Softpanorama (slightly skeptical) Open Source Software Educational Society

May the source be with you, but remember the KISS principle ;-) Softpanorama Search

Questionable costs efficiency

Prev | Contents | Next

A lot of issues at this area are still in the state of flux but costs advantages of SaaS approach in many cases proved to be questionable.  Centralization has its costs and usually costs are rising faster then capacity.  So savings from cloud model may well be fake. Here is an interesting observation  'Carbon cost' of Google revealed (BBC News)

This link added by CHS:

US physicist Alex Wissner-Gross claims that a typical Google search on a desktop computer produces about 7g CO2.

However, these figures were disputed by Google, who say a typical search produced only 0.2g of carbon dioxide.

A recent study by American research firm Gartner suggested that IT now causes two percent of global emissions.

Dr Wissner-Gross's study claims that two Google searches on a desktop computer produces 14g of CO2, which is the roughly the equivalent of boiling an electric kettle.

The Harvard academic argues that these carbon emissions stem from the electricity used by the computer terminal and by the power consumed by the large data centers operated by Google around the world.

Speaking to the BBC, he said a combination of clients, networks, servers and people's home computers all added up to a lot of energy usage. End of BBC excerpt.

Recent developments of server technology (virtualization) as well as laptop technology actually suggest quite an opposite direction: local computing can be price competitive and thus may even increase its share ("cloud in the box" in IBM terms or "datacenter in the box" in Sun Microsystems terms concept).  Mobile trailers that are now produced as "portable datacenters". While they are not very successful they can serve as a good illustration of the opposite trend (Microsoft, Sun). 

Existing companies even is more flexible models then pure SaaS are not exactly flourishing. If you look at Salesforce.com’s second quarter 10K, 6-month revenues for the first half of 2007 were $339 million, up from $223 million the year before. Marketing and sales costs were $174 million, slightly over 50% of revenue. Profits were negligible.

Cost-related problem is one of the most serious obstacle on the way to wide adoption of SaaS approach: high "per seat" price is often sited as the major problem is SaaS customer satisfaction surveys. And price trends are not favorable to SaaS. They run in direction completely opposite to Carr's vision.  Other things equal SaaS proved to be great for infrequently used applications. As application usage increases, the return on investment shrinks and might move to the point of negative returns. This is true even for simplistic mail filtering and load redistribution services like Postini: it is one thing to use Postini (now owned by Google) for the encrypted mail and the other as your global email service provider.  In the latter case paying per seat costs you so much money that for half of this price you can have a dedicated IT staff using open source applications with approximately the same quality of spam filtering (using, for example Spam Assassin). In addition you have the ability to customize your solution to better suit enterprise needs (Postini loses too many non-standard mails like automatically generated invoices, confirmation letters, etc; often those have non-standard, spam-like headers).

That means that for enterprises there is no big incentive to switch over: it order to be attractive from financial standpoint "in the cloud" services should be considerably cheaper then "in house" solutions. This is pretty difficult to achieve worth the current trends in hardware and "off the shelf" software cost. If Microsoft can low cost of Office Professional to less then $100 per seat you have Google applications hosed no matter how hard they try.  And even if it looks expensive (and in Eastern Europe it sure still is) the main competitor to Microsoft Office will be not Google applications but free Open Office.  You probably can still run office applications in Google for universities as nobody especially cares about anything but the email and Microsoft Office will be on each student's laptop anyway.  But that's about it

In order to be competitive SaaS providers should maintain substantial pool of unused resources to accommodate the spikes.  That squeezes the profit margins.  If you think about the ability to provide the same amount of resources that a regular corporate laptop has (and for a particular application they all can be used)  for many applications long term profitability of "in the cloud" software services are far from being obvious. Currently a modest one CPU virtual image from Amazon with 1.5G of RAM and 160 G of disk space costs approximately $100 a month: five time the cost of the lease of Vostro 1500 with T5270 CPU, 2G of RAM and 250G hardrives with 3 year warranty (cost $700, lease $20 per month).  And to survive as a business you need to provide a steep discount as for many applications users prefer a software installed on the laptop ;-). Just think about cost to the company for 'entertainment-oriented" users, for example daily playing their workout videos from your "in the cloud" datacenter... I would say that currently the idea of application streaming to laptops beats "in the cloud" service providers for applications like Microsoft Office hands down.  

Also "economy of scale" argument applied to service providers can be misleading. Large enterprise IT already enjoys significant economy of scale and further centralization might not add anything significant. Let's discuss a hypothetical deployment of very simple and relatively smooth for deployment in the cloud case of a single web server for corporate Intranet. Let's assume that this server supports just http and some minimal video streaming services (corporate, vendors and business partners video materials) so bandwidth is not  a big problem.  Now let's compare the cost of getting the same server in a cloud and associated costs:

• Local datacenter deployment. The total hardware costs (one Dell server with the configuration listed above) are between $5K and $10K for three years period (that includes 3 year hardware warranty from Dell). Cost of bandwidth is zero (the server is on local network).  The cost of  basic OS support which includes provision of patches is $350 a year (we assume Suse 10).
   
• Remote "dedicated server" deployment. PowerEdge 1950 with much weaker (and cheaper) configuration (2G of RAM and 2 *146G Harddrive) is approximately $600 a month and PE2950 server with 4G of RAM and 4*146GB drives is $1200 a month (see Web Hosting Dedicated Hosting Dedicated Server Upgrades) . That means that you have  to pay  $17K for software maintenance for PE1950 or 32K for maintenance of PE2950 per three year.  A typical Unix admin in enterprise environment usually service approximately 50 servers and has cost of ($70K * 3 /50)= $4.2 per server per three year.  We did not count possible additional cost of bandwidth in remote deployment and the value of having local knowledgeable people on staff "("goodwill").

The question arise does remote deployment provides any real savings even if we factor in additional building rent/electricity/ air-conditioning, etc related costs?

So far cost effectiveness of outsourced services is mainly observable only on lower level of services and for small start-ups. For example, Web hosting offerings are very attractive for small to medium size web sites. Really primitive, Spartan sites like this one benefits most. But as you move in more complex LAMP applications or god forbid into Java stack benefits gradually disappear and then quickly reverse. Especially problematic are computationally intensive Java based web sites as cost of outsourced solution here is substantially more (you generally need a dedicated server) and reliability is equal to reliability that can be achieves by using one of "small business" Internet connectivity providers and local servers with outsourced to vendor support.

Outsourced data backup is also not cheap (approximately $1 per month per 1G in case of Amazon) and benefit disproportionally small companies and individuals. 

If we think about replacing local servers to remote the cost effectiveness is also open to review. Minimal virtual instance of web server on Amazon costs approximately $100 per month (includes 30G per month traffic).  The cost of equivalent dedicated "real" server from Dell with 3 year on-site warranty can be approximated as $500. That leaves $3100 per three year for maintenance costs in local datacenter to break even with "in the cloud" provider. 

For some applications costs can be less. For example,  Joyent provides hosting of WEB applications on virtual machines with minimal cost of $45 per month which is OK price for a test or QA server. But if you want just one "real" CPU (which is desirable for production instance) costs increases to $250 a month, even more then for Amazon.  In the latter case your one year costs are equal of the cost of a nice locally installed Dell 1U server with outsourced 3 year vendor-based 24x7 hardware support. What savings we are talking about ? 

If Web application works strictly for the company and not for external customers this is a wash. For small amount of traffic Verizon Broadband provides a dedicated digital line with speeds comparative with cable modems for $40 a month service (service also includes 250MB of free online backup). For regions that do not have such service, old good T1 line is approximately $250-$300 a month.  That means that local VMware-based virtual appliances  can be cost effective alternative to "in the cloud" services even for small organizations.    Here is one telling comment to Carr's blog entry Cloud may squeeze margins, says Microsoft exec

I'm sorry guys, I just don't get it with "Cloud Computing"! Really... think about it... All this talk about dispensing with the in-house IT staff and not having to buy and maintain servers?? Passing off the application to the "Cloud"?

Come on! All this would have held true in the days when companies were spending 250K for a mini mainframe, but you can buy one heck of a server for 10,000 bucks; enough of a machine to run a 40 million a year company on. Desktops are under a $1000. Believe me, it's not a compelling argument for cloud computing.

Also, most companies have a hardware and network person visit them by the hour when needed. Even for a company using MS Server 2003, it's bullet-proof enough that it just sits there and works. It may not be Linux or Unix, but it's good enough for a reasonable price. So the "IT guy" isn't going to break the bank. Computing has become a commodity, so there ARE no big savings there - PERIOD!

Now let's say you rent software like Netsuite or Salesforce.com... You pay and you pay! Every two years you'll be paying the equivalent of a one-time software purchase for a product outright. It doesn't even make financial sense. Software rental is not cheap. What if the network or Internet provider goes down? At least with a desktop system you can still work on a local version.

Most desktop systems are rapidly becoming cloud systems anyway. Like Foundation 3000 accounting software from Softrend Systems Inc. http://www.softrend.com . This software you buy only once and you get both desktop feature and cloud features included. It's not an either/or decision.

Using Online features in software in no way makes the desktop redundant. What the heck do most people use to connect to the Internet with...? A desktop computer! Some software will be Internet-based and some will be desktop based. Most computing platforms of the future will be a mix of both. What we're really talking about here is the software pricing model - not whether your application is sitting on "somebody's" server out in a cloud somewhere.

So let's bring the discussion back to Earth and get our heads out of the Clouds!

Come on... let's debate this!

I would say that currently such services are extremely attractive for one time large computational tasks and for testing, but not so much for production.  And the main problem is higher costs and illusory advantages. sure advantages do exist but each and every advantage that SaaS model claims can be replicated cheaper and more efficiently by competing technologies (see below).

At the same time lab/test servers and related equipment in large corporations are grossly underutilized.  Here SaaS is cost effective and makes sense right now (but local VMware servers can do eseentially the same trick).

As for large computational tasks recently NYT performed conversion of its archive to digital form on Amazon for some ridiculously low price.  This is another area were SaaS makes sense right now. there are probably much more areas then those two that I know about. But to claim that this is a universal solution like Carr did is the utter level of incompetence and rogue forecasting...

Prev | Contents | Next