2008
12.22

Christmas Holidays

I’ll be off the air for a couple of weeks for the holidays.  My occasional online time will be for when I’m on call.  I’ll be back posting on January 5th.

Until then … Happy Christmas (or whatever) and have a safe New Years!

2008
12.18

Today (Dublin, 18th December, 2008) I attended the Web Hosting and Developer Roadshow in Microsoft Ireland. Laptop battery permitting, I’ll be blogging from here as the event passes like I did at TechEd EMEA recently. Dublin is the last leg in this European tour. Cormac Keogh (Architect, Developer Platform Team – MS Ireland) is the host.

The focus of today seems to be on virtualisation, e.g. Hyper-V and Novell partnership and using Windows 2008 for web site hosting. And for our feedback forms we’ll get the USB sticks with the metal covers that MS tend to use. Cool – ReadyBoost media for my laptop J

Arsim Muslija

Arsim Muslija (Business Development Manager, Hosting – Western Europe) takes over to talk about why we should use the MS web hosting platform. Funny: Apple I-Tunes pops up looking for an update over his PowerPoint slides! See, I told you Apple needs updates too.

96% of MS revenue comes from MS partners. The aim of this Roadshow is to interact more with the hosting/web development partners – educate and to listen.

Kevin Turner (COO, MS Global) email Arsim asking about what the plans were for hosting partners in western Europe. They responded with a plan but Turner wanted more. He wanted to increase hosting partner business. This is the realisation of Cloud Computing, SaaS and plans for partners to resell/build services on BPOS. And of course, MS aims to sell S+S (software + services – the rebranding of SaaS) via Azure and BPOS. There is a realisation that on-site installations are the solution of yesterday.

Wow – we’re going to talk about open source. We know MS posts stuff on CodePlex. There is the conspiracy theory that this is to avoid supporting some efforts via PSS. MS claims to have over 5000 projects on CodePlex. This includes a new blogging system from MS that was launched yesterday.

Software + Services is the MS branding of SaaS. Instead of soley relying on on-site installations, you can have completely online solutions or even a hybrid of on-site/online installations. For example, with BPOS, a partner can configure a client’s branch offices to use online hosted Exchange for onsite AD user accounts. These mailboxes would be fully interoperable with onsite Exchange mailboxes in other offices. This means the branch offices don’t need servers, just PC’s with Office. BPOS includes Exchange Sharepoint, Live Meeting and Office Communications. Additional services will add Office Live and Microsoft Dynamics. These will be hosted on the MS cloud in the new generation of MS data centres. However, these will be pretty locked down configurations where data could be located in any country. This is not possible for anyone concerned with Data Protection or the ability to configure/customise. So MS is not a threat to the hosting industry; the partners still need to deploy partner hosted solutions for those clients. There’s also the issue with client access to server solutions, e.g. you’re dealing with long distance latent connections. And of course, MS is new to the hosting industry. Some will feel uncomfortable with dealing with a gigantic company that is new to a service. There will be questions about customer responsiveness … we all know that product design is controlled by the Fortune 500, not the small/medium or even the large business.

There will be a hosting day event focusing on infrastructure in Ireland in 2009.

The MS partner hosting competency was launched this past summer. There will be training on Hyper-V and VMM 2008 that will be run in Spring 2009 for these partners. Lucky enough I won’t need it J

Summary:

  • MS is committed to work with the open source community. Office 2007 SP2 will work with more open source plug-ins.
  • MS wants to offer more options for infrastructure and application platforms.
  • There should be more flexibility in reselling products and services – Hmm… SPLA is far from flexible. In fact, it’s downright difficult and inflexible. Volume licensing is much better.

Microsoft Hosting Platform for Hosting Service Providers

The speaker is Matthew Boettcher from the USA.

The focus appears to be on shared services web hosting, e.g. the €4/month web site for enthusiasts.

Optimise the Infrastructure

  • Windows 2008
  • System Center
  • SQL 2008-12-18 IIS
  • Hyper-V

See Optimised Infrastructure on the MS website.

Easy To Manage

IIS7 is “easier to manage”. The IIS console is much different to IIS6. Everything is in one console, e.g. ASP.Net and IIS. Under the hood everything is stored in XML. The metabase is gone. That’s good news if you didn’t know already. Features of IIS can be delegated.

You can remotely manage IIS sites via HTTPS via the IIS admin console – enable the role service “Managed Service” (not on by default). In theory, provisioning of websites could be done using PowerShell and you could do away with the traditional web hoster’s control panel application. The catch is in providing a customer installation of the IIS7 console. You’d need PowerShell 2.0 or maybe W2008 R2 for this … we know new CMDLETS are on the way for IIS7. You can do all this using APPCMD.

Windows Web Server 2008

This allows SQL Server full install. It includes IIS7, WSS 3.0, .Net 3.0/3.5 and Windows Media Servers 2008. Allows up to 4 CPU’s and 4GbRAM (x86) or 32GB RAM (x64).

SQL Server 2008 Web Edition

  • Adds new PHP driver.
  • Up to 4 CPU’s.
  • No RAM or DB size limitation.
  • Has log shipping
  • There’s a database publishing wizard
  • There is policy based management
  • Performance Studio

Modular Architecture

IIS is modular and integrated. It’s a total rewrite. Only the components you need are installed. It also is more efficient with integrated components like PHP, e.g. FastCGI.

PHP on IIS7 with FastCGI is exponentially faster. Configurable caching of dynamic content increases this performance. Add in tracing and it’s a better platform than IIS6 for the programmer and web admin. This also includes Ruby and PERL.

MS recommends 1 web application per application pool. Web hosters won’t do that because there’s too much of a memory charge which would decrease price competitiveness (critical in this low end industry).

Media Services

Save bandwidth using bit rate throttling. This recognises the client/server connection speed and paces the delivery of content, e.g. just in time delivery. Broken connections are detected. This avoids wasting bandwidth.

Use server-side web playlists to hide URL’s, block skip/seek and integrate with web applications. There is an IIS7 Media Pack that adds web playlists (a .ISX file).

Smooth Streaming is new to IIS7. It’s a collaboration of IIS Media Pack, Expression Encoder and the Silverlight teams. It dynamically recognises the connection speed and adapts the picture quality to provide a smooth video instead of the “watch a bit, buffer, watch a bit, buffer” experience. There will be a public beta of this available by February 2009.

Web Farms

You can configure a master server for your XML configuration. The web servers go to this on location (clustered hopefully) to load their configuration. Now you have one configuration to set up and maintain.

Lower Cost of Infrastructure

Server Core has lower requirements … however W2008 Core Installation does not have .NET support. Wait until W2008 R2 for a stripped down version of .NET, probably 3.0 and/or 3.5.

We can run media services on a Core installation.

Security

  • BuiltinIUSR replaces BUiltinIUSR_<MachineName> … needed because the default machine name on W2008 is a huge long random-like string.
  • IUSRS replaces IIS_WPG.
  • SID injection for IIS_IUSRS group is optional.
  • You can use an application pool identity instead of anonymous.
  • Application pool isolation is on by default.

Operational Guidance

Learn.iis.net website for lots of IIS information.

There is the Hosting Deployment Accelerator kit, version 2.0. This is based on the RDP program of W2008.

Better Market Offerings and Optimised Infrastructure Due To Microsoft Virtualisation

Jan H. Haul, Expert Consultant, DNV IT Global Services.

Why Virtualisation?

Pain points for hosters:

  • Credit crunch
  • Staff
  • Uncertainty
  • Complexity
  • Drivers
  • Energy costs
  • Set-up speed
  • Licensing
  • System Management

Data centres in the world products 170m tonnes of CO2 per year. Netherlands (industrialised but green country) produces 146m tonnes per year.

Easier and better management.

Customer satisfaction:

  • Self service portal for deployment
  • Easy to clone, easy to revert to a “known good” state.

Those two sound nice, eh? The first is not possible on Hyper-V clusters. There’s an assumption that the user should know and understand the server infrastructure.  Driver isolation, i.e. the hardware is abstracted.  Centralised server administration.  Clustering of the hosts allows hardware/server fault tolerance.  And of course, Hyper-V is the virtualisation role of Windows 2008. Requires 64 bit hardware, DEP and CPU virtualisation enabled in the BIOS.

OK, this guy has made no mention of DEP. He also has refused to accept question so I can point this out. Some manufacturers hide DEP in BIOS so you cannot enable it. This prevents Hyper-V from installing/working because the VMBus/hypervisor rely on DEP for inter virtual machine security.  I’m not blogging this guy’s session any more. It’s a waste of wear and tear on the keys on my laptop.

Microsoft / Novell Partnership

Michael Croney is a sales director from MS UK.

Agreement signed on Nov 2nd (as you will have seen on my blog back then!) that was co-announced by both companies.

The components:

  • Agreement not to sue, e.g. the Linux IP issue.
  • Technical collaboration on virtualisation (Hyper-V IC’s for Suse Enterprise) and Systems Management (OpsMgr 2007 R2 Cross Platform Extensions).
  • Joint sales campaign, e.g. MS can sell SUSE Linux at a reduced cost.

MS doing this because 70% of their customers run Windows and Linux in the same data centre. They want interoperability.  There is Novell Moonlight: an open source implementation of Silverlight as a plug-in for Firefox.  Advanced management pack developed by Linux for SUSE Enterprise Server for OpsMgr 2007 R2.  “Open Server” will integrate with Active Directory. This was a more recent announcement.  Novell/MS are giving big discounts to other Linux users. Novell is providing “Co-support” for 2 years if migrating from RedHat to SUSE.  We then got a marketing video to watch *yawn*.

The Afternoon

Martha Rotter talked about Silverlight and gave plenty of good demo’s.  I was then away to meet someone in MS to plan some user group stuff for next year and that was the end of my session for the day.

2008
12.17

An Interesting Quote

A sales man of a firm that runs it’s hosting service out of a computer room here in Dublin was quoted by a local IT paper.  He was talking about Software-as-a-Service (Saas) and the importance of availability:

"If the software being delivered is key to the business using it and has real value, then like everything else that is web-based, the critical term is availability".

I nearly choked on that one.  The company is question has had at least two major outages this year … one related to networking failure and the other related to power failure.  Of course, they’re "staffed by an extensive team of fully Cisco certified staff, ranging from CCNA. CCSP, CCNA and CCIE on a 24/7 basis, we can react to threats and issues as they arise".  Funny, their senior network engineer (who is a good guy) isn’t a CCIE yet and he certainly doesn’t work 24 * 7.  And I recommend the 2am test if you’re looking at them … knock on the door unannounced and randomly at 2am and see who, if anyone, answers the door.

By the way, they’re still at 99.87% availability according to third party metrics, a far cry from the much desired 99.999% or even perfect 100%.

EDIT:

I met one of the other salesmen from this hosting company today.  I mentioned the CCIE claim and asked about that engineer becoming one.  "Oh yeah", blah-blah-blah was the response, confirming the claim on the website.  Strange, because the engineer in question hasn’t actually passed the CCIE exam yet!  Don’t ya just love this sort of sell-at-any-cost (to the customer) behaviour?

2008
12.17

If you’re in Ireland and interested in SharePoint then you should check out the SharePoint User Group that is run by MVP (another one of the magic 8), Stephen Cummins.

2008
12.16

Apple just released a patch bundle for the Macintosh.  Get this, it weighs in at 372MB!!!  That’s not a service pack, it’s just patches.  I hope all you Mac freaks have your centrally managed security update systems ready for downloading (once), testing and deploying those updates!

2008
12.16

HP Server P2V

One of the most annoying parts of doing a P2V of a HP Server for me was removing the HP SIM components.  I’ve just found a scripted solution on the HP forums that takes care of this process for you.  Apparently HP sells some software for the task … but why pay for something that someone has kindly provided a free script for?

Credit: Nelson Kaeppel.

2008
12.16

MS PFE Ireland blogged about how to configure a cluster within a Hyper-V environment, i.e. how to set up 2 virtual machines running as a cluster – not how to create a Hyper-V cluster.  They used iSCSI to configure the shared storage to connect the virtual hosts, i.e. for the quorum and the shared service installation.

Setting up a Hyper-V cluster is another story altogether.  Here’s the short version.  The hosts should have a minimum of 3 NIC’s:

  • 1 for the parent partition (on the host network)
  • 1 for the cluster (on a private heartbeat network)
  • 1 for the virtual machines (usually on a network trunk and with TCP unbound).

You’ll probably go for 4 NIC’s.  However, the virtual machine network NIC’s can’t be teamed at the moment.  MS is working with OEM’s to resolve this.  Make sure the h/w is on the cluster compatibility list.  Avoid installing any OEM server network configuration tools until there is support from them for Hyper-V.

Build the machines with Windows Server 2008 Enterprise Edition.  You can go with Core if you want but I prefer to use Full installation – the OEM’s haven’t given us accessible ways to manage their hardware from Core yet (if they ever will).  Apply security patches.  Ensure the hosts are identical.

Set up your shared storage, either iSCSI or fibre channel.  If using an even number of hosts you’ll need to configure a small quorum disk or "witness disk".  If uneven then you don’t.  However I set one up anyway in case I need to remove or add a host – VMM will scream loudly if the cluster is not in a supported state, i.e. failover will be unpredictable.

For the virtual network, you’ll need to find out what workarounds (if any) are necessary for getting the VLAN trunk (VLAN tagging) working.  See my previous posts about the NC373i and Intel NIC’s.

Configure Hyper-V and add in the patches for GUID drives and VMM compatibility (see my previous posts).  Install the clustering service as well.  Configure the firewall to allow remote administration.

Build the cluster.  If using an even number of nodes you’ll go with node majority.  If uneven then it’s disk (the witness/quorum) and node majority.  Run the cluster verification report and save the results in case PSS ask for it.  If using an EVA SAN you must remember to set the node type to Longhorn (the codename for W2008) to pass the test.

Now you can configure your virtual networks – do it in the Hyper-V admin console if you don’t have VMM 2008.  Do it in VMM 2008 if you are using it to manage the Hyper-V cluster – you need to import the cluster into VMM 2008 management first :)

Set up your VM’s and test them.  Your VM’s will be on their own LUN’s provided by the SAN.  Provision the disk and present it to all your hosts.  Configure the disk (bring online, letterless/GUID and format) on the first host in the cluster.  Add the disk to the cluster.  Note the GUID for VM deployment.  Use your administration console (Hyper-V or VMM 2008) to deploy your VM.

In VMM you should be aiming to see a healthy state.  If you’re told the cluster configuration isn’t supported then check your host virtual network, cluster verification report and VM storage configurations.

There.  That’s probably a day’s worth of training in one short blog post.  Obviously it’s a bit more detailed than all that but it’s a dump on what you should look at.

2008
12.16

Walter Chomak wrote a good article to help people avoid a gotchya when monitoring un-trusted servers over the Internet using Operations Manager 2007 and the Gateway.  This applies equally when using OpsMgr and agents with certificates.  You need to be sure that the cert is for the actual FQDN of the server and that OpsMgr addresses the agent by the FQDN of the agent server.  Set up name resolution using DNS or hosts files.

Credit: Walter Chomak.

2008
12.16

VMware’s share value continues it’s downward trend of the last year as the competition heats up.  Entrants into the market such as Microsoft (Hyper-V is a tick box away from installation) and Citrix (with the mature XenServer) have put on the pressure in terms of sales in recent months.  You could try the excuse that IT spending is down – sure but people are reducing costs by virtualising their server farms.  Now, Intel is dumping a large amount of shares.

In December 2007, a share was worth $89.85.  Today it’s worth $24.74.  Still, they’re doing better than the banks!

Credit: Virtualization.Info

2008
12.16

The big thing in the server hosting industry right now is virtualisation.  One of the buzz words is VPS or Virtual Private Server.  It’s a low end offering where a virtual machine is provisioned, quite probably on non-clustered hardware, e.g. if the host dies then every VM hosted on it goes down too.  This keeps the hosting costs low and makes the VPS machine very cheap.  You could think of VPS as being an enthusiasts solution.

There is a big market for this.  If you’re in the web site hosting business you’re familiar with using a Control Panel.  This is the system that is the interface for your customers to the hosting system.  DotNetPanel have just released their Control Panel for Hyper-V based VPS.  This allows VPS hosting companies that are using Hyper-V to sell virtual machines to customers and give the customers a web based interface over the Internet.

This looks like a great solution for selling to enthusiasts.  However, I don’t see it as a professional solution.  For me a virtual machine should be treated exactly like a physical machine (even if deployment is slightly different, e.g. different methods and quicker).  Business machines should not be cut from cookie cutters.  They should be on private networks protected by firewalls.  Firewalls aren’t one-size-fits-all.  Do you need TCP 443 open, TCP 3389, etc?  A limitation with this release of Hyper-V is that each VM on a cluster requires it’s own LUN – yes you can put lots of VM’s on a single LUN but that’s not a flexible cluster solution.  We’re waiting on Windows 2008 R2 for the cluster file system to make life easier for this sort of thing.  So VPS on Hyper-V means no host clustering in reality. 

Finally, in this sort of methodology, you cannot sell SPLA (leased by the month) licensing legally.  As a SPLA reseller, you must ensure that the correct types of licenses are being used by your customers.  For example, we have anonymous and authenticated Per CPU Windows Server licenses.  Per CPU licensing is used instead of user CAL’s.  Anonymous is fine where you have a dumb web server where Windows does not authenticate the users.  Authenticated licensing is required where Windows does authenticate the user, e.g. when a SharePoint site asks a user to log on the log on attempt by the user uses a Windows user account specific to that user.  Authenticated per CPU licenses are significantly more expensive than Anonymous.  I’ve noticed some hosting companies offer no explanation of this legal requirement and skirt around it to offer their customers the cheapest license available, despite it being illegal.  VPS sales automation places the responsibility on the customer who will not understand this so they’ll always go for the cheap option.

And don’t even get me started on people selling Windows Standard/Enterprise Anonymous licenses which are no longer available from Microsoft.  A: It’s wrong for these hosting companies to do this.  B: It was wrong of MS to pull these SKU’s from the list of SPLA licenses.

2008
12.16

MS has released SCUP 4.5 for publishing your own updates via ConfigMgr 2007/2007 R2 or SCE 2007.

2008
12.16

For all you accidental DBA’s out there, MS has released a feature pack for SQL 2005.  This contains a collection of tools and functionality add-on’s that you could alternatively download separately.

2008
12.15

Ever heard the term "MVP" and wondered what the heck they are (outside of American Sports, that is).  Microsoft has just set up a blog to help publicise their Most Valuable Professionals.  In short, an MVP is an award given to worldwide experts who share their knowledge with others.  This can be through writing, public speaking, forums, etc.  MVP’s are re-evaluated on an annual basis so MVP’s need to keep up their efforts and expertise. 

The next question is, how do you become one?  The answer is: get noticed.  Focus on one product and work on it.  Share your knowledge.  Eventually either someone inside of MS or another MVP will nominate you.  You’ll then be asked to participate in building a case which will be presented to the product group.  They decide yea or nay on your case.  If you’re lucky, you get a nice email.  There’s no exams to cram for.  This is based purely on your efforts in the community.

How many of there are us?  It seems to be split 50/50 between developers and IT pros.  There’s just over 4,000 MVP’s globally.  And there’s just 8 of us in Ireland.

2008
12.15

If you’re installing Hyper-V Server 2008 (the free stripped down OS with Hyper-V) then make sure you have configured DEP and CPU virtualisation assistance in the BIOS beforehand.  The OS will install and setup the Hyper-V role even if you haven’t configured the server hardware.  Obviously Hyper-V won’t work without these being configured.

And one thing is missing from the configuration menu: the firewall.  I’ve set up a machine for some lab work and remote admin via the Hyper-V console proved impossible until I reconfigured the Windows Firewall on the Hyper-V server.

I installed the VMM 2008 agent successfully.  The "update available" issue sprang up immediately.  Installing KB956589 and KB956774 on the Hyper-V server fixed that (reboot required).  That wasn’t much of a problem because I’ve previously encountered it.  I wonder if you can build a WDS image from the Hyper-V Server media and stream these updates?  Whenever I have free time (2011 by the looks of it) I’ll have a look.

2008
12.12

The Hits Keep Coming

My blog passed 125,000 hits this week.  Thank you for the interest!

2008
12.12

I’ve been trying out a few things, trying to push the envelope of some VM configurations to see what happens.  I’d set up a VM in VMM and forgot to mount a installation DVD image.  The machine fired up and booted up the PXE client.  It found my WDS server and was ready for me to log in when I got back to it.

A quick rummage and I found that the NIC in the default new machine template was the Legacy Adapter rather than the integration components enabled one.  So:

  1. When creating IC enabled VM’s, be aware of this.  A IC enabled NIC is more resource efficient than the Legacy Adapter.  If you mainly deal with WXP, Vista, W2003 or W2008 then have a machine template that has the Integration Component/Services NIC rather than the Legacy Adapter.
  2. This is good for heterogeneous environments where your Linux is Xen-enabled but doesn’t have IC components.
2008
12.12

It’s been a while since I’ve written any documents for my blog.  I was asked to do some stuff for the company web site so I’ve written a couple of documents.  The first is on the Microsoft Virtualisation Strategy.  I talk about how MS views virtualisation as being many things, e.g. Hyper-V/server virtualisation is just one aspect of the solution.  I also discuss what makes the MS approach different: management.

On a related note, there’s also a document on Cloud Computing and Software-as-a-Service.  Organisations are tired of becoming accidental IT firms, e.g. they buy a CRM solution and end up managing servers, racks, power, UPS, fire suppression, databases, networking, etc.  They just wanted a business solution.  SaaS gives them that solution.  It’s also good for the software developers/server providers because it simplifies product ownership and increases customer retention.  Cloud Computing is the delivery mechanism for SaaS.  And virtualisation is a core component of Cloud Computing because it breaks the link between services and hardware/geographic location.  Microsoft’s managed virtualisation gives us Managed Cloud Computing thanks to the all-seeing and all-knowing single pane of glass, Microsoft System Center.

2008
12.11

I’ve wondered this myself.  Afterall, isn’t i just an empty file?  The Virtual PC guy explains how and why MS ensures that it really is an empty file.

2008
12.09

One of my colleagues told me to look at www.microsoft.ie.  I did and I took a screen shot:

MsIEhacked

That, on the face of it, would look like Microsoft were hacked and someone had defaced the Irish site.  I checked the genuine MS Ireland URL and it was OK.  A quick lookup on DNSTools and I found this:

% Information related to ’80.93.17.0 – 80.93.17.255′
inetnum:        80.93.17.0 – 80.93.17.255
netname:        nov-sh
descr:          Novara Shared Hosting
country:        IE
admin-c:        nov23-ripe
tech-c:         nov23-ripe
status:         ASSIGNED PA
mnt-by:         mnt-novara32
mnt-lower:      mnt-novara32
mnt-routes:     mnt-novara32
source:         RIPE # Filtered
person:         Eoin Costello
address:        3, North Earl Street Dublin 1, Ireland
phone:          +35318583091
nic-hdl:        nov23-ripe
source:         RIPE # Filtered
% Information related to ’80.93.16.0/20AS31122′
route:          80.93.16.0/20
descr:          Novara Route Object
origin:         AS31122
mnt-by:         DIGIWEB-MNT
source:         RIPE # Filtered

Novara was acquired by Digiweb a while ago.  It looks like someone set up a DNS record to point to a site hosted on their shared service web servers.  Ouch!

EDIT:

This looks like a DNS hack was perpetrated on Digiweb.  I cannot say for definite but that’s what it smells like to me.  It looks like stuff that was 100% outside of MS’s control.

EDIT #2:

And for the twits wearing tinfoil hats: No, the Microsoft Ireland site was not actually defaced.  The .ie DNS record just redirects to the Ireland subpages of corporate.  That record (it looks as if it was Novara hosted but I could be wrong) was altered and a fake page on a Novara/Digiweb server was set up.

2008
12.09

Distributed Power Management

Most administrators don’t know or care about the real cost of servers: power.  A single server’s cost is much more than what you pay to Dell or HP.  The power alone massively outweighs the purchase cost.  It’s said a typical server has the carbon footprint of a car.  It’s inevitable that we’re going to see carbon taxes hitting businesses now.  Cloud computing/Software-as-a-Service mightn’t be for everyone so they need a solution.  Cloud providers also need a solution to power issues because the biggest cost they have to pass on to customers is electricity.

I found this commentary by Chris Wolf talking about an experimental feature that was included in VMware VI3.5.  This feature called Distributed Power Management (DPM) is an interesting one – one which had me nearly swinging towards VMware instead of Hyper-V.  Virtual Center monitors the usage of host resources by VM’s and using DRA and memory over-subscription it will consolidate VM’s to fewer hosts.  This allows idle hosts to be powered down or suspended.  When resource consumption grows the required idle hosts are powered back up using WOL.  VM’s can be migrated using VMotion to ensure they get the CPU and RAM (probably IO as well) resources that they need.

The commentary talks about how people are wary of powering down/up production servers.  That’s fair enough.  In my opinion however, that’s the wrong way to look at this.  The production servers are the VM’s.  In this scenario the VM’s are never powered down.  They’re offline for a few milliseconds as the VMotion across the cluster, something that VMware customers are well used to now. 

The hosts are just physical resources.  The hardware is just an enabling layer like electricity or network when you’re dealing with virtualisation.  And just like those utilities there’s fault tolerance at this layer – or there should be.  In a network that could realistically use DPM to save power there will be significant numbers of hosts.  They should be dealing with at least N+1 the number of hosts that they require, maybe even N+2.  So what happens if there’s an occasional hardware failure?  If you run an enterprise network then the hardware should be monitored and any faults will be responded to immediately.

Microsoft are currently taking a different approach to the power issue when it comes to Windows Server 2008 R2 – and logically Hyper-V.  MS are using Core and CPU Parking.  The server monitors the demand on the CPU cores every X milliseconds.  When a core is idle it is suspended, thus reducing it’s power consumption.   The CPU core is the major draw on power in a server.  It’s also the generator of heat and cooling that heat is another major draw on power.  Suspending idle Cores reduces both of those power demands.  If a Core is required then it is snapped back online.  The trick is in defining appropriate idle windows – you don’t want to suspend at millisecond 1 and find you’re always bringing it back online at millisecond 2.  That’s wasteful.  When all cores in a CPU are idle then the CPU is parked, thus saving more power.

I was at a power meeting/interview session with MS at TechEd EMEA and I brought up the VMware DPM approach.  I don’t know if it’s something MS will look at or not.  I hope they do look at it for the next release after Windows Server 2008 R2.  Right now, I have to applaud VMware for trying to do something.  They do see the hardware as just an enabling layer, not the production servers.  I think that’s the right point of view to take.  When DPM does go live I can see it saving VMware customers a good bit of money.

2008
12.08

Hyper-V How To: Balance VM I/O

This article explains how to balance I/O between VM’s on hosts that are saturated.  It’s a last resort action to resolve the issue.  Ideally you’ll be balancing your workload across a cluster, e.g. OpsMgr detects a peak load, your storage PRO tips detect the culprit and VMM 2008 balances the workload across the Hyper-V cluster.  However, that won’t work if you don’t have a cluster so using these registry edits might be necessary.

2008
12.08

Terminal Services licensing is confusing for many people.  This document by Microsoft attempts to explain it for Windows Server 2008 Terminal Services.

2008
12.08

You can now download a beta release (for testing only) of Service Pack 2 for Windows Vista and Windows Server 2008 (x86 and x64).

2008
12.05

Lots of Hyper-V Updates

Jose Barreto has listed all of the publicly available updates for Hyper-V including the RTM.  You could selectively apply them if they are applicable to your servers.  For future builds, here’s what I’m going to do:

  • I’ve got a WDS captured WIM image of my Hyper-V build.  This has everything done before enabling the Hyper-V role.  It includes the RTM release of Hyper-V.
  • I’ll download the updates.
  • I’ll slipstream the updates to my WIM build and to the WIM’s from the installation media that are on my WDS server.  That future proofs any new builds.

The result will be that any newly built Hyper-V hosts will have all of the updates in place.

2008
12.05

What’s an SLA Worth?

Imagine a data centre service provider who offers a 100% availability SLA.  That’s pretty impressive.  Most of us aim for the five 9′s, e.g. 99.999%.  99.9% is even good with no more than 60 minutes outage in a month.  You’d have to be pretty sure of yourself to offer a 100% SLA.

For a data centre, that is actually achievable but you cannot cut any corners.  I’m lucky enough to work with a service provider who can live up to their claim of a 100% SLA.  They invested heavily in the building, people and processes to have a Tier IV facility with no single points of failure.  They haven’t had an outage since they launched in 2001.  As a result their customers can build their brand name on this.  We’re able to say that our service extends their philosophy and our service hasn’t had an outage since we launched earlier this year.

But I know of others who do claim a 100% SLA.  In fact, one of them is having their 4th major outage in 2.5 years …. right now as I type this – if you’re in Ireland it’s not hard to guess who I’m talking about.  I’m not even counting the little mishaps that they have on a recurring basis … some we know about through web forums, blogs, word of mouth, etc.  I’m not going to poke fun at them.  There’s some good people in there who’ll be stressed out right now through no fault of theirs.

However, we do have to look at the people responsible for the SLA being offered.  Their clients are depending on that SLA.  They’ve reflected that to their customers, e.g. if my hosting provider gives me a 99.99% SLA then I can pass that on to my clients.  However, if my data centre is up and down like the proverbial w***e’s knickers then I look like mud to my clients, whether they’re internal or external clients.

If you’re looking at a hosting service provider then please check out their SLA. If they claim 100% then that’s very audacious.  I’m not saying it’s impossible, just very hard.  Look at their track record and see if they live up to it.  If not, then can you believe other fantastic claims about senior staff on site 24*7, huge bandwidth, "everything is possible", etc.  Here’s what I’d do:

  • Ask for the SLA.  Check out blogs.  Web masters are always quick to point out faults so their forums are a good place to check.  If they have a status site then check it.  See if the explanation for a fault stays consistent.  If not, don’t deal with them.
  • The 2am Test: Drive up to the data centre and knock on the door.  If anyone is even in, ask to speak to the senior staff the company claims is on site 24*7.  If they lie, walk away from the sales negotiation.
  • Ask for proof of certification claims, e.g. I’m an MCSE.  I have an ID number that people can check out for proof of my claim.  I’d say the same applies for CCIE’s.  If they lie, walk out.
  • The real kicker: ask them if they do XYZ.  Let XYZ = some thing you’ve just made up off of the top of your head.  If they say yes then walk out.

Am I sounding harsh?  Honestly, no.  We all expect sales people to stretch a little.  But taking things this far is too much.  Imagine if this stings you?  It shuts down your business.  If you’re a reseller your clients will blame you, not the hosting company.  You’ve got a business reputation to maintain.  If you’re in the mission critical world then outages such as this are not tolerable.  They can possible kill people.

If you’ve suffered a hosting power outage again in this 365 day period and it’s affected your business then check out an alternative.

EDIT:

The hosting company having the outage finally came back online after 2.5 hours – sort of.  Some servers are still not responding.  Assuming this is their only outage this year (and it wasn’t) that’d give them a 99.97% uptime.

EDIT #2:

I just checked that company’s web site uptime (they host it themselves in their data centre).  It’s available 99.87% of the time over the last 2 years and they had 97.74% uptime in September 2008.  Not quite 100% or even 99.9%.

Get Adobe Flash player