115-GreyBeards talk database acceleration with Moshe Twitto, CTO&Co-founder, Pliops

We seem to be on a computational tangent this year. So we thought it best to talk with Moshe Twitto, CTO and Co-Founder at Pliops (@pliopsltd). We had first seen them at SFD21 (see videos of their sessions here) and their talk on how they could speed up database IO was pretty impressive. Essentially, they have a database/storage accelerator board used to increase block store IO activity to NVMe SSDs but also provide a key-value store IO accelerator,

Moshe was very knowledgeable about the technology and had previously worked at Samsung for their SSD group. He knew a lot about what happens underneath the covers of an SSD and what it takes to speed up IO. It turns out that many in memory databases use persistent key value stores to persist data or to operate in non- (or partial-) memory-mode. Listen to the podcast to learn more.

The Pliops board plugs into the PCIe bus and accelerates IO to NVMe SSDs connected to the bus or can act to accelerate IO to JBoF that’s networked behind it. Their board uses FPGA(s), NVDimms of their own design and DRAM to accelerate database IO using NVMe SSDS.

Pliops operates in one of two modes, as a Key-Value store or as a Block store. Their Key-Value store takes advantage of block store capabilities, so we start there.

In block mode, Pliops provides inline hardware data compression and encryption. Compression requires support for variable length blocks on backend SSDs. To better support this, they pack multiple compressed blocks into physical blocks. They also use a virtualization service to support mapping host LBAs to physical block addresses (using an internal key-value store). Hardware, inline encryption is also provided on a LUN (or namespace) basis. This could enable each database to have its own key. They have a root-of-trust secret key used to encrypt customer namespace (database) keys.

They also optimize physical block layouton the SSD to reduce write amplification (doing more than one write to the NAND for every host write to the SSD).

Block mode also supports smart caching. This is especially useful for database journaling/loging which reuses a portion of LBA address space (blocks} as a revolving journal/log. These blocks are overwritten with new data often and data written to them need not be destaged to NVMe SSDs as long as it can be maintained in NVDimm storage. At some point it gets destaged but probably only when log activity slows down (if ever) or some timeout occurs.

For their key-value storage accelerator, they have implemented an API that’s similar to RocksDB, a persistent key-value store, which is used as a physical storage backend for Reddis and similar in-memory databases. However, the challenge with RocksDB is that there are lots of tuning knobs/parameters. So getting right takes some work. But all this can be avoided just by using Pliops.

We didn’t talk too much about how their key-value store works. Moshe says they optimize the key structures and key data so that all database keys can be retained in their board’s memory and just by doing that, they can have immediate (1 IO) access to any data block pointed to by those keys.

He did mention that they provide ~the same performance for a database getting 10-25% host cache hit rates using their board as that same database would support with a 80-90% host cache hit rate not using their board. Some of this was shown at SFD21 (so check out the videos above for more performance info)

A couple of other advantages they bring to the table. As they are interposed between the host and the NVMe SSDs they can take advantage of their NVDIMMs and memory to write much wider stripes than the host writes. This allows them to reduce SSD read and write amplification (due to less garbage collection) by writing more full NAND pages. All this also reduces physical host (data) writes/day which can significantly improve SSD endurance.

Somewhere in all that smart caching and data compression, they are able to also decrease response times It turns out that databases that don’t use RocksDB or depend on key-value stores can easily take advantage of all their block store functionality to improve IO performance.

They mostly market their product to hyperscalers and superscalers. His definition of super-scalers was any organization that operates at public cloud levels but is not a public cloud (e.g., big social media companies).

Moshe Twitto, CTO & Co-founder Pliops

Moshe is an expert in advanced data management and coding algorithms. Prior to co-founding Pliops, Moshe served as CTO of Samsung’s SSD Controller Development Center in Israel.

Moshe holds MSEE, BSEE degrees from Technion University, Summa Cum Laude and served in the Unit 8200 Intelligence Division of the Israel Defense Corps.

114: GreyBeards talk computational storage with Tong Zhang, Co-Founder & Chief Scientist, ScaleFlux

Seeing as how one topic on last years FMS2020 wrap-up with Jim Handy was the rise of computational storage and it’s been a long time (see GreyBeards talk with Scott Shadley at NGD Systems) since we discussed this, we thought it time to check in on the technology. So we reached out to Dr. Tong Zhang, Chief Scientist and Co-founder, ScaleFlux to see what’s going on. ScaleFlux is seeing rising adoption of their product in hyper-scalers as well as large enterprises. Their computational storage is a programmable FPGA based 4TB and 8TB SSD.

Tong was very knowledgeable on current industry trends (Moore’s law slowing & others) that have created an opening for computational storage and other outboard compute. He also is well versed into how some of the worlds biggest customers are using the technology to work faster and cheaper in their data centers. Listen to the podcast to learn more.

At the start Tong mentioned Alibaba’s use of ScaleFlux’s transparent, line speed, outboard encryption/decryption and compression/decompression. And, depending on the data, they can see compression ratios far exceeding 2:1. As such, customers not only benefit from a cheaper $/GB but can also see better NAND endurance and higher performance.

Hosts can do compression and encryption but doing so takes a lot of CPU cycles. It turns out that compression is more compute intensive than encryption. Tong said that most modern cores can encrypt/decrypt at 1GB/sec but, depending on the compression algorithm, can only compress at 40 to 100MB/sec. But in any case doing so on the host consumes a lot of CPU instruction cycles. With ScaleFlux, they can compress and decompress at PCIe bus speeds.

Most storage controllers that offer compression/decompression must have some sort of LBA (logical block address) virtualization. Because while the host may be writing 512 or 4096 byte blocks, what’s actually written to the NAND is more like, 231 or 1999 bytes. So packing these odd, variable length blocks into NAND blocks can become a problem. But most SSDs already have a flash translation layer (FTL) where LBA addresses are mapped, over time, to different physical NAND page/block addresses. ScaleFlux has combined support for LBA virtualization and FTL into the same process and by doing so, they reduce IO overhead to perform better.

ScaleFlux’s drive is an NVMe SSD, which already supports great native response times but when you are transferring 1/2 or less of (compressed) data from the host onto NAND, you can reduce latencies even more. .

Although their current generation product is based on TLC NAND they are working on the next generation which will support QLC. And the benefits of writing and reading less data should also help QLC endurance and performance.

Although ScaleFlux is seeing great adoption with just outboard transparent compression and encryption, there is more that could be done, For example,

  • Filtering query’s at the drive rather than at the host. If customers can send a search key/phrase or other filtering request directly to the drive, the drive can pass over all it’s data and send back just the data that matches that filter request.
  • Transcoding and other data format changes. Although transcoding makes a lot of sense to do outboard, Tong also mentioned format changes. We asked him to clarify and he said consider a row based database that needs to be accessed in columnar format. If the drive could change the format from one to the other, it opens up more analytics tool sets.

At the moment, ScaleFlux engineering teams are the ones that program the FPGA to perform outboard functionality. But in a future release, they plan to adding ARM cores in a SoC, which can handle more general purpose outboard functionality as code.

Because of this added complexity of compression, encryption and other outboard logic, we asked Tong what power loss protection was available at the drive level. Tong assured us that once data has been received by their device, it is maintained across a power failure with CAPs and other logic to offload it.

Tong also mentioned that Intel, AWS and the NVMe standard committee are looking at adding some computational storage support into the NVMe standard, so applications and host software can invoke and maybe modify outboard functionality on the fly. Sort of like loading containers of functionality to run on the fly on an SSD drive.

Dr. Tong Zhang, Chief Scientist and Co-fonder, ScaleFlux

Dr. Tong Zhang is a well-established researcher with significant contributions to data storage systems and VLSI signal processing. Dr. Zhang is responsible for developing key techniques and algorithms for ScaleFlux’s Computational Storage products and exploring their use in mainstream application domains.

He is currently a Professor at Rensselaer Polytechnic Institute (RPI). His current and past research span over database, filesystem, solid-state and magnetic data storage devices and systems, digital signal processing and communication, error correction coding, VLSI architectures, and computer architecture.

He has published over 150 technical papers at prestigious USENIX/IEEE/ACM conferences and journals with the citation h-index of 36, and has served as general and technical program chairs for several premier conferences. Among his many research accomplishments, he made pioneering contributions to establishing flash memory signal processing and enabling practical implementation of low-density parity-check (LDPC) codecs. He received two best paper awards and has over 20 issued/pending US patent applications.

He holds BS/MS degrees in EE from the Xi’an Jiaotong University, China, and PhD degree in ECE from the University of Minnesota.

112: GreyBeards annual year end wrap-up with Keith & Matt

It’s the end of the year, so time for our regular year end wrap up discussion with the GreyBeards. 2020 has been an interesting year to say the least. It started out just fine, then COVID19 showed up and threw a wrench in everyone’s plans and as the year closes, we were just starting to see some semblance of the new normal, when one of the largest security breaches in years shows up. Whew, almost glad that’s over and onto 2021.

As always the GreyBeards had a great discussion on these and other topics to highlight the year just past. The talk was wide ranging and hard to characterize but I did my best below. Listen to the podcast to learn more.

COVID19s impact on the enterprise

It will probably take some time before we learn the true, long term impacts of COVID19 on IT but one major change has to be the massive Work From Home (WFH) transition that took place overnight.

While WFH can be more productive for some, the lack of face2face interaction can be challenging for others. The fact that many of the GreyBeards have been working from home for decades now, left us a bit oblivious to how jarring this transition can be for newcomers.

There’s definitely some psychological changes that need to occur to be productive at WFH. Organization skills become even more important. Structured interactions (read conference calls, zoom/webex and other forms of communication become much more important. And then there’s security.

Turns out VMware and others have been touting VDI solutions for the past decade or so to better support remote work and at the same time providing corporate levels of security for remote work. While occasionally this doesn’t work quite as well as expected, it’s certainly much much better than having end users access corporate data without any security around that data or worse yet, the “bring your own device”. All these VDI solutions had a field day when WFH happened.

Many workers found they could be more productive at WFH, due the less distractions, no commute time and more flexible hours. What happens when COVID19 is vanquished to all these current WFHers is anyone’s guess.

We thought there might be less need for large office campuses/buildings. But there’s something to be said for more collaboration and random interactions through face2face meetings that can only occur in an office setting with workers present at the same time. Some organizations will take to this new way of work while others will try to dial WFH back to non-existent. Where your organization fits on this spectrum and why, will be telling across a number of dimensions.

The rise of ARM

There’s been a slow but steady improvement in ARM processors over the last almost half century. Nowadays it’s starting to make a place for itself in the enterprise. ARH has always been the goto microprocessor for low power solutions (like smartphones) but nowadays they are being deployed in the cloud and even the enterprise. These can be used as server processors but even outside servers, ARM cores are showing up in hardware accelerators as the brains behind SmartNICs, DPUs, SPUs, etc.

Keith made mention AWS 2nd generation Graviton 64-bit ARM processor EC2 instances. And yes there’s significant cost ( & power) savings that can be had using AWS Graviton ARM instances. So the cloud is starting to adopt them. Somewhere over the past couple of years I heard that VMware was porting ESX to work on ARM cores.

But apparently, it’s not just as simple as dropping an ARM multi-core processor into a server and recompiling your code and away you go. Applications need a certain amount of optimization to run effectively on ARM processors. And the speed up between non-optimized and optimized versions of an application running on ARM cores is significant.

As for SmartNICs and DPUs, these are data networking hardware accelerators that provide real time processing capabilities needed to keep up with higher speed networking, 100GbE and beyond. These DPUs perform deep packet inspection, data compression, encryption and other services all at wire speeds.. Yes you could devote 1 or more X86 cores to do this, but it’s much cheaper (and more effective) to do this outside the CPU core. Moreover, performing this activity at the network entry point to the server means that much of this data doesn’t have to be transferred back and forth through server memory. So not only does it save CPU core cycles but also memory size and memory & PCIe bus bandwidth. We published a recent podcast with Kevin Deierling, NVIDIA Networking discussing DPUs if you want to learn more.

Pat made mention at (virtual) VMworld their plans to port ESX to the DPU. Keith followed up on this and asked some other exec’s at VMware about this and they said VMware will more likely support DPUs as just another hardware accelerator in their cluster. In either case, CPU cycles should be freed up and this should help VMware use X86 cores more efficiently. And perhaps this will help them engage in more CPU constrained environments such as Telcom.

Then there’s computational storage. We have been watching this technology for a couple of years now and it’s seeing some success in being deployed to public cloud environments. They seem to be being used to provide outboard data compression. It’s unclear whether these systems depend on ARM processing or not but my bet is that they do. To learn more about computational storage check out these podcasts, FMS2020 wrap up with Jim Handy and our talk with Scott Shadley on NGD’s computational storage.

System security

At yearend, we are learning of a massive security breach throughout US government IT facilities. All based on what is believed to be a Russian hack to a software package that is embedded in a popular networking tool software solution, SolarWinds. They are calling this a software supply chain hack. Although we are mainly hearing about government agencies being hacked, SolarWinds is also pervasive in the enterprise as well.

There have been many hardware supply chain hacks in the past, where a board supplier used chips or logic that weren’t properly vetted. Over time, hardware suppliers have started to scrutinize their supply chains better and have reduced this risk.

And the US government have been lobbying for the industry to use a security chip with a backdoor or to supply back doors to smartphone encryption capabilities. Luckily, so far, none of these have been implemented by industry.

What Russia has shown us is that this particular hack is not limited to the hardware sphere. Software supply chain risk can’t be ignored anymore.

This means that any software application supplier will need to secure their supply chain or bring it all in house. Which may mean that costs for these packages will go up. It’s possible that using a pure open source supply chain may reduce this risk as well. At least that’s the promise of open source.

We said 2020 was an interesting year and it’s going out with a bang.

Matt Leib (@MBLeib), one of our co-hosts, has been blogging in the storage space for over 10 years, with work experience both on the engineering and presales/product marketing.. His blog is at Virtually Tied to My Desktop and he’s on LinkedIN.

Keith Townsend (@CTOAdvisor) is a IT thought leader who has written articles for many industry publications, interviewed many industry heavyweights, worked with Silicon Valley startups, and engineered cloud infrastructure for large government organizations. Keith is the co-founder of The CTO Advisor, blogs at Virtualized Geek, and can be found on LinkedIN.

111: GreyBeards talk data analytics with Matthew Tyrer, Sr. Mgr. Solutions Mkt & Competitive Intelligence, Commvault

Sponsored by:

I’ve known Matthew Tyrer, Senior Manager Solutions Marketing and Competitive Intelligence, Commvault for quite awhile now and he’s always been knowledgeable about the problems the enterprise has in supporting and backing up large file data repositories. But lately he’s been focused on Commvault Activate their data analytics solution.

We had a great talk with Matthew. He was easy to talk to and knew a lot about how data analytics can ease the operational burden of the enterprise growing file data environments. .Remind me not to have two Matthew’s on the same program ever again. Listen to the podcast to learn more.

Matthew mentioned that their Activate was built on the Commvault platform software stack, which has had a rich and long history of development and customer deployments. It seems that Activate data analytics had been an early part of the platform but recently was split out as a separate solution.

One capability that Activate has that many other data analytics solutions do not, is the ability to examine both online data as well as data in backups. Most analytics solution can do one or the other, only a few do both. But if a solution only has access to online or backup data, they are missing half the story.

In addition, Activate can operate across multiple data centers as well as across multiple public cloud environments to provide analytics for an enterprise’s file data where it may reside.

Given the proliferation of file data these days, data analytics has become a necessity to most large IT shops. In the past, an admin could track some data over time but with the volumes of file data today, this is no longer tenable. At PB or more of file data, located in on prem data centers as well as across multiple clouds, there’s just too much file data to keep track of manually anymore.

Activate also indexes file content to provide more visibility and tracking of the different types of data under management in the enterprise. This is in addition to the extensive metadata that is collected and analyzed so it can better understand data access rights, copies and physical locations around the enterprise.

Activate can help organizations govern their data flows in support of industry as well as government data compliance requirements. Activate Data Governance, one of the three Activate solutions, is focused exclusively on providing enterprises the tools needed to manage any and all data that exists under compliance regulation environments.

Mat Leib had worked in eDiscovery before and it had always been a pain to extract “legally relevant” data from online and backup repositories. With the Activate eDiscovery solution and Activate’s content indexing of all file data, legal can perform their own relevant data searches to create eDiscovery data sets in support of litigation activities. Self service legal extracts like this vastly reduces the admin time and cost needed for eDiscovery.

The Activate File Space Optimization solution was deployed in one environment that had ~20PB of data online. By using File Space Optimization, the customer was able to cut 20PB down to 10PB. Any customer could benefit from such a reduction but customers doing data migration would see even more benefit.

At the end of the podcast, Matthew mentioned some videos that show Activate solution use cases.

Matthew Tyrer, Senior Solutions Marketing and Competitive Intelligence

Having worked at Commvault for over twelve years, after 8 years as a Sales Engineer Matt took that technical knowledge and transitioned to marketing where he is currently serving as a Senior Manager in Commvault’s Solution Marketing team. He is also heavily involved in Competitive Intelligence initiatives, and actively participates in field enablement programs.

He brings over 20 years’ experience in the IT industry, including within the fields of data and information management, cloud, data governance, enterprise storage, disaster recovery, and ultimately both implementing and supporting those projects and endeavours for public and private sector clients across Canada and around the globe.

Matt’s passion, deep product knowledge, and broad field experiences have enabled him to translate Commvault technology and vision such that their value is easily understood in the market and amongst client and partner families.

A self-described geek-dad, Matt is an avid boardgame enthusiast, firmly believes that Han shot first, and enjoys tormenting his girls with bad dad jokes.

109: GreyBeards talk SmartNICs & DPUs with Kevin Deierling, Head of Marketing at NVIDIA Networking

We decided to take a short break (of sorts) from storage to talk about something equally important to the enterprise, networking. At (virtual) VMworld a month or so ago, Pat made mention of developing support for SmartNIC-DPUs and even porting vSphere to run on top of a DPU. So we thought it best to go to the source of this technology and talk with Kevin Deierling (TechSeerKD), Head of Marketing at NVIDIA Networking who are the ones supplying these SmartNICs to VMware and others in the industry.

Kevin is always a pleasure to talk with and comes with a wealth of expertise and understanding of the technology underlying data centers today. The GreyBeards found our discussion to be very educational on what a SmartNIC or DPU can do and why VMware and others would be driving to rapidly adopt the technology. Listen to the podcast to learn more.

NVIDIA’s recent acquisition of Mellanox brought them Mellanox’s NIC, switch and router technology. And while Mellanox, and now NVIDIA have some pretty impressive switches and routers, what interested the GreyBeards was their SmartNIC technology.

Essentially, SmartNICS provide acceleration and offload of data handling needs required to move data around an enterprise network. These offload services include at a minimum, encryption/decryption, packet pacing (delivering gadzillion video streams at the right speed to insure proper playback by all), compression, firewalls, NVMeoF/RoCE, TCP/IP, GPU direct storage (GDS) transfers, VLAN micro-segmentation, scaling, and anything else that requires real time processing to perform at line speeds.

For those who haven’t heard of it, GDS transfers data from storage directly into GPU memory and from GPU memory directly to storage without any CPU cycles or server memory involvement, other than to set up the transfer. This extends NVMeoF RDMA tech to/from storage and server memory, to GPUs. That is, GDS offers a RDMA like path between storage and GPU memory. GPU to/from server memory direct interface already exists over the PCIe bus.

But even with all the offloads and accelerators above, they can also offer an additional a secure enclave outside the TPM in the CPU, to better isolate security sensitive functionality for a data center. (See DPU below).

Kevin mentioned multiple times that the new unit of computation is no longer a server but rather is now a data center. When you have public cloud, private cloud and other systems that all serve up virtual CPUs, NICs, GPUs and storage, what’s really being supplied to a user is a virtual data center. Cloud providers can carve up their hardware and serve it to you any way you want or need it. Virtual data centers can provide a multitude of VMs and any infrastructure that customers need to use to run their workloads.

Kevin mentioned by using SmartNics, IT or cloud providers can return 30% of the processor cycles (that were being spent doing networking work on CPUs) back to workloads that run on CPUs. Any data center can effectively obtain 30% more CPU cycles and increased networking speed and performance just by deploying SmartNICs throughout all the servers in their environment.

SmartNICs are an outgrowth of Mellanox technology embedded in their HPC InfiniBAND and high end Ethernet switches/routers. Mellanox had been well known for their support of NVMeoF/RoCE to supply high IOPs/low-latency IO activity for NVMe storage over Ethernet and before that their InfiniBAND RDMA technologies.

As Mellanox came out with their 2nd Gen SmartNIC they began to call their solution a “DPU” (data processing unit), which they see forming part of a “holy trinity” underpinning the new data center which has CPUs, GPUs and now DPUs. But a DPU is more than just a SmartNIC.

All NVIDIA SmartNICs and DPUs are based on Mellanox’s BlueField cards and chip technology. Their DPU uses BlueField2 (gen 2 technology) chips, which has a multi-core ARM engine inside of it and memory which can be used to perform computational processing in addition to the onboard offload/acceleration capabilities.

Besides adding VMware support for SmartNICs, PatG also mentioned that they were porting vSphere (ESX) to run on top of NVIDIA Networking DPUs. This would move the core VMware’s hypervisor functionality from running on CPUs, to running on DPUs. This of course would free up most if not all VMware Hypervisor CPU cycles for use by customer workloads.

During our discussion with Kevin, we talked a lot about the coming of AI-ML-DL workloads, which will require ever more bandwidth, ever lower latencies and ever more compute power. NVIDIA was a significant early enabler of the AI-ML-DL with their CUDA API that allowed a GPU to be used to perform DL network training and inferencing. As such, CUDA became an industry wide phenomenon allowing industry wide GPUs to be used as DL compute engines.

NVIDIA plans to do the same with their SmartNICs and DPUs. NVIDIA Networking is releasing the DOCA (Data center On a Chip Architecture) SDK and API. DOCA provides the API to use the BlueField2 chips and cards which are the central techonology behind their DPU. They have also announced a roadmap to continue enhancing DOCA, as they have done with CUDA, over the foreseeable future, to add more bandwidth, speed and functionality to DPUs.

It turns out the real problem which forced Mellanox and now NVIDIA to create SmartNics was the need to support the extremely low latencies required for NVMeoF and GDS IO.

It wasn’t clear that the public cloud providers were using SmartNICS but Kevin said it’s been sort of a widely known secret that they have been using the tech. The public clouds (AWS, Azure, Alibaba) have been deploying SmartNICS in their environments for some time now. Always on the lookout for any technology that frees up compute resources to be deployed for cloud users, it appears that public cloud providers were early adopters of SmartNICS.

Kevin Deierling, Head of Marketing NVIDIA Networking

Kevin is an entrepreneur, innovator, and technology executive with a proven track record of creating profitable businesses in highly competitive markets.

Kevin has been a founder or senior executive at five startups that have achieved positive outcomes (3 IPOs, 2 acquisitions). Combining both technical and business expertise, he has variously served as the chief officer of technology, architecture, and marketing of these companies where he led the development of strategy and products across a broad range of disciplines including: networking, security, cloud, Big Data, machine learning, virtualization, storage, smart energy, bio-sensors, and DNA sequencing.


Kevin has over 25 patents in the fields of networking, wireless, security, error correction, video compression, smart energy, bio-electronics, and DNA sequencing technologies.

When not driving new technology, he finds time for fly-fishing, cycling, bee keeping, & organic farming.

This image has an empty alt attribute; its file name is Subscribe_on_iTunes_Badge_US-UK_110x40_0824.png
This image has an empty alt attribute; its file name is play_prism_hlock_2x-300x64.png
This image has an empty alt attribute; its file name is Spotify_Logo_CMYK_Black-1024x307.png