It’s the software, stupid
Posted in Platforms, Advisor - Tom Treadway by Tom TreadwayJon Toigo had an interesting post recently on his DrunkenData blog asking why the price of storage was increasing even though the media cost ($/GB) was decreasing. Good question. And Jon had some good answers - mostly centered on greed. I like to see Jon stick it to The Man.
I’d like to suggest that another major cost driver is the custom hardware used by these big iron companies. Their world is about to change.
If you attended SNW last week you would have noticed the continuing trend of storage vendors providing software-only solutions intended to run on standard x86 motherboards. This is the commoditization of storage that we’ve all been talking about for the last several years. There is absolutely no custom hardware in these solutions. The user, VAR or OEM simply installs the software on a 1U or 2U server, maybe sticks in a RAID card for back-end storage, and voila, you have a damn cheap storage appliance – usually providing multiple iSCSI ports for connection to the host(s). And generally the performance is very good.
The alternative is the traditional active/active, custom controller in a custom enclosure design. Very few standards exist in these designs and costs tend to be high. Software and hardware are closely tied so you can’t mix-and-match solutions from different vendors.
The SATA CabCon committee did some standardization work a few years ago for SATA controllers. Their goal was to break that hardware/software tie, but all that was quickly thrown out the window due to SAS being the new interface-of-choice.
Intel is making a new attempt at standardization of dual controller systems. They had planned a big role out at IDF a few months ago but it was pulled at the last minute. This program has been in the works for at least a year, but it looks like they want to work out a few more of the details and get more vendors buy-in before the big unveiling. BTW, when you hear about it, act surprised. They’re still in stealth mode.
But is this the right direction for the storage industry? Standard x86 systems are just so incredibly cheap. It used to be that you had to use mirroring for system redundancy and therefore you had to double the number of drives. But with SAS you can have multiple RAID heads that share drives. And if you want SATA, well, they’re practically free anyway so go ahead and buy twice as many. The drive manufacturers need the money.
Bottomline: Storage is expensive because of all the crazy custom hardware. But you don’t need that. The value is in the software. Spending money on fancy sheet metal isn’t going to make your data any safer.
TT
November 14th, 2005 at 7:30 am
Tom,
Greed and hardware complexity go hand in hand, IMHO. The original motivation for adding a lot of complex foo to arrays might have been goodness — building a better mousetrap, that sort of thing — but today array controller design and I/O pathing are used by enterprise storage vendors by and large to lock out their competitors and lock in their customers.
Look at EMC Centerra, ostensibly a platform for tracking archival data over the long haul, but in reality just a Japanese Mortgage. The vendor has taken FilePool software from an acquired Belgian company and wedded it to a proprietary controller in order to 1) jack up the cost of otherwise commodity components on the array and 2) to lock the customer into a data encoding scheme from which extrication will be near impossible. In effect, grab the customer by his data, and his heart and mind and wallet will follow.
Not to pick on EMC only (though they are notorious in this regard), there are many other examples we could cite — from HP, IBM and many others.
The need is to purpose-build storage based on what applications require. That will inevitably require heterogeneous storage (not all from one vendor). FC fabrics aren’t the solution because they are unstable (if heterogeneous) and because they lack the intelligence required to give applications the drink of storage they require, rather than a common drink of storage for all.
Thanks for the cite.
Jon Toigo
November 16th, 2005 at 6:04 am
It’s a retorhical question, but why the heck are hetrogenous FC systems so unstable. FC is an ANSI spec defining how SCSI commands are put on a serial wire. (That’s somewhat of a simplification, but not by much.) What’s so hard about that? The FC vendors definitely made enough money to afford a little compatibility testing. The mess they’ve created is just sad. Don’t even get me started about FATA drives. How silly.
I’m looking forward to seeing 10G iSCSI squeeze FC from the top and SAS/SATA squeeze FC from the bottom. Excuse me if I giggle.
TT