ORIGINALLY POSTED 6th May 2009
13,717 views on developerworks
The “modular monolith” has arrived. Oh how the EMC spin doctors must have jumped for joy when they realised the next version of the almost 20 year old Symmetrix family, the DMX-5 could also mean DMX-V in roman numerals. Better still lets try and jazz it up furter, how about DMX V-“Max”. Oh how 1990’s.
My first thought was those “chavved up Nova’s” [CHAV] [NOVA] and the mildly amusing Max Power” magazine in the UK. How to take something that wasn’t really designed to be ‘chavved up’ and do just that. While ultimately, under the covers its still a Nova.
I was on leave the week EMC did their major U-turn and decreed that SVC was the most revolutionary thing that had happened to the storage industry. Yes I did mean to say SVC there and not DMX-5 (aka the modular monolith). Let me explain…
For the last 6 years or so anyone that asked EMC about SVC would get the same pitch. I’ve seen it, I’ve got it. They basically bought some SVC nodes from us and tried to break them. So you can (if you ignore our config guidelines and setup the SAN in such a way) cause SVC problems. The same is true for any computing device. Try asking windows to defragment a disk – to you and me this would mean free space concatenation… Anyway, one of the key corner stones of EMC’s anti-SVC campaign was Would you want to put some Intel devices in the middle of your SAN?
Yes. Of course. There is nothing to say that an Intel, or AMD CPU is any less reliable than a Power, SPARC or custom ASIC with embedded ARC, MIPS or other CPU. The key advantage to using Intel or AMD is that they do all the hard work for you. You have a very high bandwidth capable CPU at your disposal. So for years we have easily combated the EMC spin by showing customers the bandwidth (memory and bus) available because we use an Intel based planar and CPU. Yet EMC have always tried to tarnish SVC as ‘not enterprise because it uses Intel’… oh how I laughed when they talked of commodity storage in their latest monolith.
Now lets think about the workings of a storage controller. They have to handle SCSI commands via some kind of interface. They need to process those commands, store the data in a cache (because disks have been slower than memory) and then do some kind of translation on the data. By translation I mean copy services, abstraction etc, then finally write or read some data from a medium (usually magnetic).
These workings are really key, because overall the namespace is important. How addressable is x from y. What kind of architecture would you want to handle this, especially moving from a monolithic to a clustered, agnostic view?
Personally I’d want an architecture that has been designed from the ground up as a modular, commodity, global namespace solution. Not some almost 20ish year old legacy firmware thats been butchered to try and cope with the latest and greatest that hardware can provide. Granted, EMC has finally given up on(and I believe withdrawn) Invista from the market as a bad idea. Oh how things change, and now they are basing their enterprise solution on a modular Intel platform. Oh, you mean like SVC – of course SVC has the advantage of being designed from the ground up to use this modular high performance solution – how much legacy junk does EMC have to keep in said firmware to make sure old Symm users can upgrade, thats a lot of overhead for new hardware to live with.
Of course you lucky folks to just recently buy a DMX-4 can’t upgrade to a DMX-5 and you still need that painful host based EMC services migration plan to get from DMX-4 to DMX-5 – unless you have an SVC at your disposal. Oh and even when you do, now you are locked into an EMC upgrade plan – don’t be fooled by the used of the V – V means nothing more than 5. It certainly doesn’t provide the heterogeneous virtualization of SVC or even USP, you are still stuck to that monolith.
I have to take my hats off to the marketeers, they did a great job of making it sound like something new, but its the same old global cache, same old Symm firmware, same old monolithic approach. OK, so now you can add quadrants as you need, up to octets I guess, but its still vendor lockin.
As for the super advanced functions like migration (yawn – 2003 we did that) and hot spot management – announcements of future plans is one thing, delivery and actual end user benefit is another. Don’t be fooled by the crap that they were first – first to announce, but that patent I filed back in 2004 stands the test of time – you need a solution that chunks, extents up the storage to do that and SVC was the first to do just that. If anything they are playing 6 year catch-up after the wasted effort that was Invista.
Don’t take the spin for granted, think about the underlying architecture and if it was designed to do what it needs to do today. As I’ve said before, the future is bright, the future is truly virtual, not some cobbled together monolith.