He who shall not be named sent me this thumbnail today. At any rate, the cat was already let out of the bag last month by Dave Raffo and several folks on Twitter but it looks like the ink is dry and there’s no turning back on VF Cache as the official name for Project Lightning.
What is Known About VF Cache?
At EMC World in May of 2011, both EMC Chairman and CEO Joe Tucci and President/COO Pat Gelsinger provided some details about Project Lightning indicating it would be available in 2011. In addition, several bloggers outside of EMC have speculated about Lightning (Chris Mellor in some detail) and some insiders as well (e.g. Chad Sakac and Chuck Hollis) have blogged and spoken about the project. In addition, Gelsinger and Sakac “showed more leg” at Oracle OpenWorld in September, holding up a PCIe card for the audience and providing some additional detail.
Here’s what we know (or think we know):
- Project Lightning is a flash-based, server-side cache solution designed to front end EMC backend arrays via a PCIe server card.
- It has been widely reported that the flash group is a separate organization inside of EMC run by Senior Vice President Mark Sorenson.
- EMC began shipping a beta version of the product last summer.
- EMC has talked about three phases to the project: 1) Caching software married with a PCIe card; 2) Deeper integration with EMC’s FAST solution and VNX/VMAX backend arrays; 3) A distributed cache coherency capability with shared LUNs across multiple servers.
- At OpenWorld Gelsinger/Sakac said the card shown was 320GB while subsequent reports cited somewhat smaller capacities.
- EMC has announced that it will announce Project Lightning on February 6th.
- It has been widely speculated that the flash technology card is supplied by Micron. In addition, on its most recent earnings call, LSI CEO Abhi Talwalkar told analysts:
“We are pleased to be participating in the EMC Lightning program, which we expect to begin ramping this quarter.”
When contacted, EMC officials had no comment.
As well…EMC has announced plans to deliver all-flash VMAX and VNX arrays and embrace MLC (mulit-layer cell) flash technology as opposed to only SLC.
Where Does Project Lightning Fit?
In 2008, EMC landed a haymaker and completely changed the performance dynamic of the high end array marketplace. At the time, Hitachi Data Systems was a major thorn in EMC’s side, claiming “top gun” performance in the array marketplace and effectively putting Symmetrix on the performance defensive with it’s 1 million IOPs claims. EMC completely neutralized HDS marketing with that announcement and surprised the industry with the addition of flash to Symmetrix. Meanwhile, startups like Fusion-io were coming out of stealth with a new vision to totally disrupt traditional approaches to IO management.
For the past twenty years, storage function (e.g. performance/data management, data protection, copy services, etc) has steadily moved away from the server toward the SAN. The trend has been reversing since flash hit the enterprise. Having a persistent non-mechanical resource closer to the processor is swinging the momentum back to the server side. In my opinion, EMC downplayed this trend and was caught napping. EMC was smart to create a separate business unit to accelerate its time to market. The giant has awakened and Project Lighting is all about EMC extending its core value proposition towards the server, protecting its huge installed base and staking its claim in the increasingly granular storage hierarchy (see chart below).
David Floyer has written several truly amazing posts describing this new storage hierarchy and how systems and storage design is changing. Floyer describes five layers in this new storage hierarchy, three physical and two management:
- Local working flash layer - close to the processor using atomic writes and eliminating all physical and logical (e.g. SCSI protocol) constraints of hard disk drives.
- Active Data Management – a management layer that controls the integrity and movement of data between the top and middle physical layers.
- Note, this layer is a reference model for what could be the later stages of Project Lightning with a plan to integrate tiering and cache coherency between the servers and storage arrays.
- Distributed Shared Flash Layer- a low latency layer shared locally and remotely by server clusters.
- Note, this layer could be a reference architecture for a combination of flash-only arrays from vendors such as SolidFire that include flash IO tiering and management and federated arrays from HP 3PAR, EMC, NetApp, etc.
- Archive Management layer – to manage the interface between the Distributed Shared Flash Layer and the Archive Disk Layer.
- Distributed Archive/Long-Term Backup Layer – the “cheapest and deepest” layer.
From the information we have thus far, Project Lightning is aimed at the “fat middle” of this emerging hierarchy– initially as a front end cache to spinning disk. Over time, given its ambitions, VMware asset and huge portfolio, one would expect EMC to attempt to own the entire emerging storage hierarchy; from virtual server to the lowest cost archive layer.
Make no mistake – this graphic conceptually describes the storage battleground for the next two decades and EMC has every intent of trying to dominate the playing field.
They key for EMC is its Fully Automated Storage Tiering (FAST) software. EMC will position that secret sauce with its largest customers as the most cost effective, most logical, safest way to secure and manage data throughout the new storage hierarchy as it evolves. Indeed, when it comes to buying storage, EMC is perceived by the majority of storage practitioners as the safest bet.
Based on the speculation in the marketplace, and information in the public domain, what’s missing from Project Lighting is a way to play in the very top layer. Only Fusion-io, with its VSL technology and atomic write capability has a solution at that level today. Many observers hear “PCIe” and think Fusion-io knockoff, but PCIe is not the 10X leverage point. Rather eliminating what Intel’s Pauline Nist calls “the horrible storage stack” is.
After Fusion-io’s most recent earnings call, the stock dropped due to margin pressures and speculation about Project Lightning. It’s not surprising but generally these solutions aren’t head-to-head in the near term. Project Lightning is all about extending EMC’s storage value proposition closer to the processor. Fusion-io is all about radically changing application performance– and for that you have to pay a premium. Over time, however these franchises are on a collision course. It’s why Fusion went public – it’s not looking for a quick flip, it’s going after the Superbowl of brass rings. Fusion-io has first mover advantage. EMC has…well it’s so obvious I don’t even have to state its advantages.
Will this rising tide lift all flash ships? Specifically:
- Will EMC’s extension of flash up the hierarchy lend more credibility to an already frothy flash market (I’d bet yes)?
- For which server vendors will EMC announce it’s PCIe solution has been qualified? Cisco’s a lock right? Will HP, Dell and IBM drag their feet and risk being less competitive? Interesting question.
- Will EMC’s initial PCIe solution freeze the market and buy it time to extend its value prop or will it accelerate competitive pressure (my bet is yes and yes).
- Over time, will all active data will reside on flash (yes)?
- Where is the new point of control for managing data? Is it the disk array? An independent software layer? The server (dunno)?
- How will system suppliers respond? Well I guess that’s pretty obvious too – systems company focus, database expertise, “we have our best people working on it…”
- Where is the best bang for the user buck (need to see the announcement first)?
- Will cost/IO take over from cost/TB and become the new metric for storage purchases (hmmm. Should be but No)?
- How will the current disk drive supply shortage impact the adoption of flash?
Join us for the Discussion
The Wikibon community is gathering on February 9, 2012 to dissect and analyze the Project Lighting announcement. I hope you can join us. Here are the details of the Peer Incite Research Meeting. Please come on the call and share your perspectives with the community.