The last (half) day of VMworld 2017 US. As always, it's bitter-sweet. There's no coverage of the General Session this morning as I was touring the Switch data centre here in Las Vegas instead.
Tour of the Switch Data Center, Las Vegas
This morning I was invited by a collection of companies to take a tour of the Switch data center here in Las Vegas. I joined a group of about 20 or so other invitees, who all boarded a shuttle from the Mandalay Bay convention center that would take us to the tour. As soon as we approached the complex it was clear that Switch takes their business seriously. The shuttle passed through a gated entrance to a loading area next to one of the Switch buildings, which surrounded by fencing about 10-12 feet high with a discouraging spiky top. Switch branded security staff (all of whom are highly skilled and trained former military folks) were on hand to guide us as soon as we stepped out of the shuttle.
The procedure for gaining entrance to the facility was fairly standard based on my previous data center experiences: DC staff hold on to your ID and give you a visitors badge, and you are then badged in through a formidable-looking full height turnstile. The entrance area and proceeding hallways were very nicely designed, from an aesthetic viewpoint. Coloured lighting, clean considered flooring, and clever designs made from the powder-coated cable trays and metal piping that's used in areas of the DC itself.
The group was lead to an event room which consisted of tiered home-theatre style seating (plush reclining chairs linked into rows), with very modern A/V gear up front including a large projection screen. The various companies involved in arranging the tour, Evolve IP, HPE, Veeam and Ingram Micro, each gave us a brief overview of themselves before relinquishing the floor to a Switch rep, who provided information about their operation and fielded questions.
Switch's Las Vegas facility was claimed to be one of the most advanced data centres in the country. It's the west coast home for Switch (their east coast home is in Philadelphia). By square footage, it's the largest data centre in the world, as about 2.2 million square feet, including building 11 which is not quite open yet (our tour was in building 7). Each building on the campus is about 300 thousand square feet (several football fields worth), and each is capable of dealing with 100 megawatts of power.
Due to innovations pioneered by Switch, and in fact designed by Rob Roy, the owner, who holds 350 claims and patents, Switch is able to reach higher density in their DCs compared to other providers. In fact, they can squeeze 38-42 kW per rack, and some clients are working on achieving about 60+ kW per rack. This is achievable chiefly due to Switch's unique circulation and cooling system.
The air handlers, which are key to circulation and cooling, have been designed in a modular fashion so that they can be produced and integrated into the facility on-demand. Each air handler also has a weather station built-in, so that it can programatically decide which of six available modes it will run in based on current environmental conditions.
The data centres are built with catastrophe in mind. For example the walls are ballistics resilient to protect against possible projectiles from an explosion of some kind, or from war, or perhaps even a meteor strike. Due to this engineering approach, the Switch facilities have not contributed to a single second of customer impact over the 17 years they've been operating.
It was clear throughout the tour that a lot of thought and consideration has gone into the design and production of these facilities, and that same mindset is in place in dealing with it operationally as well. It was an impressive facility, and an impressive tour. Costs for leasing space in the facilities weren't discussed, but I'd say it's safe to say that if you can afford it, you can't go wrong.
vSAN 6.6: A Day in the Life of an I/O [STO1926BU]
The last(!) session attended at VMworld was presided over by John Nicholson and Pete Koehler. John explained how the vSAN datastore is an object store. It allows for granular availability and can meet business performance requirements. Each object in the object store is made up of one or more components. Data is distributed across the vSAN cluster based on the defined VM storage policy, which means you have some control over how distribution is handled.
The typical objects are:
VM home (core VM configuration, i.e. the vmx file),
VM swap (note that it's reserved by default out-of-the-box), and
Virtual disk (VMDKs).
Make sure that your vSAN is designed to take advantage of fault domains. You can create fault domains to increase availability within your vSphere/vSAN cluster. Fault domains help protect your environment against such infrastructure related issues like rack failures. An example given was how fault domains could be designed and implemented to make sure the cluster can tolerate a single rack failure.
Fault domains can be extended into nested fault domains for vSAN stretched clusters. Nested fault domains provide redundancy both locally and across sites. vSAN is smart enough that it will leverage the nested fault domain design to reduce I/O between sites, preferring local data to that at another site.
As the session was quite technically dense, here are a few points gleaned from the rest of the presentation:
vSAN caches data based on the frequency of data access and data locality.
vSAN leverages checksums and disk scrubbing for maintenance.
vSAN provides deduplication and compression (note that you need vSAN Advanced or Enterprise for this).
Dedupe and compression effectively happen during the transition of data between the caching and capacity tier. Not strictly in-line, but very close and provides more benefits around workload performance impacts.
Pete Koehler also discussed vSAN backend storage I/O, which was a little over my head, especially on the last day of the conference.
If you're interested at all in how vSAN does what it does, I highly recommend you watch this session once it becomes available online.
Now It's Time to Say Goodbye
As always, saying goodbye to VMworld is tough. There was so much learned this week, so many folks caught up with, so many steps put on pedometers. It was wonderful. It was exhausting. I'd do it all over again. Until next year, ciao!
![VMware Explore 2022: A New Dawn [Day 2]](/_next/image?url=https%3A%2F%2Fcdn.hashnode.com%2Fres%2Fhashnode%2Fimage%2Fupload%2Fv1737277173084%2F27aa6c63-c1ff-4df5-8074-7bf666ef46fd.jpeg&w=3840&q=75)
![VMware Explore 2022: General Session Live [Day 3]](/_next/image?url=https%3A%2F%2Fcdn.hashnode.com%2Fres%2Fhashnode%2Fimage%2Fupload%2Fv1737277115028%2F62e1ebe4-7d09-46a8-8ac6-aca2c17e2487.jpeg&w=3840&q=75)
![VMware Explore 2022: The More Things Change [Day 1]](/_next/image?url=https%3A%2F%2Fcdn.hashnode.com%2Fres%2Fhashnode%2Fimage%2Fupload%2Fv1737277227689%2Fe6cdc90f-c453-4673-afb3-9dba59700e47.jpeg&w=3840&q=75)
