Windows 2008 Server: An Early Look

What is Hyper-V?

2008-09-06T09:33:29Z

Hyper-V, Microsoft’s newest entry into the virtualization arena, is built on the Windows 2008 Server platform. Hyper-V is installed as an add-on role within the 64-bit version of Windows 2008 Enterprise and Database Center editions. It allows for the creation of multiple child partitions (VM's) to run simultaneously on a single, physical server. Each child partition is capable of running a different operating system.

Child partitions are given the illusion that they are exclusively sit on-top-of the physical hardware. This is accomplished by taking advantage of hardware assisted virtualization. This hardware assisted virtualization allows the child partition to access the CPU and memory without having to first go through the parent partition. (The parent partition is the operating system installed directly onto the physical server). To access the CPU or memory both parent and child go through the same 300Kb hypervisor. Placing the hardware closer to the child partition is the secret to Hyper-V performance gains.

Only certain operating systems are able to take full advantage of the hypervisor. These operating systems are referred to as Enlightened OS’s. An example of difference between enlightened and non-enlightened operating systems would be how they access disk resources. An enlightened operating system can have physical disks directly “passed-through” from the parent partition without having to involve the rest of its disk model. Alternatively, an operating system without the enlightenments, attempting to access disk resources must pass through the full disk model of the parent partition. This additional (emulation) overhead substantially reduces throughput.

To date the following operating systems are enlightened:

Windows Server 2008
Windows Server 2003 SP2
Windows Server 2000 SP4
Windows HPC Server 2008
SUSE Linux Enterprise Server 10 SP1/SP2
Windows Vista SP1 (except home editions)
Windows XP SP2/SP3

Oracle 10g Optimizer Behavior for JDE EnterpriseOne

2008-07-15T09:01:13Z

by Roger Carter, DB Connect Solutions

Background

As of Oracle 10g the Rule-Based Optimizer (RBO) has moved to an unsupported state. Although the RBO is not completely deprecated as of version 10g, its days are clearly numbered and there is one and only one path forward – the Cost-Based Optimizer (CBO).

As many battle-weary Oracle DBAs can attest, the move toward the CBO has been a long and painful process for many applications and application vendors. On the one hand the traditional RBO was extremely naïve in its approach to SQL query optimization, yet it yielded very stable and predictable results. Conversely the CBO had intricate algorithms written and rewritten over the years with the end goal being an implied silver bullet toward SQL query optimization. The implicit promise from the propeller brains at Oracle was an internal engine that could and would make intelligent decisions based on data distribution, data volume, and even the speed of the underlying hardware subsystems.

Nice goal indeed.

What many application vendors struggled with over these intervening years was that despite the advertisements to the contrary, Oracle’s CBO was inherently unstable. For many good reasons that went largely uncommunicated to us mere mortals who actually supported production systems, Oracle struggled internally with exactly how to make the damned thing work as advertised.

Think about it -- How could any algorithm yield perfection (or near perfection) when it wasn’t based on a set of explicitly-defined rules such as those present in the traditional RBO? True, it might yield fabulous results 98% of the time, but those other 2% -- wow could they be painful, if not excruciatingly so! It didn’t matter that an ERP system using the CBO ran perfectly most of the time. What mattered is that ERP systems relying on the CBO didn’t perform well some times!

Many ERP vendors resisted the push toward the CBO for this very reason. It was better to adjust to “acceptable” performance using the RBO approach when results were predictable; it was not sufficient to accept even minimal performance problems using the CBO approach. As all DBAs know quite well, end users don’t particularly like to wait 36-48 hours for month-end reports, yet they can adjust to it if the results are predictable. The worst possible scenario, one which leads normally slack-jawed quiet users to mutate into homicidal DBA killers, is a set of standard reports that run in 5 minutes one month and 5 hours the next month. When the DBA says “But the database is fine!” that’s um, sort of a lie. And it makes excuses for instability introduced by the CBO.

Yet the move toward the CBO wasn’t and still isn’t a matter of debate. It’s sort of like a Microsoft edict with a database twist from Sir Larry Ellison himself:

Thou shalt embrace the Cost-Based Optimizer or die trying!

This author is fairly certain a number of DBAs have gone ahead and died, perhaps by their own Oracle OCP hands, when they couldn’t get the cost-based optimizer to well, optimize. The purpose of this whitepaper is to help prevent the same fate for a few of my JDE DBA comrades. For the non-JDE DBAs, well, you are what you eat I guess.

Cost-Based Optimizer: Oracle 9i per chance to 10g?

In anticipation of the move to a universal CBO paradigm, most major packages, including JDE EnterpriseOne, have transitioned to the Cost-Based Optimizer (CBO) as of Oracle 9i. This required that Oracle statistics be kept up-to-date so the CBO could produce optimal (and consistent) execution paths. If the CBO did not have accurate statistics, performance would suffer and the JDE DBA would have to go into hiding or do a little work.

This meant the DBA had to run scripts to periodically refresh statistics. The problem with this process is that the statistics gathering process was typically an “all-or-nothing” approach, meaning that statistics were refreshed for tables whether they needed to be refreshed or not. Some DBAs took advantage of a new feature in 9i called table monitoring that allowed the Oracle database to track which table statistics needed to be refreshed, but this required even more scripting to accomplish. More work (job security) for the Oracle DBA.

As of Oracle 10g, however, the process of identifying which statistics to refresh as well as the statistics refresh process itself, is completely automated (as long as the init.ora parameter STATISTICS_LEVEL is set to either TYPICAL or ALL). The internal Oracle Scheduler comes predefined with a job named GATHER_STATS_JOB which automatically refreshes statistics on tables whose statistics have gone stale. Less work (job insecurity) for the Oracle DBA.

The CBO in Oracle 9i was a small step forward in the new optimization paradigm because the default behavior of statistics collection was not to create histograms (a mechanism for profiling the data distribution within a table). This meant that the CBO in Oracle 9i was a little like the RBO with a twist of CBO thrown in.

But Oracle 10g became a CBO leap of faith. As of Oracle 10g, the default behavior is to collect histograms when table statistics are generated. (This is controlled by the method_opt parameter of the statistics collection mechanism.) Histograms provide the CBO with more detailed information about the data distribution within the columns of data in each table, and therefore allow the CBO to make an even more thorough analysis regarding the proper execution path for a particular query.

However, therein lies the CBO rub. When the CBO has detailed data distribution statistics available, it can do really stupid things to JDE Oracle queries.

This is because the presence of histograms can lead to performance instability for certain queries that use bind variables against columns containing skewed data. To understand this anomaly and to formulate a strategy to address it, we must delve deeper into the CBO and understand the concept of bind variable peeking.

Peekaboo! Your Bind Variables Have Been Peeked

In Oracle 9i the CBO introduced the concept of bind peeking. This is essentially a step performed during the hard parse of a SQL statement. (A hard parse occurs when the SQL statement does not exist in the shared pool.)

When parsing the SQL statement, the CBO “peeks” at the actual values of the bind variables as the SQL statement is loaded into the shared pool. The CBO uses these actual bind variable values to make a decision on the execution path to utilize for the statement. If the values within the column(s) corresponding to the bind variables are uniformly distributed, the result of the bind peeking doesn’t particularly matter regardless of the actual bind variable value. (Think of a statement such as SELECT * FROM EMPLOYEE WHERE EMPID = :1; the CBO will always use the primary key index on the EMPID column regardless of which employee id is used in the first execution of the statement.)

However, if the values within the column(s) corresponding to the bind variables are not uniformly distributed, the result of the bind peeking can lead to interesting and sometimes unexpected results. (Think of a statement such as SELECT * FROM EMPLOYEE WHERE STATE_CD = :1 where nearly all employees live in New Jersey. If there’s an index on STATE_CD and the very first execution of the statement is specified as SELECT * FROM EMPLOYEE WHERE STATE_CD = ‘AK’ the bind peeking would lead the CBO to choose the index, and the result would be an efficient execution for the first execution. After the first execution, the SQL statement and the execution path are now cached in the shared pool. If a subsequent execution of the statement is specified as SELECT * FROM EMPLOYEE WHERE STATE_CD = ‘NJ’ the index would be used again, even though a full table scan would be much more efficient.)

Within JDE, this bind peeking performance anomaly is seen quite frequently. It is typically seen on large tables, especially the F0911, F42119, and sometimes the F4211. It can impact UBEs, especially jobs that process the entire table a set of data at a time (for instance, processing the entire F0911 by a column that has a relatively small set of values). It can also be seen in interactive queries depending on the search criteria and the specific values chosen by the end user. At times it can seem to “come and go” depending on how long the offending SQL statement is cached in the shared pool. (This depends on the size of the shared pool, how frequently the SQL statement is run, and how much dynamic/un-shareable SQL is issued to the database.)

This issue was not as prevalent in Oracle 9i because the creation of histograms was not the default behavior. Because the CBO did not have detailed knowledge of data distribution in 9i, the risk of bind peeking anomalies was much lower. At the same time, the CBO was not as intelligent in Oracle 9i.

Recommendations for running JDE with the CBO

Do not turn off the Cost-Based Optimizer because you just can’t fight City Hall.
Although it is possible to disable bind peeking with a hidden init.ora parameter, this is not yet recommended. There may be some future guidance on this issue from Oracle or other know-it-alls like this author.
Do not turn off histograms collection across the board. Overall, the presence of histograms helps more than harms performance. Individual performance anomalies must be addressed instead.
It may be beneficial, however, to test the removal of histograms on individual tables, for example the F4211 and F42119. (If histograms are removed manually they will be subsequently recreated via the GATHER_STATS_JOB unless this is overridden via exec dbms_stats.set_param(‘method_opt’,’FOR ALL COLUMNS SIZE REPEAT’
Use the default GATHER_STATS_JOB provided in Oracle 10g. However, do not use the default sample size of DBMS_STATS.AUTO_SAMPLE_SIZE which can lead to poor histograms on skewed columns. It is strongly recommended that at least 30% be used for the default sample size. This can be achieved by setting the default via the command exec dbms_stats.set_param('estimate_percent',30)
In Oracle 10g the recalculation of statistics using DBMS_STATS will no longer automatically invalidate dependent cursors as happened in Oracle 9i. (This is because the no_invalidate parameter of DBMS_STATS defaults to DBMS_STATS.AUTO_INVALIDATE.) This means that Oracle will decide when to invalidate cursors so this behavior must be kept in mind when trying to force a new execution plan by recalculating statistics.
Because statistics collection is drastically different in Oracle 10g versus Oracle 9i it may be beneficial to transition to the new statistics mechanisms after upgrading to Oracle 10g. This is discussed in detail in Metalink Note 465787.1 “Managing CBO Stats during an upgrade from 9i to 10g”.
Individual queries that suffer from unstable execution plans can be stabilized via the creation of stored outlines or SQL profiles.
Despite all best efforts, occasionally new Oracle statistics will cause the CBO to choose an entirely new execution path, resulting in poor performance. Oracle has provided a mechanism to “rollback” changes made to statistics via the DBMS_STATS.RESTORE_TABLE_STATS package.

Summary

From an Oracle database perspective, the Cost-Based Optimizer is here to stay. Despite this author’s many rantings about its instability, it is possible to manage the CBO quite effectively in an Oracle JDE setting. This involves an understanding of the changes to the CBO between Oracle 9i and Oracle 10g, and it likewise involves the Oracle DBA understanding the performance profile of the underlying application. This is all the more important as the size of Oracle JDE databases scales over time. As Oracle has demonstrated time and again, nothing ever stays the same.

About the Author:

Roger Carter of DBConnect Solutions, Inc. has over 20 years of experience in the Information Technology arena working as an Oracle DBA for nearly all of his career. He brings a world-class approach to customer service that is lacking in the Oracle DBA arena in general and more specifically within the realm of ERP implementations. Roger is highly-regarded as a subject-matter expert in the area of Oracle tuning for JDE, with many years of experience managing and tuning multi-terabyte Oracle databases.
Roger is the founder and past president of the Iowa Oracle Users Group and has recruited, trained, and mentored dozens of Oracle DBAs over the course of his career, many of whom are now highly-regarded in their own right.

Making Windows 2008 less Vista-ish

2008-05-14T20:31:01Z

One of the main objections many people have to adopting Windows 2008 Server, petty as it may be, is that is feels too much like Vista. I sure this has nothing to do with Windows 2008 being built on the Vista code-base. So for you creatures-of-comfort out there let me submit a few humble suggestions:

1.) Set explorer to classic mode. Do this by right-mouse clicking on the START orb and choosing properties. Then choose classic mode.

2.) Stop Windows from locking your session after 15 minutes of ideal time. Go into Global Policy’s and modify Windows Settings -> Security Settings -> Local Policies -> Security Options and then select Microsoft Network server: Amount of idle time required before suspending sessions. The default value is 15. I’d change it to 360 as long as you’re SOX auditors aren’t watching!

3.) Be able to shutdown the server without having to log in first. Go into Global Policy’s and modify Windows Settings -> Security Settings -> Local Policies -> Security Options and then select Shutdown: Allow system to be shut down without having to log on. Remember though, anyone have physical access to the server, including the janitorial staff, can now shutdown your server.

4.) Get rid of those pesky approval boxes. The Mac commercials nailed this one. We all hate those pop-ups. Go into Global Policy’s and modify Windows Settings -> Security Settings -> Local Policies -> Security Options and then select User Account Control: Run all administrators in Admin approval mode. Choose disable.

Lessons from DR

2008-05-13T10:11:53Z

I arrived in Philadelphia late Sunday afternoon and departed for home Wednesday morning. During this time span I managed to get a cumulative total of 12 hours of sleep. "What accounts for my sleep deprivation" you ask? "Too much caffeine in my system?" Well, yes, but the main culprit is our yearly disaster recovery test.

Our test began Monday morning at 8 a.m. and ran for the next 48 hours. During this time we recovered an ERP system, a database cluster, a data warehouse, Exchange, and miscellaneous third-party applications. I'm happy to report we were largely successful with our recovery meeting all of our primary objectives.

Below I will outline principles we followed to make this exercise as successful as it was. I also will lay out several suggestions that had we followed would have made it even more efficient.

See the full article at: http://informedcio.com/2008/05/13/lesson-from-dr.aspx

Data Encryption: TrueCrypt

2008-03-25T09:20:16Z

Technorati Tags: security,encryption

Protecting your customers data against accidental leakage is perhaps one of the most important concerns facing CIOs today. Almost every week there is a news story concerning some organization who inadvertently lost a laptop containing Social Security numbers, backup tapes containing credit card data, or some other type of personally identifiable information. When these data leakages occur it not only results in loss of credibility for your corporation but also places you in serious legal jeopardy.

I am familiar with this issue firsthand. About a year ago a customer requested copies of backup tapes for their ERP system. We duplicated the tapes and sent them via UPS. The tracking numbers show that the tapes made it to the local warehouse however after that point completely disappeared. In our case as the data was already an encrypted format so we had little concern over the missing tapes. Had encryption not been involved this could have been a disaster as it contained all of the companies financial, sales, and procurement data.

When looking at various encryption products available one stands out because of an important feature it offers: it's free! In an era of ever shrinking IT budgets the words "open source" and "free" are like music to the CIO's ear. Another really important advantage of going with an open source solution for encryption is that you can be relatively sure the product in question lacks backdoors as its source code is being reviewed by literally hundreds of developers.

TrueCrypt, http://www.truecrypt.org, works using a very similar concept found in server virtualization. It creates a file on your local machine that will later act as a virtual hard drive. This file can be formatted with either a FAT32 or NTFS file system. This file will be encrypted based on the algorithm you select and then can only be accessed via password. Once you have created this file you use the TrueCrypt program to mount it as a drive letter you simply access it as though it were any other disk resource attach to your machine. Once you've copied the data to be encrypted drive you can then dismount it making it unavailable to anyone else to view without the password.

This particular technique is very useful if you are planning to send data to a third party via e-mail or FTP. All the third party requires in order to unencrypted data is a copy of the TrueCrypt software and the password.

Another option for using the software is creating a larger volume on your workstation or desktop and using it to store sensitive data such as your Outlook PST files, spreadsheets, and MP3s! One of the advantages to this arrangement versus the whole hard drive encryption scenarios is that you are only focusing on encrypting data that you're worried about and not expending CPU and IO resources on making sure no one steals a copy of your Microsoft Office DLLs. The downside, of course, is that it relies on you to ensure that everything that needs to be protected is going to that encrypted volume. (It's been my experience that any system relying on discipline of end users is doomed to failure.)

For the price conscious CIO I would highly recommend this product. It doesn't offer all the features of commercial products such as PGP however it does go a long way towards mitigating potential legal issues you could face with a data breach.

Citrix, A Fond Farewell?

2008-02-26T15:56:47Z

For the past 10 years Citrix, in all its various versions, has proven itself to be my faithful friend. Together we have wheather all of the trials and tribulations that IT can throw at a person. And furthermore, we've been successful. Citrix has, in large part, been responsible for the success of my ERP career. That's why, with much sadness, I realize this once great product does not have much of a future.

Microsoft has had remote desktop offering since 1998 when they released Windows NT 4.0 Terminal Server Edition. Its functionality, while extremely useful, was somewhat limited. If you just needed to publish a single application or do server load balancing in a meaningful way Citrix was the answer. Don't even get me started about the bandwidth benefits of the ICA protocol versus RDP. But alas, this is all ancient history.

With Windows 2008 Server Microsoft has added into terminal services all of the functionality that once made Citrix great. The most compelling feature new feature added is support for Remote Applications (aka published apps). With remote applications you no longer have to have your users running in a full desktop session to take advantage of remote computing power.

Now, you can simply publish out an individual application and it interacts seamlessly with your desktop giving the illusion that it is installed locally.

The second traditional advantage Citrix has always had was the ability to publish your applications to the Web. Microsoft also has an answer for this now with it's TS Gateway. The model is you place a gateway server on your network perimeter and your Internet users users connect to it using RPC over HTTP. The beautiful thing for the end user is that no VPN is required.

In interest of fairness I should mention one of the main limitations are ever encountered with this new version of terminal services. You requirement for connecting to the new terminal services is that your workstation or laptop be running either XP SP3 or Vista SP1. Many enterprises have chosen to go with extremely thin clients such as Winterms or are running Linux on the desktops. At this time such an approach is not possible although I'm sure third parties will very quickly stepped in to fill these gaps.

So is Citrix dead? Only time will tell the answer.

AMD's New Phenom Chip Causes x64-Windows to Disappear

2008-02-16T12:56:37Z

For years, I've run their processors in all of my home machines. Their stock has shown up in my portfolio several times over the last decade. I've loved the David and Goliath tale of the small, upstart company taking on evil Intel…

This is why it breaks my heart to tell you now to sell AMD's stock short.

The story begins about three weeks ago, when I decided to build a new system for the purpose of testing Windows 2008's vitalization technology. Although still in beta, HyperV promises to make Windows a mainstream for vitalization. The catch with it, however, is that it must run on a 64-bit operating system. I made the decision to build my test lab on AMD's new Phenom processor line.

I built my lab server using the Phenom 9600 quad-core chip, running at 2.3 GHz with 8GB of RAM, and eagerly installed Windows 2008 Enterprise, 64-bit edition, on it. Everything went smoothly until I attempted to enable the Terminal Services role. When I did, it blue-screened on me with the message: "a clock interrupt was not received on a secondary processor within the allotted time." My heart sank.

I rebooted the server, went back into Server Manager, and this time successfully installed the Terminal Services role. I thought maybe I was going to be okay after all. My hopes were short-lived. As I attempted to install the HyperV role, it blue-screened again with the same message.

Having BSOD’d twice within 30 minutes, I decided to spend some time Googling the error, and came up with several hits concerning TLB erratum within the Phenom processor. Supposedly, this erratum could be mitigated by a bios upgrade. I verified that my motherboard manufacturer had an update for the TLB erratum, upgraded my bios, enabled the TLB erratum setting, and tried again. Oh, the agony I felt when, yet again, I crashed with the same error! I tried under-clocking my CPU as well, but to no avail.

Further Googling pointed me toward the strong possibility that I had a bad chip. According to the rumor mill, the three-core chips AMD is now selling are actually refurbished, four-core chips containing a bad core that has been disabled via microcode. I contacted my supplier and eagerly awaited the replacement CPU.

It arrived yesterday evening and I installed it this morning. Once again, I received the clock-interrupt error. This time I contacted AMD technical support directly. I was no more than two sentences into my explanation when the engineer interrupted me and said that it was a known issue with Vista’s 64-bit edition. (Note that Windows 2008 Server is built on the Vista code base.) He also told me that they were not sure if it was a Microsoft problem or an issue with the processor. He then told me they would contact me as soon as they had any more information.

Bottom line: this processor should have never been released for the 64-bit environment. Even a cursory level of quality assurance on Windows Vista’s 64-bit edition would have demonstrated the problem. AMD has lost a serious amount of credibility with me, and perhaps with the IT community at large. They should halt sales of this processor until the issue is resolved, or at least issue strong disclaimers concerning problems with 64-bit systems.

I'm really sad to see this happening with what has been one of my favorite companies.

Windows Server 2008 Tested by Hundreds of Thousands of Consumers

2008-01-27T10:44:52Z

I've been eagerly waiting for Microsoft to release SP1 for Vista but decidedly against the idea of running it on my laptop until it's completely out of beta and officially released. (I deal with enough buggy software every day without having to go out and look for it.) All of this changed this weekend while I was working on a Windows 2008 Server Terminal Services demo. When I went tried launching remote applications from my laptop I got a message informing me that I must be on Vista SP1 or XP SP3. With much mumbling I broke down and installed SP1 RC on my laptop.

Three hours later when the process was done (that included download time) I noticed something very interesting. My Vista desktop was now showing that it is running Build 6001.

This build number sounded suspiciously familiar so I went poking around my lab servers. Much to my utter amazement I found that the build numbers were identical too my Windows 2008 lab boxes.

I knew that Windows Server 2008 was built on the Vista SP1 code base but I never realized how truly tightly coupled they were until now. This is is one of the sneakiest and most brilliant things that Microsoft is done in quite a while. They now have hundreds of thousands quasi-professional consumers testing their server product.

Anybody care to bet on a release date for Vista SP1? I'm going to put it within two weeks of the release of Windows Server 2008.

Windows 2008 Virtual Cluster Lab

2008-01-25T11:34:25Z

Want to check out the clustering changes in Windows 2008 but dont have a SAN, servers, and a lot of time? No problem, the folks at Microsoft have got you covered. They are providing a virtual lab for you to play with all this until your heart is contented.

http://msevents.microsoft.com/CUI/WebCastEventDetails.aspx?EventID=1032345932&EventCategory=3&culture=en-US&CountryCode=US

Title

TechNet Virtual Lab: Windows Server 2008 Enterprise Failover Clustering Lab

Overview

For the exercises in this lab, you will be responsible for the configuration of Windows Server 2008 Enterprise failover clustering for your organization. This lab will teach you how to add nodes to the cluster and configure the cluster to support a variety of high-availability roles such as file, print, and DHCP server clusters.

Dude, Who Stole my GUI

2008-01-25T11:33:07Z

For too long Microsoft has continually been lowering the bar in terms of who was capable of administrating Windows servers. Long ago the Mystical Order of the Computer Priesthood was strong and healthy but alas, as ever more have come into IT, we have become overpopulated and weakened. We are no longer nerds and geeks. It is time for a return to the old ways. It is time for us to admit something that our UNIX brethren have known for decades... "Real operating systems dont' have GUI's."

Ok... enough of the melodramatics... Let's talk about Microsoft Windows 2008 Server Core installations.

A core server installation is a scaled-back version of the full server install. All system administration is done via command prompt. There is no GUI. Once the server is on the network though it can be administrated remotely via Microsoft Management Console (MMC).

A server core will not include.Net Framework, Internet Explorer, or any other feature not critical to server operations. After installing the core the adminitrator has the ability to configure it for one or more of many basic roles: domain controller, DNS server, DHCP server, file server, print server, Windows media server, terminal services Gateway, Web server, and virtualization server.

There are two main compelling reasons why we will see the entire industry rapidly adopting core installations. The first has to do with security and the second with availability.

By introducing core installations in Windows Server 2008 Microsoft has dramatically reduced the possible surface area for an attack. In previous versions, by default, the operating system was installed with the vast majority of its functionality enabled. This meant, that features installed on a server, but not actively being used, might be exploited to gained access to the system and it's data. Personally, I feel that getting rid of the GUI was taking this to a bit of an extreme, however it does show how serious of a commitment Microsoft is made to security these days.

The second reason to consider core installations is server availability. One of the largest causes of scheduled system unavailability is the requirement to keep your servers patched. When Microsoft reduced the surface area of the operating system it also reduced what needed to be patched. Now, instead of having to apply patches for Internet Explorer or IIS, you only have to patch for the core install, any roles that have been added, and software running on top of the operating system. This could potentially cut in half the amount of patching required.

The bottom line is that in spite of the whining you will hear from your system administrators core installs are going to be the way to go. (It's also quite possible regulatory considerations and auditors will give you no choice in the matter.) Any inefficiencies introduced on the system administration side will be more than offset by the savings you will gain from increased system availability.

Below is a list of configuration commands I used when setting the core installation demo:

netsh interface ipv4 show interfaces

netsh interface ipv4 set address name="2" source=static address=192.168.1.4 mask=255.255.255.0 gateway=192.168.1.1

netsh interface ipv4 add dnsserver name="2" address=192.168.1.3 index=1

netdom renamecomputer win-43215454 /NewName:lab2

For a step-by-step guide on configuring a server core go to the following link:

http://technet2.microsoft.com/windowsserver2008/en/library/47a23a74-e13c-46de-8d30-ad0afb1eaffc1033.mspx?mfr=true

VMWare and High Availability

2008-01-25T11:32:38Z

Virtualization is one of the most exciting technologies in the computer industry at the moment. The reason for this is quite simple. Most servers run at a very low level of resource utilization. Virtualization allows companies to consolidate physical servers into virtual ones thus reducing their overall hardware expenditures and simplifying overall system administration. This article will focus on one of the aspect of simplifying system administration, that being high-availability.

High-availability is the holy grail for the system administrator. Aside from answering the obvious questions of what happens in case of a physical failure, an event that should happen very infrequently, it also helps solve the more common issue of how to get a scheduled outage window to perform system maintenance. As companies increasingly automate business processes getting these outage windows becomes proportional more difficult due to business costs associated with system unavailability.

In the video above I demonstrate how to use the VMWare Server in a clustered environment. More important than simple failover of a VM from one node to another is the ability to save state on one node and to restart it on another node. To put this in English, any applications that were running on the VM on one node will still be running on the other node when it's brought back up.

Below is a list of the commands I used in the video:

vmware-cmd d:\testvm\testvm.vmx start

vmware-cmd d:\testvm\testvm.vmx suspend

cluster group vm /moveto:sclabent11

In the demo above I simply used Notepad as my application to demonstrate the saving of state and failover. One of my next projects will be to build out a complete ERP system, Oracle’s OneWorld, on VMs and test how well real applications failover.

Please let me know any thoughts or questions you may have on the topic of virtualization and high-availability.

The Good, the Bad, and the Ugly about VMware Server 2.0 Beta 1

2008-01-18T17:21:05Z

One of the best things about starting Informed CIO is that now I have a legitimate excuse to start playing around with software that typically I would not be touching until at least it’s first service pack is released. Such is the case with a VMware's newest entry into the virtualization wars, VMware Server 2.0 Beta 1.

The Good…

This new version of VMware, although still in beta, certainly is a contender. It has now added support for Windows Vista Business and Ultimate editions as well as Windows Server 2008. Additionally it now supports Red Hat Enterprise Linux 5 and Ubuntu 7.x. (I note this least the Linux zealots spam me to death!)

In addition to supporting new operating systems VMware Server 2.0 has made huge leaps over the 1.X versions in terms of scalability. Previously a guest VM was capped to using no more than 3.6 gig of RAM. In the current version a guest VM can use up to eight gig of RAM and two processors. This added capability should move virtualization much farther up the food chain. I seriously believe that once this product GA’s we will begin seeing an exponential increase in production servers being virtualized.

The best news about VMware Server 2.0 is that it will continue to be free. VMware will begin offering paid support options for the software once it is officially released however there will be nothing to stop you from legitimately installing as many copies on as many servers as you want.

The Bad…

For Windows, VMware Server 2.0 continues to only be a 32-bit application. For Lunix they do have a 64-bit offering however it will place them at a serious competitive disadvantage as Microsoft's HyperV will be being released about the same time. In terms of virtualization any serious contender must be running the host on 64-bit code, period. I won't delve into the differences between memory architecture for 32-bit and 64-bit systems suffice to say 32 bit is going to have serious performance limitations.

Another negative is that I found VMware’s a new web interface a bit cumbersome and sluggish. In the interest of being fair I should point out that it also found me to be a bit cumbersome and sluggish. We're both in counseling and hopefully we will be able to make this relationship work.

The Ugly…

One final cautionary note about VMWare’s new beta: after uninstalling version 1.0.6 and installing version 2 .0 I was forced to reactivate my copy of Windows 2003 Enterprise edition. I'm not positive that VM was the culprit however I find the timing very suspicious.

Please let me know about your experiences. Feel free to add any comments, questions, or corrections below.

Backup and Recovery in Windows 2008 Server

2008-01-18T17:06:32Z

The backup process in Windows 2008 server has taken a significant step forward in terms of usability from the Automated System Recovery (ASR) process of Windows 2003/XP and even more so when compared to the legacy NTbackup.exe process of yesteryear. The new backup process integrates into the server manager console. When you first open it up it immediately shows you a listing of all recent backups for the server. The available actions to perform have also been greatly simplified and streamlined. You are given the choice between performing the backup immediately, scheduling a backup, or performing a restore.

Earlier today I was able to make a backup of my test server with just a few simple mouse clicks. (See video above.) I then took the backup and restore it to an alternate server. (See video below.)

This by far has been the easiest system recovery I have ever performed outside of using imaging products such as Norton Ghost or Acronis. For enterprise customers you're still going to want to consider some form of centralized backup product however for the small to mid enterprise customer using of the native tool in Windows 2008 is the easy, and cost-effective way to go.

Extending a Boot Volume

2008-01-18T17:08:38Z

In Windows Server 2008 you now have the ability to extend and shrink volumes... or so I thought. When I installed my test lab with RC1 I only created an 8GB boot volume on the VM server. This proved to be a bit shortsighted... I quickly ran out of space... When I went to extend it only then did I realize that I had ran into another limitation with this new disk management functionality. You cannot extend a boot volume across more than one disk. If you have additional space on that disc you can extend it however you can't make it span discs. And no, DISKPART does not help.

The video above shows my traumatic experience in trying to make the boot volume span drives.

Shrinking and Extending Volumes in Windows 2008 Server

2008-01-18T17:09:14Z

One of the most anticipated features in Windows server 2008 surrounds disk management. In the Disk Manager you have the ability to create, extend, and shrink simple and spanned volumes. In prior releases of Windows you have the ability to extend volumes through the rather clunky diskpart command. In Windows 2008 server this is all done via a graphical, wizard driven interface. The ability to shrink a volume is completely new with this release. (But it's not without its bugs, as I demonstrate in the video.)

One major limitation that Microsoft still has in this product is its inability to extend and shrink stripes. Any applications that require sustained levels of high disk I/O will require striping. The main difference between a spanned volume and the stripe is how its data gets written to the disk's. With a spanned volume the in a and I is that it is not in a a is written to the first disk until its full and then begins writing to the second disc. This is useful in that if you lose one of your disks you have not lost all of your data, or so the theory goes.

With the stripe files are written to all of the disks meaning that you can take your I/O characteristics for each individual spindle, add them together, and, assuming you don't saturate your controller, that should be the sustained IOPS level. Using the example from the video of four discs they should have four times better performance characteristics within the four discs configured as a spanned volume.

If you have a system requiring higher levels of disk IO and you want the ability to add and shrink stripes I would recommend going with the VERITAS storage foundation product. It provides the ease of management that you're looking for but still in a way that optimizes performance. (Just don't spend too much time looking at its price tag. It might you make cry.)

Enhanced Internet Security in Windows Server 2008

2008-01-18T17:10:02Z

Enhanced Internet Security is perhaps one of the most annoying features in Windows. This feature was introduced in Windows Server 2003 and could be disabled by uninstalling the option via add/remove programs. Many system administrators, finding the continual pop-up notifications maddening, would immediately remove this feature from the server. This, of course, left their servers exposed to malware from the internet.

In Windows Server 2008 Microsoft has introduced more granular control for Enhanced Internet Security. Now, by default, it is enabled for all administrative accounts but disabled for less privileged accounts. In terms of balancing security against usability this seems to be a very sensible trade-off. The administrator can log on with a less privileged account, complete their business, and then log back as an administrator to complete any installation or patching activity.

Enabling and disabling enhanced Internet security is done from the new server manager console. The video above will show a quick demonstration of this.