We don’t normally comment on commentary at the 360is blog. Most of our postings are more substantial, whitepapers, guides, seminars, or software we’ve produced or packaged. However there is a good posting (Cloud Hosting & Service Provider Forum) by Richard Talaber (ex-CTO’s office at VMware) on LinkedIn and thought I’d reflect on it here.
For those of you who
aren’t on LinkedIn, or don’t want to join the group in question, Richard was
responding to Beth Pariseau’s article
at TechTarget on VMware’s recent abolition of the vRAM tax.
Richard makes the case
that VMware is probably not going to be the right virtualisation stack for a
commodity “$75” per-month per-VM hosting/cloud service, vRAM tax or not. He
also believes that VMware still makes sense for the smaller market for high-value
fully-managed VM hosting, with load balancing, monitoring, fault tolerance, and
high performance throughput bells and whistles.
Mostly I agree with
Richard, he makes sensible arguments supported by good assumptions but in-spite
of this I still see more than a few problems for VMware in the short and medium
term. Let me explain...
All service providers,
including Cloud & Hosting providers, need to own their own infrastructure
if they are to have a sustainable business. Whether it be a fibre optic network
(in the case of a carriers) or their provisioning, monitoring, and management
stack in the case of Cloud providers. Unless a service provider owns his own infrastructure,
he cannot exercise full control over his costs. Cost control is vital in a
service provider business, as profits depends upon volume, and any tax on
profits is unwelcome, especially one that grows in-line with that volume.
Hardware must be paid
for (hopefully using inexpensive long-term debt secured against assets you also
own), but if there is even a chance of finding a cheap or inexpensive
hypervisor-and-management-stack then Cloud providers have to take it. This is
why you see so many of them with Xen, XenServer or KVM. This is why the list of
users of OpenStack reads like a who’s who of Telecoms and Hosting. For those
that argue I'm not considering TCO, think about this. I've hired good technical
guys, pay them well, they work hard for me, I'm supposed to be a player in Cloud. If my guys can’t engineer something solid, maintainable, and cost
effective then what am I doing in this business? These guys and the platform
they build from KVM, Xen, or whatever, are my long-term competitive advantage.
At least until I'm big enough to be building my own data centers from scratch.
If I'm a Cloud
provider and my maintenance renewal (or any other per-unit-customer cost) jumps
even 1%, its a big deal, I've got a bazillion systems after all.
For parts of the
infrastructure that can’t be wholly owned, or cannot be had for free (with effort from
my hardworking DevOps guys), the Cloud provider's only weapon is over-subscription. Pay for the product then figure out how to dilute that cost
by maximising over-subscription, balance providing a great service with running
the hypervisor and server hot. It is a very difficult balance. This is one of
the reasons the vRAM tax was so wildly unpopular with Cloud providers, it took
away one of their profit levers, or at the very least shortened it.
Hosting/Cloud
providers are currently at a very immature stage in life, they are talking
about RAM, and IOPS, and vCPUs or is it CPUs or is it Cores or Threads? What do
any of those things mean? What about transfer rates, or latency, what kind of
cores? What is a thread? These things don't mean much to business people so good
luck trying to explain them. When business sees 3 prices for what they perceive
as essentially the same service, they are going to go for the cheapest and find
out later if it was appropriate. By Richard’s own admission, 80% of workloads
are relatively modest in demands. VMware will not manage to get across an
argument based upon performance; a performance lead of the size VMware achieves
(or even aspires to achieve) is not a sustainable advantage for the Cloud provider
market.
Richard's last
sentence is key: "Perhaps VMware should consider a public cloud price that
is significantly less expensive than a private cloud price.". Logical. If
the public cloud VMware price were somehow so small as to barely matter, then service
providers would not spend time fiddling around with the competition or knocking
up free alternative platforms....but how to segment the product? I think it is
too late now, there are alternatives in-use, there are engineers out there with
experience of building these infrastructures for service providers, and even if
you can't afford to buy such people you can probably rent them. Talk to us.
- Better performance is not a sustainable advantage for VMware.
- A richer feature set probably isn’t either.
- Cloud providers dislike anything that handicaps their ability to over-subscribe.
- Cost that grows in-line with customer volume is a no-no, unless it is absolutely impossible to avoid.
VMware could make
their product more appealing to commodity Cloud providers, but in order to do
so they’ll have to start thinking more like them. Or talk to someone who does. 360is has helped companies like CheckPoint, HP, and Microsoft understand the Cloud service provider market. You know where to find us.
Posts
No comments:
Post a Comment