Essential Technology: A GCS Blog

Your intrepid reporter recently went behind the scenes at a national cloud hosting provider.  Once I pulled back the curtain, would you believe I found:

1) Single points of failure at the firewall and switch level. Yep. For an organization that spent millions on infrastructure and management tools, why not configure your firewall and switches for High Availability. The devices they used all supported HA configurations.

2) No backup of static configurations. If one of the single-points did fail, surely they could recover quickly? Right? So where do they store the configuration backups of these devices? The staff I spoke to had no idea. Reconfiguring a shared firewall and switches from scratch would be no fun at all.

3) Lack of spare parts. Anyone would stock replacement gear if they designed in a single points of failure. Not this cloud provider. Though the staff asserted that spare switches and firewalls exist, we were unable to locate the equipment. No spare firewall. No spare switch. Supposing it does exist, the replacement process apparently starts like this:

"Step 1: Dig through mountain of parts hoping to find replacement gear."

I'm as optimistic about the cloud as anyone, but caveat emptor.

I've often called out Google and other cloud computing vendors that promise cost savings. The truth is that the SMB market makes better decisions with their IT dollars than larger businesses. The often touted savings just don't exist! Cloud computing is new-enough and its vendors proud-enough that most offerings have a very large margin priced in. Enter SoftLayer. SoftLayer delivers a range of managed hosting options as wells as CloudLayer. CloudLayer is a virtualized, highly available infrastructure priced to compete with on-premise solutions in the SMB space.

Unlike other cloud infrastructure providers, SoftLayer features technology frequently found in the SMB. They utilize Fortinet, Dell's Equallogic and Citrix's Xen Server. You'll also find SuperMicro servers and Array Networks load balancers. A powerful web interface enables rapid provisioning and easy management across SoftLayer's three data centers. Pricing for their Public Cloud option starts at only $99 per month for 1 Core + 1GB RAM + 100GB Storage. Bandwidth is included up to 2TB of outbound traffic while inbound and server-to-server is not metered.

SoftLayer data centers are SAS 70 Type II certified and feature bandwidth or peering with many of the top telecoms. They claim 20,000 servers and more than 5700 customers. In a market dominated by enterprise focused companies, it is exciting to see a cloud solution priced for the SMB. I'm planning a trip to Dallas for the tour shortly and an extensive evaluation. I'll fill in more details when available.

Moving large amounts of data offsite is a difficult thing to do. It is difficult to beat tape drives and an automobile for rapid data replication. Obviously this method has serious limitations. With the rise of Shared Storage, most replication today occurs at the Storage layer. Though application layer replication offers major advantages, it is usually pursued in addition to storage replication.

What is storage replication? Most SANs and many NAS devices offers licensed replication as a feature. GCS' two favorite devices, the HP Lefthand and Dell Equallogic, include these licenses at no cost. By placing a SAN in-office and one at a remote data center, then connecting the sites with sufficient bandwidth, the SANs will replicate data between the two at the block level. This is an efficient method. Unfortunately, the SANs cannot differentiate between legitimate data changes (ie, a saved Word doc) and temporary data changes such as SQL log files. Everything is pushed across the connection. This tends to require large amounts of bandwidth.

The HP Lefthand offers both asynchronous and synchronous replication options. With asynch replication changed data is pushed at scheduled intervals. With synchronous replication the devices attempt near real-time replication. The HP Lefthand only offers synchronous when there is less than 20ms of latency between the sites. The HP Lefthand also offers bandwidth management on the SAN to help restrict the amount of bandwidth consumed on your site-to-site connection.

A readily apparent benefit of SAN replication is virtual machine replication. If my VMs are stored on my SAN, they are replicated to my remote sites. This dramatically reduces recovery time in a variety of outages.

GCS also recommends Microsoft DFS-R for replication of file stores. Simple, reliable and efficient, this technology allows Windows NAS devices to replicate between sites and provide failover in the event of an outage.

We've been installing these ProLiant G6s for a while now and were really impressed with the noise reduction on the 1 Us. One of our technicians, Nikolai Wunderlin, had his camera with him on an install and captured a few minutes of the silence for you. Don't let the first shaky seconds deter you - he gets on a roll about 5 seconds into it! Thanks Nik!

This brief guide will outline the different SAN classes available to the SMB buyer.

Entry Level: If you're buying a SAN, you need a few basic features. iSCSI support and vendor certification (Microsoft, VMware, Citrix) will meet minimal needs. These will enable your virtual environment to utilize high availability features in any of those vendor's environments. HP's MSA 2012i G2 and Dell's MD3000i both meet these criteria. Expect to spend as much as $15k on these devices. The HP will let you mix and match SAS and SATA drives in a single chassis for best use of your SAN dollars. Neither of these solutions offer thin provisioning or clustered storage. What's that mean? Less efficient per GB and much greater risk of failure. If you're concerned about putting all your VMs in one basket, and you should be, then look to the Mid Level, below.

Mid Level: This is what you want, if you can afford it. Two major features enter play here: thin provisioning and clustered storage. I'll touch on both of these now with more to come soon. Thin provisioning allows for oversubscription of storage. Don't worry about it, just do it. Clustered storage is like clustered servers. Two, or more, boxes configured for failover. The HP Lefthand provides the lowest entry cost to true clustered storage. Dell's Equallogic, Compellent, NetApp, Xiotech and others each offer some unique features. Expect to spend at least $30k on this device.

Enterprise: Forget about it. You can't afford it and wouldn't fit in your server closet if you could. Vendors such as HP, EMC, IBM, NetApp and others live here.

Virtual Storage Appliance: For those organizations that may already have a large investment in internal storage in servers or direct attached storage, a VSA may be the best bet. This software solution aggregates storage across your servers into an iSCSI SAN with similiar feature benefits to a full system. You will be able to support High Availabilty and VMotion/Live Migration with this solution. Obviously since you are only buying the software the entry cost is much lower than a hardware solution. HP's Lefthand offers a VSA for VMWare (Xen and Hyper-V are coming). StorMagic has an interesting option that currently supports only VMWare as well, but Hyper-V support is coming.

A not-too-brief market overview should whet your appetite. Look for more info on the features and key differences between vendors to come soon.

We're finally seeing the ProLiant G5s and virtualization-enabled PowerEdges become widely available in the used server market. Many clients are asking for our recommendations. In a virtualized environment with VMware's High Availability or Microsoft's Failover Clustering, is used hardware an acceptable option?

First let me say that I've long been opposed to used hardware in traditional environments. The hardware represents a small portion of the overall cost of the implementation. The installation and configuration, whether performed in-house or by a consultant, is wasted when the hardware has to replaced. I would rather see that service cost spread across four of five years of use. Why save a few thousand dollars to sacrifice a lot of ROI? Cash is the only valid reason. The difference between new and used gear is rarely enough to determine the fate of a project.

Now we have a highly available, virtualized system. We have architected enough capacity to easily run in the event of a single server failure. Are we more comfortable pushing the typical refresh cycles of servers? YES! If we're only refreshing from a fear of hardware failure, then used equipment should be an option considered. In two recent projects we've been able to go from 20ish servers to two servers without buying a single new server! We're reusing equipment from the organization in these cases, but that compares well to procuring used hardware.

Make sure when ordering that your hardware is identical to take advantage of the high availability features. SANs are still new enough to have limited availability on the used market. If you can knock $20k or $30k out of the upfront cost by using used servers, why not?

HP's Proliant G6 maxes out at 144GB of RAM. 16 core processors are due out in 2012. How many VMs can you cram in a quad, 16-core box with a 144GB of RAM? Enough. Arthur Cole describes trends toward both mainframe-style aggregration and grids of smaller systems. Which is right your for you? One factor in determing the best strategy is:

Additional servers reduce the overhead for virtulization failover. Much like RAID configurations, with 2 servers you must allocate 50% overhead for failover capacity on each server. Three servers only need 33% overhead and so on.

Back in August, I introduced the various components of the System Center family. I also pointed out that the cost would of licensing each product, or even licensing a single product for multiple virtual machines, would be prohibitive for most smaller organizations. However, Microsoft has responded with the Systems Center Management Suite license that includes the licenses to manage multiple virtual machines using the full System Center family for a far more approachable price.  Here is what is included:

·         A license to run the Virtual Machine Manager server.

·         Management license for:

o   Data Protection Manager (DPM)

o   Operations Manager

o   Configuration Manager

o   Virtual Machine Manager (VMM)

This license is sold in two flavors:

·         The Enterprise flavor is licensed per physical server and allows you to manage 4 virtualized operating systems on that server. 

·         The Datacenter flavor is licensed per CPU (a minimum of 2 processors) but allows you to manage an UNLIMITED number of virtualized operating systems on a physical server.  When licensed for two CPU’s the Datacenter license is only about 25% more than Enterprise, and most likely if you have more than 4 VM’s you’ll have two CPU’s and will save money by going with the Datacenter edition.

But for small environments the biggest value of the license comes when using DPM for backup. But to explain the benefit let’s use a small virtual environment example. 

We have a Hyper-V server with 4 virtual machines:

·          A domain controller/file server

·         An Exchange 2007 server

·         A SQL server

·         A SharePoint server

The DPM protection agent also comes in two flavors:

·         Standard Data Protection which allows you to perform basic file level protection of a server.

·         Enterprise Data Protection which is required to provide protection for advanced applications like Exchange, SQL, SharePoint and Hyper-V

One of the benefits that drove this small organization was the ability to backup an entire virtual machine, which means we would need an enterprise DPM license for the Hyper-V server. When a Hyper-V server is protected with the enterprise license you are also permitted to deploy a standard DPM license inside any of its virtual machines. This allows us to backup the domain controller and file server, but the other servers would each require their own Enterprise protection license. So we would need 4 enterprise licenses.  Similarly, to properly protect all these servers with alternate backup products would require a special Hyper-V agent, SQL Agent, SharePoint Agent and Exchange Agent.

But, if this company was to purchase the Enterprise Server Management Suite they would be entitled to all the Enterprise DPM licenses they needed. And because of the new pricing, they would pay less.  Just 3 enterprise DPM licenses would cost more than the enterprise suite license. And on top of that you get management licenses for all the other System Center products.

There are two gotchas with this management suite:

·         The only System Center Server license included is the VMM server license. So you still need to purchase the server licenses for DPM, Operations Manager and Configuration Manager to take advantage of the management licenses included in the suite.

·         Despite all being part of the same product family, the System Center server product cannot all run on the same server.  Only VMM and DPM can coexist.

Because of these limitations it is likely that smaller environments will only take advantage of the VMM and DPM components of the suite, but it still offers a compelling value to consider.

When should I recommend Cloud solutions to my clients? I don't know. Today I'm going to start figuring it out. I'm going to compare an on-premise build with a cloud based solution for a hypothetical 50 employee organization. Allow me to skip a lot of details and say it will provide roughly equivalent features and security. Anything required in both solutions was a wash and therefore ignored. What you're left with is mostly a lot hardware (servers, SANs, firewalls, switches for iSCSI, etc), deployment services and hardware-level support.

NOTE: All pricing is retail and rounded. Individual proposals may vary. A greenfield is assumed - no data migration. Everything that was common to both solutions was excluded so this may represent a small portion of the overall project. 

Cloud Offering:

In the cloud we'll procure 20Ghz of processor capacity, 50GB RAM, 3 TBs storage with 10 Mbps of bandwidth to host 15-18 Virtual Machines. Onsite backup for 1.5 TBs. This will run about $10,000 per month.

On-Premise Offering:

On-premise will provide provide about 60Ghz (over 24 cores), 96GB RAM, 3TB direct attached storage for backup, 3TB iSCSI SAN. We'll plan on hosting 25-30 VMs on this platform. You'll get this for about $160,000 including 200 hours for installation and configuration. See my build here. Note we dropped 25% of capacity for high-availability.

To even the playing field we have to factor in datacenter costs for the On-Premise Offering. I see a rack, 10Mbps of bandwidth, and 40A of power. Say $3000 per month?

Conclusion:

Let's compare 3 years of costs. In the cloud you're out $360,000. At home, including the data center costs, you're looking at $268,000. The $92,000 delta has to cover the hardware support and maintenance for 3 years. There is some difference in capacity and reliability. I hope it helps your evaluation to say that the cloud offering, in this case, is 35% more expensive.

Click image for single page PDF

In today's post on the CIO.com blog, Michael Hugos raises several interesting points about transitioning IT asset costs from fixed costs to variable costs. He boldly predicts that doing so will herald a rapid economic expansion. Though I'm not willing to go so far as he is, I am a big proponent of emphasizing variable costs. The justification is simple dollars and cents.

Even in small organizations IT asset acquisition is well into the tens or hundreds of thousands of dollars per year.  Dollar for dollar it is hard to beat rolling your own virtualized environment.  Do-it-yourself, if you're able, is the cheapest method. There are several mitigating factors to this:

1) The competition for those budgeted dollars is diverse. Instead of spending $100,000 this year on assets, what if you spent $35,000 on cloud-based offers and moved $65k into a marketing program, new sales reps or other projects that increase top line revenues.

2) The opportunity cost of building and maintaining that system interferes with other IT projects. By migrating routine and time-consuming tasks into optimized environments that are designed for those tasks IT departments can focus on projects that require highly specialized knowledge.

3) New cloud offerings from Terremark, Amazon and 3Tera (among others) are radically altering the scale at which IT exists in small and mid-sized businesses. The cost savings that come from purchasing a small piece of a big pie are significant when compared to your home baked pie.

Many organizations would do well to investigate methods that take IT assets off the books and put those capital dollars into more revenue generating tasks. In any economy that is just good sense.

I needed to find a home for seven wayward servers. The prospective owner had decided to stop owning hardware just as the project to deploy these severs was underway. My job seemed easy enough - replicate a basic Microsoft stack (Exchange, MOSS, SQL, Dynamics GP) for a fixed monthly fee in a land far, far away. I've been closely following the development of hosted/managed/cloud systems for years, have built numerous virtualized environments and I have a very health monthly data center bill. Still the last few days has opened my eyes to the breadth of the cloud's offers. In the end this dynamic market, with a vast pool of options, won over a skeptic like me.

I want to put that last statement into the proper context. I own a value-added-reseller business that has generated millions of dollars in revenue selling hardware and licenses. Last year almost 40% of our revenue was equipment resold from vendors such as HP, Dell, Cisco, Sonicwall and more. I suspect that in the next few years that number will dwindle substantially.

After extensive deliberation I have come up with one singular reason to embrace the 'cloud' that applies to IT departments as well as VARs. We all have better things to do. The time spent procuring, installing and supporting hardware and software detracts from time spent integrating, training and customizing. What if a typical IT project directed a bigger part of the budget at these services? Businesses will derive more value from the technology. The employees will use more of the functionality. The system will work better with the other technology around it. It will more closely match business processes.

Far from competing against the services offered by most IT departments and VARs, cloud computing offers the opportunity to expand these organizations. Those that successfully transition their skill sets can expect to see revenues and budgets increase as businesses receive more consistent performance from existing systems and lower costs to add new systems. This transition will not occur overnight, but in next decade we will see fewer and fewer servers in offices, more servers in the clouds. Think hard about investments in on-premise equipment. There are compelling options out there.

A recent project has forced me to analyze two cloud computing providers - Terremark and Rackspace. In this post I'll discuss my analysis of the two services and use a recently completed Private Cloud project to provide cost comparison.

Terremark's Enterprise Cloud service fit the needs of my client best. This is a fully virtualized environment that completely divorces the client from hardware concerns. The Enterprise Cloud enables rapid deployment of Virtual Machines. All management services from the OS up and managed by the client.  It consists of HP ProLiant servers with a 3Par SAN. Redundant firewalls and load balancers are included in every package. All equipment is shared. Virtualization is performed using VMWare ESX 3.5. Server management occurs through a web application and includes many tasks that you would find in VMWare's management tools. Terremark offers packages at fixed Ghz and GB or RAM (entry is 5Ghz and 10GB RAM). You then add storage, backup and bandwidth. There is no incremental charge per VM, only for the resources it consumes.

Rackspace's Platform Hosting takes a different approach. They supply a mix of dedicated and shared resources to create a similar environment. Firewalls, load balancers and servers are dedicated. SAN resources are shared. Rackspace allows access to the VMware infrastructure tools so all functionality is exposed. Rackspace's price plans vary based on hardware redundancy and resources required.

Some interesting points:

• Terremark offers technical support 12 hours per day. Rackspace's support is 24/7/365.
• Terremark is hardened to withstand a CAT 5 hurricane. Good thing since they are based in Miami.
• Pricing favored Terremark by 25-30% when comparing environments with similar features.
• Network configuration is limited at Terremark. There is a maximum of 4 VLANs.
• Both solutions offer use of existing licenses or their hosted licenses
• Neither were excited about voice applications due to QoS concerns.
• Rackspace does not offer a backup service in the Platform Hosting option. Backup will require a dedicated server and regular offsite downloads.
• Rackspace included 6 TB of download and unlimited upload in their base plan. They claim it was the equivalent of 20Mbps sustained for the entire month.

Honestly I'm afraid I have to end this blog post by changing the name to a braindump. I am beginning a 30 day demo with Terremark immediately. I will continue to assess the two providers and post the information here. Look for follow up posts soon.

When small and midsize businesses find the need to invest in a storage area network solution they are faced with lots of options.  The modular, clustered, iSCSI SAN products offered by EqualLogic (Dell) and LeftHand (HP) have become highly attractive solutions for organizations moving into virtualization that need robust solutions with integrated replication.  But, these are complicated solutions and gathering information to compare these offerings can be a challenge.

The Info Tech research group has posted a great PDF (registration required to view) comparing these two product lines and explaining what make clustered SANs different from the more traditional model.  For even more information, check out the blog equallogicversuslefthand.blogspot.com.

Microsoft has long offered a number of system management products. Recently these products have been grouped into the System Center family of products.   The founding members of the family were SCOM and SCCM.

·         System Center Operations Manager (SCOM) is a highly extensible monitoring system. Because of its expandability SCOM can be configured to monitor an entire environment from network and server hardware to detailed monitoring of applications.

·         System Center Configuration Manager (SCCM), formerly known as SMS, is a powerful system management product capable for end to end management of both servers and client.

Historically these products have been targeted at medium to large environments. While smaller environments could benefit from the enhanced management capabilities, they could rarely justify dedicating servers to host these products or the cost of the server software. However, newer additions to the System Center family combined with virtualization have altered the playing field to a point that smaller environments should re-evaluate the potential of System Center.

Virtualization is probably the biggest game changer for the small environment. On the one hand virtualization gives smaller environments the flexibility to host additional isolated servers, like system center servers, without requiring additional expensive hardware.  The downside to this freedom is the increase in the number of systems that need to be managed.  If virtualization is added to a small environment without additional systems management capabilities the potential for problems and poorly designed environments increases drastically.

Data Protection Manager (DPM) and Virtual Machine Manager (VMM) are the newer members of the System Center family.

·         DPM is Microsoft’s foray into the backup software market. More than just another multipurpose backup solution, DPM is designed by Microsoft specifically to backup Microsoft systems and applications using the new concept of continuous data protection.

·         VMM allows IT professionals enhanced management capabilities and centralized management of virtualized environments large and small. VMM can be combined with SCOM for even more functionality to ensure optimal placement of virtual workloads.

As smaller environments begin to build virtualized infrastructures, DPM and VMM can become powerful, indispensible tools. But the cost of licensing of these products individually would seem to keep them out of reach for these environments.  Microsoft has release a new license called System Center Server Management Suite that addresses this issue, puts System Center within reach of small environments, and will be the focus on the next post in this series.