system. Through the network, the Linux client interacts with the WekaIO kernel-based VFS (virtual file system) driver which coordinates the Front End (grey box in upper right corner) to the Linux client. Other client-based protocols such as NFS, SMB, S3 and HDFS are also supported. The Front End then interacts with the NIC (which can be 10/100G Ethernet, Infiniband, and NVMeoF) through SR-IOV (single root IO virtualization), bypassing the Linux kernel for maximum throughput. This is with WekaIO’s own networking stack in user space. Continue reading

Sexy HPC storage is all the rage

HPC is sexy

There is no denying it. HPC is sexy. HPC Storage is just as sexy.

Looking at the latest buzz from Super Computing Conference 2018 which happened in Dallas 2 weeks ago, the number of storage related vendors participating was staggering. Panasas,, Excelero, BeeGFS, are the ones that I know because I got friends posting their highlights. Then there are the perennial vendors like IBM, Dell, HPE, NetApp, Huawei, Supermicro, and so many more. A quick check on the SC18 website showed that there were 391 exhibitors on the floor.

And this is driven by the unrelentless demand for higher and higher performance of computing, and along with it, the demands for faster and faster storage performance. Commercialization of Artificial Intelligence (AI), Deep Learning (DL) and newer applications and workloads together with the traditional HPC workloads are driving these ever increasing requirements. However, most enterprise storage platforms were not designed to meet the demands of these new generation of applications and workloads, as many have been led to believe. Why so?

I had a couple of conversations with a few well known vendors around the topic of HPC Storage. And several responses thrown back were to put Flash and NVMe to solve the high demands of HPC storage performance. In my mind, these responses were too trivial, too irresponsible. So I wanted to write this blog to share my views on HPC storage, and not just about its performance.

The HPC lines are blurring

I picked up this video (below) a few days ago. It was insideHPC Rich Brueckner interview with Dr. Goh Eng Lim, HPE CTO and renowned HPC expert about the convergence of both traditional and commercial HPC applications and workloads.

I liked the conversation in the video because it addressed the 2 different approaches. And I welcomed Dr. Goh’s invitation to the Commercial HPC community to work with the Traditional HPC vendors to help push the envelope towards Exascale SuperComputing.

Continue reading

The Network is Still the Computer

[Preamble: I have been invited by  GestaltIT as a delegate to their TechFieldDay from Oct 17-19, 2018 in the Silicon Valley USA. My expenses, travel and accommodation are covered by GestaltIT, the organizer and I was not obligated to blog or promote their technologies presented at this event. The content of this blog is of my own opinions and views]

Sun Microsystems coined the phrase “The Network is the Computer“. It became one of the most powerful ideologies in the computing world, but over the years, many technology companies have tried to emulate and practise the mantra, but fell short.

I have never heard of Drivescale. It wasn’t in my radar until the legendary NFS guru, Brian Pawlowski joined them in April this year. Beepy, as he is known, was CTO of NetApp and later at Pure Storage, and held many technology leadership roles, including leading the development of NFSv3 and v4.

Prior to Tech Field Day 17, I was given some “homework”. Stephen Foskett, Chief Cat Herder (as he is known) of Tech Field Days and Storage Field Days, highly recommended Drivescale and asked the delegates to pick up some notes on their technology. Going through a couple of the videos, Drivescale’s message and philosophy resonated well with me. Perhaps it was their Sun Microsystems DNA? Many of the Drivescale team members were from Sun, and I was previously from Sun as well. I was drinking Sun’s Kool Aid by the bucket loads even before I graduated in 1991, and so what Drivescale preached made a lot of sense to me.Drivescale is all about Scale-Out Architecture at the webscale level, to address the massive scale of data processing. To understand deeper, we must think about “Data Locality” and “Data Mobility“. I frequently use these 2 “points of discussion” in my consulting practice in architecting and designing data center infrastructure. The gist of data locality is simple – the closer the data is to the processing, the cheaper/lightweight/efficient it gets. Moving data – the data mobility part – is expensive.

Continue reading

The Malaysian Openstack storage conundrum

The Openstack blippings on my radar have ratcheted up this year. I have been asked to put together the IaaS design several times, either with the flavours of RedHat or Ubuntu, and it’s a good thing to see the Openstack interest level going up in the Malaysian IT scene. Coming into its 8th year, Openstack has become a mature platform but in the storage projects of Openstack, my observations tell me that these storage-related projects are not as well known as we speak.

I was one of the speakers at the Openstack Malaysia 8th Summit over a month ago. I started my talk with question – “Can anyone name the 4 Openstack storage projects?“. The response from the floor was “Swift, Cinder, Ceph and … (nobody knew the 4th one)” It took me by surprise when the floor almost univocally agreed that Ceph is one of the Openstack projects but we know that Ceph isn’t one. Ceph? An Openstack storage project?

Besides Swift, Cinder, there is Glance (depending on how you look at it) and the least known .. Manila.

I have also been following on many Openstack Malaysia discussions and discussion groups for a while. That Ceph response showed the lack of awareness and knowledge of the Openstack storage projects among the Malaysian IT crowd, and it was a difficult issue to tackle. The storage conundrum continues to perplex me because many whom I have spoken to seemed to avoid talking about storage and viewing it like a dark art or some voodoo thingy.

I view storage as the cornerstone of the 3 infrastructure pillars  – compute, network and storage – of Openstack or any software-defined infrastructure stack for that matter. So it is important to get an understanding the Openstack storage projects, especially Cinder.

Cinder is the abstraction layer that gives management and control to block storage beneath it. In a nutshell, it allows Openstack VMs and applications consume block storage in a consistent and secure way, regardless of the storage infrastructure or technology beneath it. This is achieved through the cinder-volume service which is a driver most storage vendors integrate with (as shown in the diagram below).

Diagram in slides is from Mirantis found at

Diagram in slides is from Mirantis found at

Cinder-volume together with cinder-api, and cinder-scheduler, form the Block Storage Services for Openstack. There is another service, cinder-backup which integrates with Openstack Swift but in my last check, this service is not as popular as cinder-volume, which is widely supported by many storage vendors with both Fibre Channel and iSCSi implementations, and in a few vendors, with NFS and SMB as well. Continue reading

The leapfrog game in Asia with HPC

Brunei, a country rich in oil and gas, is facing a crisis. Their oil & gas reserves are rapidly running dry and expected to be depleted within 2 decades. Their deep dependency on oil and gas, once the boon of their economy, is now the bane of their future.

Since 2000, I have been in and out of Brunei and got involved in several engagements there. It is a wonderful and peaceful country with friendly people, always welcoming visitors with open hearts. The country has prospered for decades, with its vast oil riches but in the past few years, the oil prices have been curbed. The profits of oil and gas no longer justify the costs of exploration and production.

2 years ago, I started pitching a new economy generator for the IT partners in Brunei. One that I believe will give a country like Brunei the ability to leapfrog their neighbours in South East Asia, which is to start build a High Performance Computing (HPC)-as-a-Service (HPC-as-a-Service) type of business.

Why HPC? Why do I think HPC will give a developing country like Brunei super powers in the digital economy?

Continue reading

Storage dinosaurs evolving too

[Preamble: I am a delegate of Storage Field Day 15 from Mar 7-9, 2018. My expenses, travel and accommodation are paid for by GestaltIT, the organizer and I am not obligated to blog or promote the technologies presented at this event. The content of this blog is of my own opinions and views]

I have been called a dinosaur. We storage networking professionals and storage technologists have been called dinosaurs. It wasn’t offensive or anything like that and I knew it was coming because the writing was on the wall, … or is it?

The cloud and the breakneck pace of all the technologies that came along have made us, the storage networking professionals, look like relics. The storage guys have been pigeonholed into a sunset segment of the IT industry. SAN and NAS, according to the non-practitioners, were no longer relevant. And cloud has clout (pun intended) us out of the park.

I don’t see us that way. I see that the Storage Dinosaurs are evolving as well, and our storage foundational knowledge and experience are more relevant that ever. And the greatest assets that we, the storage networking professionals, have is our deep understanding of data.

A little over a year ago, I changed the term Storage in my universe to Data Services Platform, and here was the blog I wrote. I blogged again just before the year 2018 began.


Continue reading

The rise of RDMA

I have known of RDMA (Remote Direct Memory Access) for quite some time, but never in depth. But since my contract work ended last week, and I have some time off to do some personal development, I decided to look deeper into RDMA. Why RDMA?

In the past 1 year or so, RDMA has been appearing in my radar very frequently, and rightly so. The speedy development and adoption of NVMe (Non-Volatile Memory Express) have pushed All Flash Arrays into the next level. This pushes the I/O and the throughput performance bottlenecks away from the NVMe storage medium into the legacy world of SCSI.

Most network storage interfaces and protocols like SAS, SATA, iSCSI, Fibre Channel today still carry SCSI loads and would have to translate between NVMe and SCSI. NVMe-to-SCSI bridges have to be present to facilitate the translation.

In the slide below, shared at the Flash Memory Summit, there were numerous red boxes which laid out the SCSI connections and interfaces where SCSI-to-NVMe translation (and vice versa) would be required.

Continue reading