Oh – if ONLY there was a way to…

The lament of many cloud compute providers as we engage with them…   “I wish someone would offer a storage subsystem:”

• Was transactional (which VMware needs today to host VMDK storage) – but scaled, so I could manage at scale with less:
• scaled out linearly to large clusters (lets say 16 nodes – for a total of 128 intel cores for dealing with the IO load), using commodity x86 servers, but commodity components built for a service provider.
• where any resource in that cluster – brains, ports, caches could all service any IO – meaning it’s one big pool.  A datastore literally is everywhere, but also dependent on no one element.
• that could scale up to huge capacity (let’s say petabytes) without increasing my FTE cost because it’s designed for management at scale (hey, just like Cisco UCS!)
• where everything could be wide-striped across the entire backend of the array – so it acts as one big aggregate pool of ports, IO, brains, IOps and GBs.
• that was deeply integrated with VMware:
• offering vCenter plugins for end-to-end visibility
• complete automation of all path management – period.
• delegated storage provisioning in vCenter – where storage could be allocated to datacenter objects and cluster objects, and the VMware admins could self-service provision their own storage as needed?
• vStorage API for Array Integration support and much more…
• that was uber-efficient:
• enabling use of massive 2TB SATA disks (lowest $/GB), as well as solid state (lowest $/IOp)
• had massive scale-out cache models to absorb common workloads
• Supported pool expansion and rebalancing, as well as simple and easy zero reclamation.
• could auto-tier – which coupled with VMware SIOC = DRS for storage
• Could be coupled by that same vendor with cloud object storage at huge scale for non-transactional use cases and next-generation application use so they could compete with Amazon S3.
• Could be purchased and supported by someone with global reach and scale so as they grow, and need new stuff in places like Brazil, they could be counted on as a partner.
• Was built like a tank – an outage for a service provider is very bad.

And perhaps most of all:

• could support their business model – meaning that they could acquire it for:
• low dollars/GB if they like a capex-oriented model
• low cents/GB/month if they like an opex economic model
• Ultimately able to support services in the $0.10/GB/month band as end-user pricing with good business margin, and simultaneously also power higher-value services in the $0.25/GB/month and higher range.

If only EMC made something like that – well, the service providers would buy it by the truckload…

Well… WE DO.   And, they ARE.

Read on….

So – what is that product?  It’s a VMAX.

Symmetrix as a “brand” is synonymous with “Rolls Royce Storage” – the stuff you use when it’s GOT to work.   In a way – that’s a good thing.  The brand is based in reality that many customer have trusted, and continue to trust Symmetrix with their most precious data.  It’s local and remote data protection technologies are second to none.   I met a customer yesterday that uses EMC Symmetrix to power critical functions in our financial system, and depend on SRDF Star.  It’s not perfect, and yes, while I know VERY many happy customers, there are some we need to work better on.

On the whole – customers speak with their wallets – and Symmetrix is selling very, very well – growing 32% year over year.  

BTW – if you’re wondering “when should people go EMC Unified then?” – the answer is analagous to “why do people choose rack-mount vs. blade systems?   Sometimes it’s about managing at scale and cost at scale, sometimes it’s about lower cost of entry.  

Unified storage platforms are generally chosen by enterprise customers for their virtual environments – particularly before they start virtualizing the largest, and most mission critical of their apps.  Again – customers are voting with their dollars – our Unified product family grew at 33% year over year.  See Q2 results here).  

But – at the same time that strong brands can help, they can hinder.   The fact of the matter is that the VMAX was a major architectural change in the Symmetrix – making it keep all the stuff that people associate with Symmetrix, but also making it able to also meet the “Cloud Scale-Out Transactional Storage” description above.  

In those use cases, often while they like the idea that they could use SRDF, Timefinder and other cool value-added features – often, to start, they don’t.   And that’s perfectly OK.

Terremark and other service providers use EMC VMAX to power those use IaaS cases, and often simultaneously evaluate Atmos for cloud object storage to support next gen cloud apps – some choosing both, some choosing just one.

Terremark was kind enough to let us write up a whitepaper based on our initial engagement with them.  They put us through our paces, and compared us with the other storage platforms they were using for similar use cases.   You can read that doc here.

 

The same goes for many service providers – SAAVIS, Alphawest have picked VMAX over strong, respected, and sprited competitors.    Just last week, another VERY large service provider just selected EMC VMAX and/or VCE Vblock 2 for their cloud offerings.  

BTW – I’m not implying that any of those customers only uses EMC.   Cloud service providers have choices.   All I’m saying is that I see many of them having specific requirements.  What were their requirements?  Same as the above.   What were their reasons for picking EMC and EMC VMAX?  The same as the above.

We should probably should brand this way of using a VMAX as “Cloud Edition” :-)

Comments

[Marc Farley]

Chad, well written and you were careful not to misprepresent things, but readers might not read it the same way. For instance, readers could assume that Terremark's Enterprise Cloud was being run on VMax systems. That is not what you were saying though.

[Chad Sakac]

Thank you Marc - to be explicit - they use a whole whackload of products (and from what I hear, like many of them).

I feel weird being too explicit about a customer, but here's a more explicit list.

In their managed hosting business - there's our friends from Sunnyvale (NetApp), along with tons of EMC and boatloads of everything else - managed hosting tends to be dominated by "whatever the customer has".

In their cloud compute business, they use EMC, 3PAR and NetApp :-) Terrremark in Amsterdam is using NetApp (for now :-) In the Americas, it's mostly dominated by 3PAR and EMC (key technology partners are reflected on Terremark's site here: http://vcloudexpress.terremark.com/)

BTW - the architectures of 3PAR and VMAX are similar in some ways(while not the same) - both well suited to the technical decision criteria listed above. I would argue that both are very well suited (and we often compete furiously - in fact that "recent win" is actually at a MASSIVE cloud provider and was a toe-to-toe with 3PAR that EMC won in a tight battle, NetApp fell out in the hardware failure analysis phase of the evaluation.)

One thing that led to additional EMC use was Terremarks' expansion into Latin America, where they wanted all the technical capabilities, but also a partner that could support them globally.

This is a new competitive landscape, and one that it's great to have strong competition in :-)

[marc farley]

Chad, the link you posted above isn't working.

You know I have to ask, of those VMAX systems in the Americas, how many VMAXs are presently located in North America where most of their Enterprise Cloud business is. There are Latin American countries where 3PAR does not conduct business, which puts us at a disadvantage there.

There is a difference between Terremark's Enterprise Cloud (http://www.terremark.com/services/cloudcomputing/theenterprisecloud.aspx) and vCloudExpress (http://vcloudexpress.terremark.com/) offerings. It's not unusual for IAAS providers to offer different service levels comprised of different levels of infrastructure products. The link you gave looks like it links to vCloudExpress.

Finally, you mentioned a recent win in your comment above. I know that it does not refer to Terremark and I want to make sure readers are not confused by that. Hmmm, using social media to announce wins is probably a bad idea, but I can appreciate your excitement considering how dominant 3PAR's market share is in the IAAS market. Hey now! Should I start announcing 3PAR wins against EMC in traditional data centers? Just kidding :)

[Jonas Irwin]

Cool paper. I was a little suprised that it didn't talk about any of the advanced QoS (cache partitioning, priority controls) features in the Symm though.

[JohnFul]

Hi Chad,

Interesting paper. I did find a reference to a single V-MAX system in the Terremark vCloudExpress environment on page 8, although the location is not stated. That seemed a bit odd; I was assuming it took more than one to do SRDF. At the Terremark vCloudExpress link you posted, the feature set was very interesting. I did not, however, find any reference to V-MAX or EMC. Are there any further public references that would clarify the exact configuration and location(s) where V-MAX is used by Terremark that you could share?

Thanks

JohnFul