Hardware

Websphere on Windows Outperforms Websphere on AIX by 37%, Who Knew?

Rick G. Garibay — Thu, 30 Apr 2009 14:39:17 GMT

In a conference call with Steven Martin yesterday, the head of the products division that includes .NET, Azure, IIS, ASP.NET, Dublin and Oslo shared some very impressive news.

Led by Greg Leake, a team of researches in the Microsoft Connected Systems Division set up two rigs in the lab. One was a 64-bit IBM WebSphere 7 Application Server running on an IBM Power 570 (IBM Power6/AIX 5.3 platform) and the other was a Hewlett Packard BladeSystem C7000 with Windows Server 2008 64-bit. Each were set up to run comparable versions of the Stock Trader reference application which has long been a reference standard for performance benchmarks.

In his paper, Greg compares the performance and price of these two IBM WebSphere platforms to the equivalent workloads developed using the Microsoft .NET Framework and deployed to the Hewlett Packard BladeSystem C7000 with Windows Server 2008 64-bit operating system.The results are astounding.

Of course, as you would expect, the ported version of the application, for .NET outperformed the J2EE/Websphere version by 57%, because it is no surprise that .NET is leaner and meaner than the Java platform, especially when it comes to distributed communication and Service Orientation leveraging WCF.

What is even more impressive, is that the IBM reference application, designed for Websphere runs 37% more efficiently on Websphere on Windows than Websphere on AIX! It gets better. The AIX rig, which Microsoft acquired on the regular market, retailed in at a total system cost of $215,000. This figure includes the IBM Power 570 hardware cost plus cost of Websphere middleware.

The alternative? The platform that has long been regarded as inferior for large enterprise application scenario? The platform that has spanked the IBM equivalent by 57%? The Windows Server 2008 system, running HP BladeSystem C7000 has a total hardware cost of, wait for it… $50,161.That’s an 81% system cost savings for a configuration that outperforms the 6-figure-monster by 57%.

Now of course, this would be news in itself. But it doesn’t stop there. Greg’s group found that the identical reference application, deployed for Websphere on Windows Sever 2008 outperforms the IBM/AIX rig by 37%, and all for a 66% cost saving. The savings of course is reduced from the 81%, because unlike the Windows Platform, Websphere, well, isn’t free. Tacking $37,000 on for Websphere middleware licensing costs brings the Websphere on Windows system cost up to $87,161.

Now of course, the questions you have to ask, given this detailed information are the following:

Why would you choose an inferior platform to run you mission critical enterprise applications when it underperforms when compared to an equivalent hardware and middleware configuration, especially when the IBM/AIX configuration costs 81% more?
If you are running Websphere, why would you not opt to save 66% of your total system costs by moving to the Windows platform, especially if doing so would result in over a third better total application performance?

These are some big questions that Microsoft hopers Big Blue will attempt to answer.

In fact, Microsoft has prepared a site dedicated to informing and educating the industry on its findings: http://www.websphereloveswindows.com/

Now, while the performance benchmarks and cost savings speak for themselves, I have some breaking news. The report is flawed. Severely flawed. The report doesn’t speak to other costs such as administrative costs, maintenance or professional services. If you are an IBM/Websphere/AIX shop, you know too well how expensive keeping these systems running is. From tooling, to readiness, the Microsoft platform just has a far better story when it comes to administration and maintenance. And, if you’ve ever hired at Websphere consultant, you know what a bargain services from even a premier managed partner like Neudesic are.

References:

Hyper-V Parent Partition Filesytem Access Not Working?

Rick G. Garibay — Fri, 05 Sep 2008 16:30:27 GMT

With Hyper-v, if you want to access the parent partition (host) file system, or attached USB drives, the only way to do so is through a UNC or IP connection from the child partition (guest) to the parent (host). To do so, you will have to create a network and bind it. You can read about how to setup an Internal Network in a previous post.

At times, you may find that despite being able to ping the parent from the child, and even resolve DNS names, you get a networking error:

For whatever reason, I have found that restarting the "Server" service on the parent resolves the issue. As you might imagine, the Server service is responsible for managing file sharing, among other things. What is strange is that it is already running, but every time I run into this issue, restarting the service is the fix.

Hope this helps someone out who could otherwise spend hours pulling their hair out debugging networking issues.

Hyper-V Internal Virtual Network- The New Loopback

Rick G. Garibay — Sat, 23 Aug 2008 05:36:20 GMT

In Virtual PC and Virtual Server, one way to build a private network of VMs was to create a loopack adapter using the Microsoft Loopback Adapter. This allowed you to specify an internal IP address (such as 192.168.1.1) and as long as the IP and subnet mask on the VMs were configured properly, you could communicate from one VM to another. This is such a common scenario, that I often wondered why one had to go through the contortion of creating a loopback adapter. I am not a Virtual PC/Server or Hyper-V expert, but the need for this seems to be quite common.

Well, leave it to Microsoft to eventually comooditize key user stories.

With Hyper-V, creating a network of VMs is as simple as creating an Internal Virtual Network and binding it to each of the child partitions. When you do this, an internal virtual NIC is created on the parent partition which you can then configure with the appropriate subnet. Match the settings to each child partition in your network, and viola, you are golden.

To create an Internal Virtual Network, click on "Virtual Network Manager" in Hyper-V manager.

In Virtual Network Manager, select Internal in the list of network types and click Add:

Give the new Virtual Network a name and click OK.

That's it. You're done.

Of course, you must bind the new Virtual Network to your Virtual Machine. To do so, select the Virtual Machine within Hyper-V Manager, click Settings, and select the name of the network you just created.

On a side note, I learned about this by unlearning what I had already learned with Virtual PC. I had actually gone through the process of creating a loopback adapter when I realized that I get one for free by creating an Internal Virtual Network. It is amazing how so often, to get the most out of new technology, you must first unlearn old tricks.

Merging Differencing Disks with Hyper-V

Rick G. Garibay — Sat, 16 Aug 2008 06:26:51 GMT

A differencing disk is a disk that is a child of a parent disk. Differencing disks are very helpful in keeping disk images small, manageable and consistent, because you can create a base parent disk- such as a Windows 2008 Standard base image- and use it as the foundation for all other guest virtual machines and disks that will be based on Windows Server 2008. For example, I have a Windows Server 2008 guest that I use exclusively as sandbox for development. I am in the process of building out another guest based on Windows Server 2008 that will be for some TFS 2008 demos that I am working on for an upcoming series of talks. Rather than copy the Windows Server 2008 guest VPC over and over again, I can simply create one differencing disk for my development environment role and one for my TFS role. The result is a VHD that represents the intersection of the base/parent disk (in this case, a barebones install of Windows Server 2008 Standard) and any additional software I've installed or configuration changes I have made. This not only conserves disk space, but also saves me a lot of time in copying hefty giga-some-odd vhds around.

Sometimes it is necessary to merge a differencing disk back to it's parent or into a new disk. For example, you may be moving VHDs around as I did recently to a new, high speed E-SATA drive. My old drive hosted a vhd that I used as my development sandbox that used a parent on the old disk. I certainly don't want to depend on my clunky old USB 2.0 drive for the parent (the IO cost alone would be just silly), and at a minimum, there is state on the differenced guest OS that I do not want to lose.

The first thing to do is copy over the parent VHD, create a new differencing disk based on the same parent, but in the new location.

Next, since the differenced guest VHD has state that you want to move over (lest you lose it), it is necessary to merge the state of the "old" differenced guest VHD with the new copy. To do so, under Server Manager, in the Hyper-V Manager, click "Edit Disk", and locate the disk that you want to merge into a new differenced disk:

On the next screen, under Action, select "Merge":

Select "To a new virtual disk", and choose a name and path for the new disk that you created in the initial copy:

The "old" differenced disk, which is based on the original parent disk plus state from the "old" differenced disk is merged into the new disk on the drive you specified:

That's all there is to it. Differencing is a powerful feature in virtualization, and there is very nice support for migration of differenced disks right within the Server Manager.

Windows Server 2008 Update

Rick G. Garibay — Fri, 25 Jul 2008 15:46:43 GMT

OK. Honeymoon is over. I have been experiencing some degradation in performance/responsiveness recently, so started looking for optimizations which have had a positive result.

Here is what I have done:

Set any non-essential services to "Manual" and stop them. This has been somewhat of a trial and error exercise, but I've been able to free up some memory and CPU utilization in the process.
Adjust Visual Effects for Best Performance. This can be accomplished by going to to System Settings->Advanced ->Performance Tab->Settings Button-> Visual Effects. The only visual effect that I couldn't live without was smooth fonts, so I turned it on.
Uninstalled Trillian. This thing does not seem to play well with Windows Server 2008 64-bit as it was constantly churning CPU and memory. I noticed a marked improvement in CPU utilization when I turned it off, and have since removed it thanks to this post which pointed me to a previous version of Windows Live Messenger which installed just fine.
I also enabled "Superfetch" per this post.

I am also finding that my USB 2.0 hardrive is becoming a bottleneck since I use it to store all of my VHD files. I might go out and get an ESATA drive this weekend and see if I notice a difference.

I will report back in a few days with an update.

Renaming SQL Server 2005

Rick G. Garibay - The more I learn, the less I kno — Wed, 16 Jan 2008 23:22:46 GMT

When using VPC, it is common to clone machines and rename servers, which can pose a problem for platforms such as BizTalk Server which are configured to connect to a given machine name at installation/configuration time.

The first step in managing a rename is to reconfigure BizTalk using the BizTalk Configuration Tool, but there is an additional step that is required from a SQL Server perspective.

If you perform the following query within the master table in the SQL Server that houses the BTS databases, you will get a record that includes the following:

select * from sys.servers

server_id name
=============================
0 CurrentServer

Where the server name has changed to something like "NewServer" it is necessary to not only rename the machine and assign it a new SID, but also to rename the SQL server name.

To do so, execute the following statement which drops the server name (don't worry your databases will still remain intact):

exec sp_dropserver 'CurrentServer' GO

Next, execute the following statement, providing the new server name:

sp_addserver 'NewServer', LOCAL GO

Now, run the same query as above and the result should include the new server name:

server_id name
=============================
0 NewServer

Once this is complete, you will more than likely need to unconfigure and reconfigure your BTS installation. There are some components where this is not required, but I haven't quite figured out which are which yet.

The Attack of the FAT 32 Partition

Rick G. Garibay - The more I learn, the less I kno — Mon, 14 May 2007 03:55:50 GMT

I just picked up a 500 GB USB 2.0 external drive to handle the back ups for my domain controller, application and database server that I run in my SOHO network which runs on a Windows 2003 AD domain (I can't believe that a half terrabyte can be had today for less than $200). My backup strategy has been pretty consistent over the last couple of years; full backup on Sunday and Wednesday with incremental backups every day in between. This allows for fault tolerance in the event of a botched full backup, while providing recoverability up to the day before a potential failure.

The drive came pre-formatted in it's own enclosure, so replacing the existing 80 GB drive was a breeze. The idea was to keep the current backup schedule intact by keeping the same drive letter, etc.

The number one rule of disaster recovery is to test your backups, so the first thing I did was run the full backup job that typically runs on Sunday. Strangely, even though the partition that is backed up is over 30 GB, the bkf file was only 4 GB. Baffled, I tried running the job a number of times interactively with the same result. The file was exactly 4 GB in size.

The first thought was that somehow, the backed up attribute had somehow been set on what, 26 GB worth of files? This made no sense, because a full backup doesn't set (or clear) any flags, but when strange things happen, so too does the mind wander.

Well, it turned out that the drive was formatted at the factory with FAT 32. One of the limitations of FAT 32 is a maximum file size of 4 GB. I formatted the drive using NTFS (which has no such limitation), bringing it inline with the rest of the machines on the domain and the backup is operating as expected, creating a bkf file comparable in size to the source partition.

So, if you ever run into something similar (as rare as encountering FAT 32 these days might be) , check the file allocation scheme on your disk and if possible, re-format in NTFS. Not only does this remove the file size limitations, it is far more efficient and secure.

New Rig Rev 2.0

Rick G. Garibay — Mon, 05 Feb 2007 21:49:41 GMT

In an earlier post here, I provided the specs to my new system that I built during my 1 week sebatical before starting my new job.

Well, in just over a month, I've already upgraded.

"I ordered two idential 250GB WD drives and configured them in RAID 1 mode (mirroring)."

Yep, its true.

While at first, my inner geek was pleased with my new rig- with a dual core processor and 2 GB of schaweet DDR2 RAM- how could it get much better?

Well, I will 'splain.

First, after tooling around on my new machine, I quickly realized that while a 400 GB hardrive is mondo big and would provide for enough storage to allow me to do all the video editing I could imagine as well as keep me from having to uninstall old games to reinstall new ones, it was impractical. The reason being that while 400GB is very accomodating, it also becomes incresingly volitile as the capacity fills up. From a risk perspective, anything above and beyond game files would just be a liability in the event of a HD crash. Sure, I could back up the 400GB monster, but to what media? Another 400GB (or better) monster, just sitting there, under-utilized?

So what did I do? You guessed it! I dropped another $160 and got a second 400GB drive! No, not exactly. For about $60 more than I paid for the Seagate drive, I ordered two idential 250GB WD drives and configured them in RAID 1 mode (mirroring). This should be perfect for gaming, video editing and development, especially since I have a 120 GB external USB 2.0 drive also installed for the purpose of staging video and content before it goes to production. The mirroring will provide 100% fault tolerance in the event that one of the disks goes bad. Although it is possible that both disks could fail at once, this possibility is purely theoretical. Today, RAID is becoming a staple in mid-range to high-end machines and with disk prices this cheap, there really isn't a good reason not to protect your data.

The second addition was completely unexpected. While Asus' flagship AM2 board comes SLI ready, I wasn't sure that I'd be adding a second card anytime real soon. For anyone that is unaware and curious, SLI stands for Scalable Link Interface. It is a technology developed by Nividia that allows for two cards to be bound together as a single output. Essentially, this is the equivelent of dual processors on a board, which is similar in concept to dual core, where both processors reside on the same processor wafer. Well, during one of my almost weekly visits to Fry's, I caught the week's add circular which stated a whopping $80 mail in rebate and a $30 instant rebate on eVGA Geforce 7600GT cards! A pefect pair to my existing eVGA Geforce 7600GT card for 70 bucks?! Deal! So, my system is now in SLI mode with two eVGA Geforce 7600GT cards. I can't wait to run some benchies.

On a side note, one of the rules I adopted some time ago with regard to video cards is to never, ever buy the latest spec. Chances are, as I have several times before, you'll get burned. New technlogy is completely expendable, and when you are paying for the latest spec, you are going to overpay because the minute you pop that card into your new rig, you will read about the next big thing which will immediately devalue your investment. Regardless of what zealots may say, there is more to dropping $250 to $300 on a card than the mere fun and satisfaction that comes with enjoying the card as soon as you get it home. It is painful to see the card drop $100 two months later forever relegated to the obsolete bin. So my rule is to try to buy last year's hot card. In this case, back in 2005 it was the Geforce 7600GT chips that was hot. Now, as you can see, I can pick up a second card for $70- I only wish my first card would have been that cheap!

So, all said and done, here are the revised specs of my new build:

Englight Midtower Case (This thing is built like a tank. I've owned it since 2001!)
HIPER HPU-4S580-MS ATX12V 580W Power Supply
ASUS M2N32-SLI Deluxe
AMD X2 3800+ (AM2 Socket)
Corsair XMS2 2Gb Dual Kit @ DDR2-800
2 eVGA Geforce 7600GT in SLI Dual Mode
Plextor PX-716SA DVD/CD RW Optical drive
Sony 24X CD-R Optial drive
2 Western Digitial Caviar 250 GB SATA Hardrives in RAID 1 (Mirroring)

CPU-Z Specs

(Click to Enlarge)

CPU

Motherboard

Memory

Memory SPD