In reading this article, you might think it runs along the lines of the still popular Hulk vs Superman Debate except in this case, it’s storage related and this storyline is Storage Mojo vs The SSD Guy.
Self-professed ‘Storage Mojo’ is none other than Robin Harris while ‘The SSD Guy’ seemed to be a tag given to me years ago. I will even help in driving some traffic to Robin’s Blog because it is there that this amusing tale begins, through an article he aptly named, ‘Are SSD-Based Arrays a Bad Idea.’
It spoke to the ideal that using NAND flash arrays in a custom design made more sense than that of the present SSD based solutions that have taken the storage community by storm. I was so taken by the article, in fact, that I wrote a comment which was replied to, yet never posted. Read carefully as it seems to speak to the intellect and open nature of Robin Harris himself:
MY COMMENT SUBMISSION TO STORAGE MOJO
Am I missing something here? This article seems to take into comparison the similarity of comparing a car without an engine, IMHO, to one that has an engine. The cost of a flash based array, in the end, would not be significantly lower as you then need to consider your means to an end when creating the speed for the particular environment which will add hardware and software to the mix. We are also forgetting about such things as garbage collection, error correction, TRIM, power management, caching and on and on. I must be missing something here and have no problem in asking what I am not seeing and putting my name to it???
ROBIN HARRIS (AKA STORAGE MOJO) E-MAILS RETURN:
StorageMojo is talking about using flash SSDs in enterprise storage arrays. No one is forgetting about such things as garbage collection, error correction, TRIM, power management, caching and on and on¦ So yes, I’d say you are missing how flash SSDs designed for consumers might be less than ideal for enterprise storage. But that isn’t your beat, is it? Also, I’m curious: is English your second language? I see odd constructions in your writing that make me curious. Regards, Robin
I have to admit that I thought that Robin’s response was well worth the press and might help many in the community to understand his point of view in the first place. In any case, here are a few thoughts, to help Robin along in understanding that, there is a great deal more to the story than simply throwing flash into a system.
We also don’t seem to be alone in wondering what exactly Robin was thinking as Pure Storage seemed to be just as bewildered in their article ‘SSDs in enterprise flash arrays‘as was StorageIO with Why SSD Based Arrays and Storage Appliances Can Be A good Idea.
AN SSD IS A COMPLETE SYSTEM
I think the main thing that that even the most non-storage techie should understand is that there are key differences between buying raw flash and buying an SSD for these types of applications. To start with a conclusion, both raw flash and a packaged SSD have a place in the enterprise, and they offer different value propositions.
An SSD, however, is a complete system and equating it to raw flash is like comparing buying a car without an engine to one that has an engine. Heres a quick table to describe some of the differences in capabilities between an SSD and raw flash:
|Error Protection|| |
|Garbage Collection|| |
|Power Management|| |
|Smallest form factor|| |
|Lowest cost per unprotected bit|| |
The core components of an SSD are the NAND flash memory, DRAM buffer memory (in many cases), the controller, and the firmware. The ‘array’ of companies that have completely rewritten the book of flash through their SSD products is large and it seems a new door (in both consumer and enterprise’ markets) opens each day with respect to possibilities that are non-existent through flash alone. Flash memory based SSDs are increasingly adopted by almost all the critical segments of the industry that need to maintain 24 x 7 uptime, such as rack-mounted servers, RAID systems, SAN storage and NAS servers.
On the flip side, raw flash offers the optimal form factor because it isnt standards-based. For enterprise applications, raw flash needs a controller with firmware and software added to it to make a complete system. As semiconductor process technology advances, creating these algorithms for flash control and management will become increasingly challenging.