Archive for June, 2009



Up or Out?

Friday, June 26th, 2009

I recently read a post that took a new twist on the long-term debate over whether it’s better to scale up (buy bigger servers) or scale out (buy more servers).  Traditionally this battle has been fought mostly on the technical considerations only.  ”Which is better for processing the real-time inventory of my growing  Dippin’ Dots empire vs. the fast serving of web  pages on my trendy social site?”

DIPPIN' DOTS BANANA SPLIT

The conversation is often reduced to raw number crunching power vs. the benefits of highly parallel processing or high availability.  But in this era of sacrifice, we might want to take a look at the oft-overlooked cost factors lurking behind the curtain.  Following the framework of the aforementioned post, let’s consider the costs using IBM hardware.

First, we fire up IBM’s server configuration tool and build a big dreadnought-class server with an x3950 M2:

  • 4 CPU sockets (using 6-core processors)
  • 32 memory sockets
  • 4 drive bays
  • 2 power supplies
  • 4U

Total Price:  $68,429  MSRP.  In Pennsylvania, we throw on another 6% for the Governor, so the number rounds to $72,500.   Sure enough, scaling up has the hefty price tag one would expect.

If instead we were to scale out, what kind of horsepower could we get for the same money?  Taking a trip to the other end of the product line, we find the modest x3250 M2:

  • 1 CPU socket (using a 4 core processor)
  • 4 memory sockets
  • 2 drive bays
  • 1 power supply
  • 1U

Total Price: $2,431 MSRP.  Allowing again for the Governor, we come in at $2,575, which means that for the same $72,500 we could buy 28 of these unassuming smaller servers.

So, if we decide to go shopping to IBM with $72,500 as our budget, what can we get for our money?

: ———— Scaling Up           Scaling Out

CPU’s               24                          112

RAM              256GB                   224GB

Disk               1.2TB                      28TB

It would seem that scaling out puts more resources in our data center for the same money.  Score 1 for scaling out.

Now let’s take a look at things from the software angle:

:————————————— Scaling Up             Scaling Out

Windows 2K8 Server*            $2,515                     $20,524

SQL-Server                              $7,400                    $16,800

Our quick mental math says that scaling out costs nearly 4 times as much in software.  Score 1 for scaling up.

And now for the tie breaker, let’s examine operational power costs, assuming the boxes run on average at 50% of peak and without factoring in cooling:

——————————-: Scaling Up           Scaling Out

Peak Watts                1440w                     9,828w

Power Cost/Year        $441                       $3,013

Scaling out is an order of magnitude higher in power costs.  Final score: Scaling up appears to win in a narrow 2-1 victory.

Having seen the costs, which approach seems to make more sense?   If you object to this question, you’re quite right to do so.  From a strictly financial point of view, scaling up seems to be way the go, unless you decide to level the playing field by zeroing the software costs with open source (e.g. Linux and PostgreSQL).  Scaling out becomes more financially appealing when open source is in play, which is what we often find in places like Google.

Of course, the decision can’t be made solely from a financial point of view, but prior to this exercise  have you ever even considered these hidden-in-plain-sight costs?  Ultimately the decision still does still come down to your particular business needs which must be discussed on the technical requirements involved.   Watching your team whiteboard the various options can sometimes be more tedious than reading Klingon poetry,

klingon2

but you need to let the team work through both the technical and cost considerations to arrive at the best solution.

There are two take-aways from this example.  The first is that when the technical requirements don’t point hard in either direction, you may be able to appeal to cost to help arbitrate the decision.  The second is that you really don’t need to make these types of decisions anymore.  The infrastructure utility trend is already in motion and is gaining momentum.   Before investing significant capital of any scale, consider deploying new applications in a professional hosting data center. Outsource these ongoing scaling decisions to others while you focus on the bigger picture of providing the right applications for your business.

//spk

* Please don’t flame me with comments like “How’d you get those prices, we only pay $20 for Win2K8 server?”   I haven’t spent the four years of education required to be fully conversant in the Microsoft  Licensing program, which is more complex and complicated than ancient Hebrew Law.  These prices were based on recent customer quotes and internal pricing from our distributors.

Post to Twitter Tweet This Post to Delicious Delicious Post to Digg Digg This Post Post to StumbleUpon Stumble This Post

Information Bankruptcy

Thursday, June 18th, 2009

I may need a government bail out. At the rate information pours into my email and PDA, I fear that my planned time away from the office tomorrow will leave me hopelessly buried in electronic obligations that I will be unable to pay. I’ll likely have to declare Information Bankruptcy. This is an advanced form of  Email Bankruptcy, in which I seem to be finding myself more and more.  This is such an insidious problem that I feel compelled to provide the following public service announcement:

To say one has “information overload”  has become cliché, yet the multi-headed beast of email and social networking that hounds us daily will not go away and either has to be fed daily, or dealt with in some other way.

Social networking can be easily controlled with a little willpower – Just Say No, as Nancy Reagan would advise.  Drop the mantra of “Live to Tweet, Tweet to Live.”   Inbound email, on the other hand, is not within our control, so Lawrence Lessig’s prescription for declaring email bankruptcy is very tempting:

1) Collect the email addresses of everyone you haven’t replied to. Paste them into the BCC field of a new message that you’ll send to yourself.

2) Write a polite note explaining your predicament.  Genuinely apologize and promise to keep up with your email in the future. Try to sound credible.

3) Ask for a resend of anything particularly pressing, and offer to give such messages special attention.

 
And yet having done that, I’m reminded by another veteran of the email wars that this is only a temporary fix because the underlying problem has not been addressed:

A one-time erasure of communication debt would give temporary relief, but the basic challenge remains; the net number of requests for my attention exceeds my ability to provide that attention by at least an order of magnitude. And the disparity around my ability to thoughtfully respond to my pile may be ten or more times worse still. The scale is insanely out of whack.

 
So, email bankruptcy may not be enough to save us. The problem is that human beings simply don’t scale. We just can’t add another processor or more memory. Now throw LinkedIn, Facebook, Twitter, et. al. on the fire, and the blaze is completely out of control. The next logical step up would be total Information Bankruptcy, which would include any form of electronic communication that demands our near-constant attention.  It would be extremely difficult to pull this off without committing occupational suicide, but there are things we can do to improve our lot.  Does your Blackberry really need to go off everytime a network interface flaps?  Better to adjust the monitoring system than to get awake every 5 minutes from 2:49 AM to 4:30 AM during maintenance windows.

No, we can’t pull completely out of the Information Matrix, but I do believe taming the monster is possible.  It requires good email hygiene and the cooperation of others, which starts with us. At risk of stating the obvious, email is not the only way, and certainly not always the best way to communicate.  So the call to action is this: Take better advantage of IM, texting, and (shock) the telephone.  And if the target of your message is in the next office or cubicle, get up and deliver the message in person!   Scale your communication method to the type of bandwidth required.   If you’re typing more than a paragraph or have to hit Page Down to review your message, you probably ought to be picking up the phone instead.

If you’ve emailed me recently and I haven’t responded, please don’t be offended.   I’ll get back to you soon…really. That is, of course, if you don’t get a bankruptcy notice from me first.

//spk

p.s.  If you agree with me on this, please send me an email.

Tired of dealing with your infrastructure?  Why not put it in a professionally managed data center?

Post to Twitter Tweet This Post to Delicious Delicious Post to Digg Digg This Post Post to StumbleUpon Stumble This Post

Worse Than Failure

Friday, June 12th, 2009

Whether you have your own shop or host your gear somewhere else, this week’s horror story at VAserv should serve as a wake-up call if you’re responsible for safeguarding vital company data, especially your customer’s data.

To briefly sum up the story, hackers took out 100,000, (yes 100,000) web sites, many of them permanently, in an evening’s worth of work.   Just restore the backup, you say?  Not so fast.

VAserv basically offers low-cost Web hosting services using virtualized private servers based on HyperVM. As of Wednesday morning, it was not clear how many of its customers — many of them based in the U.S — had irretrievably lost data in the attack. That number could be high, though, because half of those affected had apparently signed up for an unmanaged service that doesn’t include backups, according to the Register. [emphasis added]

 
And for those customers that did sign up for backup?

A note on VAserv’s Web site, which is now just a text document with details on the company’s restoration efforts, claimed its staff had been working “tirelessly” over the last 48 hours. “However, we have now reached the end of all of our servers, and as such, if your server is not currently up, or not partly up, then it is unfortunate that you will have lost your data due to this third-party attack,” the note said.

Oh the humanity, indeed. ComputerWorld’s Jaikumar Vijayan receives this week’s Master of the Understatement award:

The continuing fallout from a hacking incident at U.K.-based Web hosting company VAserv should serve as a powerful reminder that companies need proper data backup and disaster recovery procedures. The incident, which could result in a fire sale of VAserv to another hosting provider, is also an especially stark example of the kind of havoc that a malicious attacker can wreak on businesses.  [more emphasis added]

 
Can you say ‘class action lawsuit?’

Attempts to reach Rus Foster, VAserve’s director via e-mail and phone were unsuccessful. But the terse updates on the company’s Web site and the thousands of customer posts on a discussion forum painted a picture of total chaos.

I’ve personally reached the end of my physical and emotional tether” Foster wrote in one post on the discussion forum late Tuesday evening. “We have worked pretty much continuously for the last few days firefighting.”

Foster wrote in a post that suggested he was putting the company up for sale. In his note, Foster said he had two options: Do what’s best for the customer base by getting “some big boys in behind” to help get things back up and running. The other he said was to simply “Run away and hide and just say to everyone “good bye”"

 
Run away and hide?  When did that become a viable option for gross negligence?  No one can outrun the long arm of the Bar Association.

matrix42

I’m reminded of a line from The Architect’s classic speech: “There are levels of survival we are prepared to accept.”   There are clearly plenty of folks that seem comfortable managing their IT shops that way.  We see it all the time when we look at their backup strategies and disaster plans, if they have any.  It seems to me that being totally wiped out or having to sell our companies because of something so easily preventable as failed backups is not one of those acceptable levels. But wait, it was those scoundrels the hackers, wasn’t it?  They caused the problem, and they killed the company.  No they didn’t. To be sure, the hackers wreaked havoc, but what they really did was expose the ultimate game-ending event: no backups. Had proper backup procedures been in place and restores regularly tested, the incident would have been merely one of  downtime and possibly SLA penalties. (Yes. I know credit card data was also stolen, but that’s not necessarily a game-ender.)

Being an infrastructure company, we routinely preach about the need for proper backup and restore procedures, and the need to test them.  Sadly, it often falls on deaf ears, and while we do occasionally read an obituary like VAserv’s,  death-by-no-backups is happening all the time in companies you’ll never hear about.

There’s another quote I like from the Matrix: “You hear that Mr. Anderson?  That is the sound of inevitability…it is the sound of your death…”   If you aren’t testing your ability to restore your backups (you do have them don’t you?) , the sound of inevitability may be tolling for you.

Hope is not a strategy.  If backups and restores stress you out, or you’re just hoping they’ll be there when you need them, consider handing it all over to a group of people who live and breathe it. They actually enjoy backups, and they’ll take good care of your gear too.

//spk

Post to Twitter Tweet This Post to Delicious Delicious Post to Digg Digg This Post Post to StumbleUpon Stumble This Post


Twitter links powered by Tweet This v1.6.1, a WordPress plugin for Twitter.