Lukas Beeler's IT Blog

Since December 2008, i’ve used my ThinkPad W500 as my work laptop. We’ve bought this as part of a promotion package.

The W500 i had had a 15.4″ 1920×1200 Panel, which wasn’t too great. While the high fidelity was certainly nice, the screen was very, very dark. It could only be used indoors, and required you to darken the room on sunny days.

Today i’ve had the chance to upgrade from the W500 to a T510, which i did. So far, i’m very much impressed with the changes Lenovo has do to this device. The W500 is running Windows 7 Enterprise x64.

• New controls for volume, microphone mute. Much easier to use than before
• New bigger and multitouch capable touchpad. As i prefer the touchpad over the TrackPoint, this is something that helps me tremendously
• Integrated Camera and eSATA connectivity
• Improved connectivity layout

There’s only one thing that i don’t like very much right now – the redesigned keyboard. As part of my job i deal with IBM’s IBM i platform, which still makes use the Function keys – which have all been shifted to the right for one key. So i regularly press F3 instead of F4, but chances are i will get used to it.

There’s one thing that worked very well – moving my Windows installation from the W500 to the T510. I’ve disabled Bitlocker protection, removed the OCZ Vertex SSD from the W500, placed it into the T510, booted it up, Windows installed several new drivers. Then, i installed the Intel LAN drivers from an USB stick, rebooted once more and installed the rest of the necessary drivers from Lenovo’s driver matrix. The whole process was done in less than half an hour, and reenabling Bitlocker protection was a breeze.

Windows 7 automatically reactivated by contacting our KMS servers, and i’ve had to reactivate my Office 2010 Beta manually, which also worked flawlessly.

While this portability is great (and also existed with Vista), it’s something I was able todo with Linux back in 2004 (assuming of course that the kernel had the storage drivers you required).

I’ve been using ThinkPads exclusively since 2004 – my first ThinkPad was an R51 and my first new laptop (my first laptop was a Compaq Armada i’ve bought used for 50.- CHF). When Lenovo took over the brand, i wasn’t to sure what to think of it, but having gone through several iterations of ThinkPad devices now (R51, T60, W500 and now the T510) i can see that Lenovo is commited to provide further well built, high performance devices.

Both the T60 and the W500 are still in service, neither of them are broken. The T60 is used by my apprentice and around 3 or 4 years old. We’ve replaced the Mouse and Keyboard to mitigate the wear and tear of several 40 hour work weeks on the device, but aside from that it stills works great.

A long time ago, i wrote a review of the HP ML110. In the comments, Paul indicated that the Performance of the E200 controllers was pretty bad, and i promised i would do benchmarks of that. Now we have a year later, and i indeed finally got the time and did those benchmarks.

For the benchmarks, i’ve used the free version of HDtune. I’ve benchmarked four systems, and five different disk configurations. Note that the free version only does benchmarks for disk reads, and it’s a not a very pervasive test. None of these benchmarks are scientific. They should serve as a general indicator of performance, not as a final world on this topic. I don’t have that much clue about benchmarking.

The first system is my computer at home: It has an i7-920 CPU at stock speed, with 3x2GB RAM at 1333 Mhz (which is a slight overclock, but within the spec of the memory i purchased). Attached to it’s ICH10R controller are an Intel X25-M G2 160GB (Firmware 02HA) and a WD1001FALS (1TB, 7×24), running Windows 7 x64.

The next system is my work laptop, which is a ThinkPad W500 with a 2.53 Ghz T9400 C2D CPU, with 4GB of RAM. Attached to it’s onboard controller is an OCZ Vertex 120GB (Firmware 1.40), running Windows 7 x64.

The third system is our Exchange Edge server, on which i dared to install a benchmark utility. It’s an IBM x3250 with two 70GB 15kRPM 2.5″ SAS drives installed, attached to an onboard LSI1064E SAS controller. The system has a Xeon 3040 2.4Ghz Dualcore CPU and 5 GB RAM. It is running Windows Server 2008 x64 SP2.

And the final system is a HP ML110 G5 with a 2.33 Ghz Xeon 3065 CPU, 8GB of RAM and a E200 with the latest firmware (1.78). Attached to that are 4 WD1001FALS drives in a RAID10 configuration. The E200 has a backup battery and 128MB of cache installed. The system is running Windows Server 2008 R2.

Please note that none of these benchmarks are scientific. They were done on real systems, with workload minimized as much as possible, but virus scanners and other mandatory background applications active. Both the laptop and the desktop have not been formatted since Windows 7 RC was installed (i migrated to Windows 7 RTM using Windows.old), but the ML110 was freshly setup and the only application that’s been installed so far is the HP ACU and Forefront Client Security. The Exchange Edge server has been in use since May 2008. As such, the ML110 is the “cleanest” machine out of these four.

Intel’s X25-M G2 160GB on an ICH10R (AHCI Mode)

This is how a graph should look. It’s nice, it’s clean, it’s fast. Intel’s X25-M G2 shows how a modern SSD and storage subsystem should behave. Clean, predictable performance.

OCZ’s Vertex 160GB on an ICH7 (AHCI Mode)

Here’s the OCZ Vertex. It’s running on a machine that’s a lot slower than the one the X25-M is attached to, and it’s storage controller is also quite a bit older. It still shows remarkably good performance. It should also be considered that this Vertex is quite a bit older – it was bought in May 09. It’s still very fast and responsive and a good SSD.

2x IBM’s 73GB 15kRPM 2.5″ SAS Disks on an LSI Logic 1064E SAS Controller

As you can see, this is the performance you get from the server hard disks on an entry-level controller in an entry-level system. It’s not astonishing, but the performance is very well acceptable.

Western Digital’s 1001FALS 1TB on an ICH10R (AHCI Mode)

Here’s how the Western Digital disk behaves on a proper controller. Please note that this is a single disk, not part of a RAID array. The performance is quite good.

4x WD’s 1001FALS 1TB on an HP E200 in RAID10

And here’s how it shouldn’t look. Compare this to the stand-alone disks above, which exhibits better performance. HP fucked up bad on this one, and there’s no fix in sight. Stay away from the E200.

And as a final word: I really don’t have much of a clue about benchmarking. If you see an obvious error here, please state what you think. If possible, i will try to correct it.

Update: As requested in the comments, i upgraded the E200 to Firmware 1.84 and redid the benchmark. It looks roughly the same.

The Midrange Wiki is a good way to get started with the IBM i platform. I’ve started writing a short Getting Started guide there, which may be of interest to you.

If you work on the IBM i platform, the Midrange Mailing Lists may also be a place to visit and subscribe. Also, check out the IMHO Midrange Blog.

In the light of the recent announcement how Hyper-V Server 2008 R2 will be licensed, i thought about familiarizing myself with the Clustering & Live Migration capabilities, using the RC release of Hyper-V Server 2008 R2.

I have to admit that Failover Clustering isn’t exactly the field i have a lot of experience in (in other words, i have never used it in producation). But after seeing that i wouldn’t be drowning in work this friday afternoon, i decided to give it a whirl.

So, in order to get started i needed two machines that were able to support running Hyper-V Server 2008 R2. One of them was HP ML110 G5, about which i wrote a few months back. Unfortunately, i could use only one of them. So my next choice was an old HP desktop, which fortunately had a VT compatible CPU.

Next, i needed a storage backend. Of course i had to use a software solution, but having no experience and only a very old PIV era IBM SFF PC, i just picked the first Google search result which supported SCSI-3 Reservations, which are required for WS08 clusters. I’ve downloaded and installed Open-E DSS.

For networking, all i was able to find was a 100mbit 3com 24 port hub. Yes, this looked like one of the most ghetto environments i put together yet, but interestingly i got it all to work.

Open-E DSS installs to an USB stick, formatted with FAT32. You just unzip the installation file, run an .exe on the stick to make it bootable, and then you can run the system directly from USB. In my case, using rather outdated hardware, everything was recognized by the Linux kernel. Of course, the machine only having a slow 40GB 5400RPM hard disk wasn’t exactly the fastest on the block, but configuration was surprisingly simple. Unfortunately, installing and activating the Lite license required two reboots, after which it lost all it’s iSCSI settings (but no data!)

Installing Hyper-V Server 2008 R2 on the ML110 was a breeze. Using sconfig, the machine was quickly joined to the domain, remote administration enabled, failover clustering enabled and using the graphical iscsicpl iSCSI was configured, the volumes formatted and attached.

Next was the HP desktop machine. Installing was fine, worked perfectly, all the necessary hardware was recognized. Unfortunately, the machine only had 1GB of RAM, which meant that i couldn’t do all that much fancy stuff with it. I was in for a nasty surprise here later, because i didn’t enable Intel VT in the BIOS (which is hidden in the “Security” Options). I think the Hyper-V Server setup should give you a warning here if the feature isn’t enabled.

Next i created the cluster. I’ve used this blogpost and TechNet to get a basic overview on what i needed to do. In just a few steps through the cluster configuration wizards, my cluster was configured and ready. I was able to bring my VM online on the first node (the ML110) and decided to install Windows XP, since i only had 1GB of RAM on the second node. I gave the VM 256MB of RAM and ran through the setup (which took ages – iSCSI over a 100mbit Hub to an old PIV with a 5400rpm hard drive isn’t a good idea anymore).

Next, i decided to setup VM networking, created the appropriate VM interfaces on both machines, restarted my XP VM and tried to do a live migration. Which failed. “Insufficient system resources”. Turns out i needed to adjust the amount of memory reserved for the root partition using PowerShell – all described in this Clustering and High Availability blog post.

After running (get-cluster HV01).RootMemoryReserved=128, it failed again. This time with these event log entries:

‘Test-VM’ The switch port connection for “Network Adapter” (BE62B93F-1490-4F7E-8229-FA18D50DC974) is invalid.

‘Test-VM’ Microsoft Synthetic Ethernet Port (Instance ID {BE62B93F-1490-4F7E-8229-FA18D50DC974}): Failed to Power on with Error ‘The system cannot find the path specified.’ (0×80070003).

Failed to connect NIC ’9144ED30-35D9-4E5F-8012-70AC436EC603–BE62B93F-1490-4F7E-8229-FA18D50DC974′ to port ” on switch ’0734959D-3′, status = C000003A.

I disabled networking in the VM altogether, and tried Live Migration again. It worked! The next was spent with searching the internet for information about my issue, about which i found nothing. Obviously the network interfaces should be named the same in all cluster hosts, but that was the case. Yet, no matter what i did it didn’t work!

I was starting to doubt my hardware, added a second pair of NICs since the configuration of using the same NIC for everything wasn’t really recommended, but when reading the error message it really didn’t sound like that was my issue. Of course adding the second pair of NICs didn’t help.

So i did what i always did: i started guessing, and after quite a bit of time i got it rights. Turns out you must not use the Hyper-V MMC to manage the VM configuration, and instead the “Settings” button in the failover cluster manager. Only issue is that the failover cluster manager has a much more prominent button labeled “manage virtual machine”, which opened the Hyper-V MMC.

After that, everything worked. I was able to live migrate my machine including the network from host to host. I tested running a Top Gear clips through RDP, while live migrating the machine.

Migrating from the slow HP desktop to the ML110 gave about 2 seconds of video outage, but migrating from the ML110 to the HP desktop just resulted in a slow hiccup. My assumption was that this would probably be completely invisible on more modern hardware.

So what does this mean? Microsoft has made Live Migration and Clustering a feature available to everyone, at (almost) no cost. Administrating such a cluster requires Active Directory, and either a WS08R2 server or a Windows 7 machine with RSAT installed.

This means we can finally have decent virtualization features without paying thousands of francs in licensing fees. I hope this makes it possible to create a few virtualization projects for our customers, which are mostly in the small business range.

Hyper-V Server R2 should be available around mid-August, at which i’ll need to rebuild my Ghetto setup here. I’m of course hoping to get some more cash in order to move or internal virtualization setup from a single-host to a SAN-hosted cluster, but somehow i doubt that will happen quickly.

Update:

I’ve played around with Expression Encoder a bit, and created a Video of a Live Migration. I’ve put the probably most boring video on Youtube – Live Migration of Pinball.

So, IBM’s x3650 M2 have been out for some time and in the meantime i’ve deployed three of them – two with SBS 2008, and one with Windows Server 2008. No pictures, since the camera in the office is broken.

The x3650 M2 comes with a new IMM (Integrated Management Module) that replaces both the BMC and the RSA II Slimline. In order to get remote KVM capability, a physical licensing key must be installed into the server. Standard features like remote power on/off are available without the licensing key – which is more expensive than the physical RSA II slimline adapter. With the IMM also comes UEFI, as a replacement to the aging BIOS.

The x3650 M2 also gets rid of the Adaptec ServeRAID 8k controllers, and introduces us to the ServeRAID 10 series manufactured by LSI. The ServeRAID 8k series have been plagued by several extremely heavy bugs that never caused me loss of data (but several other people), but nevertheless cost me a lot of my nerves.

The two standard RAID controllers in the x3650 M2 are the ServeRAID BR10i, which is the baseline controller without BBWC and without support for RAID5. The ServeRAID MR10i is the better model, which comes with 256MB BBWC and support for RAID5/RAID6. Unlike the ServeRAID 8k/8k-l story, these are entirely standalone controllers, that are located in a special daughterboard position with a standard PCI-e x8 interface.

Both controllers support only 8 drives – in order to get 12 drives, you need a special enabler kit that comes with a SAS expander hard and several of other stuff that doesn’t look all that trivial. I’ve used such a kit, and so i can’t comment on how it works exactly.

The power supplies have gotten a lot smaller, the server seems to look a lot more organized, the 2.5″ SAS HDDs are no longer as finicky as they were in the x3650 and now fit very well into their slots, the Lightpath diagnostics panel now looks like it belongs to an expensive server and locks into place securely.

Of course, all the usual changes that come with the new Nehalem based Xeons, triple channel DDR-3 memory, both processors needed for using more than 8 slots, using to many memory modules will downgrade the speed, etc.

So much for the general rundown – now for my assorted observations:

• The ServeRAID BR10i seems to be a slightly newer variant than the SAS RAID Controllers found in the x3250 and x3250 M2. The configuration interface is simple, but it works reasonably well.
• The ServeRAID MR10i is a controller i haven’t dealt with before. It does not offer a standard character based interface for configuration, only a graphical interface called WebBIOS. It’s completely awful and half done – half of the buttons have no text on them, every button press requires several seconds until something happens. Configuring a controller with this interface requires you to guess actions based on the manual, since they aren’t labeled onscreen.
• As usual, both LSI based controllers use the MegaRAID management software. Compared to the old Adaptec software, it’s really awful. It runs extremely slow – even on these new servers, is much more complicated than the old ServeRAID software and offers less options in terms of notification.
• The IMM webinterface has gotten even slower than the already slow RSA II interface. Web2.0 style “loading” icons have been added, but viewing the status screen can take up to a minute now. This is retarded, and clearly a step back. At least IMM standard now comes with every server.
• The IMM’s KVM capability have gotten a lot better. Instead of a java applet running in the browser, a java application is launched using java webstart. While IMM itself is slow, the remote KVM capability is actually very fast, and even works with decent speed through a VPN connection.
• While the ServeRAID 8k with newer firmware usually spent 2-3 minutes looking for the drives, the new LSI based ServeRAID 10 series now only takes a few seconds. This is compensated by UEFI which now takes roughly 2 minutes instead of the 10 seconds the old BIOS took. With this, IBM is successfully keeping the server at roughly 4 minutes until OS boot
• The IMM connects to the OS using an USB LAN interface. This is a real problem on Windows, since it confuses the Windows Firewall (switching it into “Public network” mode) and the Windows DNS client. Install the driver and disable it in Windows. Ensure you never enable it on DCs! Run the IMM firmware updates from CD. Clearly a step back.
• The UEFI configuration screens act a lot slower than the old BIOSes. But the options available are decent.
• You can install Windows Server 2008 Standard in UEFI mode. I did that on our internal x3650 M2 which is going to run our WebFOCUS deployment
• You cannot install SBS 2008 in UEFI mode. It will work fine with legacy BIOS emulation, though.
• The IMM can’t send alert messages to email addresses with a – (Dash) in them. Retarded.
• The machine is extremely silent. Compared to the Power 520, which will kill your ears within minutes, they’re a blessing.
• Just like the RSA II interface, the IMM web interface has a tendency to lock up randomly and stop working. Requires a physical power cycle on the server to fix.

That’s it for now. Lot’s of negative stuff in here, but the machines are actually extremely good performers. I hope that IBM will fix those outstanding bugs soon.

I’ve had my share of experiences with ZyXEL equipment, like the ZyWALL vs. Exchange post i did a few years ago.

But today i experienced the most grave issue with their equipment that critically impacted a customers business.

The customer has two sites – an HQ with an SBS 2008 and a branch office with two Lenovo SFF machines running Windows Vista Business. Both sites are using 20/2 VDSL lines from Swisscom, with ZyXEL P-2802HWL routers.

There is an IPsec VPN configured between these two sites. This has been working fine since January.

Now, about a month ago a telecom service company installed VoIP telephones in the branch office, and enabled QoS on both ZyXEL routers.

Since then, Outlook was unable to synchronize correctly with the SBS server. Unfortunately, the customers personnel isn’t that technically savy, so they weren’t able to tell that they had a problem – because smaller e-mails were able to successfully synchronize, but larger ones failed. This led to very inconsistent states of the OST files, with some mails there and some mails not there.

When i arrived at the branch office i didn’t have a single clue what the issue was or may be. At first i suspected an Outlook problem, so i deleted the OST file. But from there on, nothing happened – Outlook wasn’t able to download anything.

Next, i tried to copy a 50kbyte Excel file from a share to the local computer. This worked. So i tried a 2 megabyte Word file. This failed about halfway through, with Explorer just hanging there and doing nothing. From that point on, i suspected a network issue, but the fact that copying a 50kbyte file worked and a 2 megabyte file didn’t was very odd.

Using Outlook with Outlook Anywhere also worked (when the VPN tunnel was downed).

Whenever i’m confronted with strange network problems, i suspect MTU issues (which was my first “real” network problem i solved back on my first ADSL line – took me weeks for a simple fix). ping -l 5000 CUSTSBS01 worked. ping -l 15000 CUSTSBS01 worked, too. So thought it wasn’t an MTU issue.

Disabling QoS on the ZyXEL router fixed the issue, but made the phones unusuable while Outlook was filling it’s OST files.

So i ran through the usual check points – tcp checksum offloading, chimney, receive window autotuning, reboots, etc. Nothing helped. At the end i was just changing network settings at will. But nothing helped.

Out of any reasonable ideas, i changed the MTU to 1300. That fixed it – with QoS enabled and the NIC MTU of the two machines, everything was working as it should. File transfers worked, Outlook worked, Phones worked.

Don’t buy ZyXEL.

With the release of the new generation of System x servers, IBM also revamped it’s tool offering.

Central point of the new IBM offering is the ToolsCenter, which serves as a starting point for all important IBM tools.

The two most important tools, which every admin dealing with IBM System x servers should know are now available in new versions, which offer improved functionality.

UpdateXpress System Pack Installer

UpdateXpress is now available in version 3. Pictured to the right is the new user interface, which offers much needed improvements. The previous versions looked like a leftover from the cold war.

UpdateXpress allows you to update all your System x drives in one automatic swoop, without the need to meticulously check the IBM web site for newly released drivers.

Dynamic System Analysis

DSA is now available in version 2.20. While the handling of the tool hasn’t changed much, there is now a 64bit version available. A few bugs i’ve encountered on 64bit systems are fixed with this new release.

Today i attended an interesting product presentation from IBM, about their Nehalem product line up. In addition to the information i’ve already gleaned from the IBM web page, i’ve learned several things that are equally important.

If you’ve never heard about the x3650 M2, i suggest you to read my introduction post first.

Planned availability, other new products

Planned general availability (GA) of the new x3650 M2 is the the 20th of April, or in pretty much 3 weeks. Orders, configuration and pricing is available – so if you want to buy a new server now and can wait for three weeks, you should order a x3650 M2.

At the end of April, the M2 versions of the x3400 and x3500 will be announced. Judging from the current timeline, this will put GA of those new products near May.

4 CPU machines are planned for Q1 2010. I don’t care much about those since we run all our heavy DB workloads on IBM POWER.

Positioning of the new Intel 5500 Xeons

Just like the Core Microarchitecture brought many changes to servers, the new 5500 Xeons bring even more changes. It’s especially important for system administrators to understand the differences, and even more important if you’re selling systems to customers.

Here are a few key differences:

• Memory speed depends on the CPU purchased and the amoung and type of memory installed into the server
• Memory slots are only usuable if the associated CPU is installed
• HyperThreading is reintroduced in 2/3rds of the CPUs – systems will show twice the amount of logical processors
• TurboBoost is a new functionality that allows the CPU to run at higher clock speeds, depending on load and cooling

To the right you can see Intel’s official spec sheet. Intel introduces a “garbage bin” of CPUs that you should never use – the E5502 / E5504 / E5506 models. These CPUs do not support HyperThreading, TurboBoost, 1066 Memory and only have 4 instead of 8 megabytes of cache. Make sure to use E5520 or faster CPUs to ensure best performance. The performance difference between an E5506 and an E5520 is 15-20%, while the price difference is much smaller! In my opinion, the E5520 is currently the sweet spot between price and performance.

Order the right memory configuration

With FB-DIMMs, memory configuration was simple, because FB-DIMMs were slow no matter which way you put it. However, with the new integrated memory controllers, memory of much higher speed is now available. Now, as a technician or sales things will get more complicated.

I wrote in an earlier post that i didn’t understand why IBM only put 16 DIMM slots into the machine, while HP installed 18 DIMM slots – the reason is that in most cases it makes little sense to populate all DIMM slots, because this will heavily reduce the bandwidth available, as the memory must run at lower speeds.

While HP has decided to offer registered and unregistered memory for their DL3xx G6 models, IBM only offers registered DIMMs. The x3650 M2 COG Guide offers a lot of in terms of possible configurations. The most important thing is keep the numbers of memory modules down – this makes it easier deploy them correctly.

Another important part is that memory is no longer ordered in pairs, but again separately as it was a few generations before. Most servers ship with two 1GB modules standard. This is not an optimal configuration, since you have three channels that could be used.

There are two ways to deal with this: either add a third 1 GB module and then add the rest of the memory you need. Or discard the two memory modules that come with the servers and just install the higher capacity modules you bought. My recommendation would be to discard the 1GB modules and install three 4GB modules – for most SMB environments, 12GB of memory suffices for almost all services.

Other part changes

The onboard RAID-Controller is gone, there is a new specially positioned but otherwise standard PCI-E slot for the RAID controller. The system ships with a non-BBWC controller called the BR10i. In most cases, it makes sense to replace the standard RAID-Controller with a ServeRAID MR10i.

Some configurations also require an enablement kit to drive all 12 disk slots. Only 8 disk slots are standard.

The RSA II Slimline is gone. IMM now offers a lot of the RSA II functionality by default, but the most important functionality, remote KVM still requires a so called “Virtual Media Key” (as it enables Remote Media as well). In general, if a customer has used RSA II Slimline up to now, also include a Virtual Media Key. This enables full IMM functionality.

SSD offerings have also been added. Currently, the pricing for the 50GB SATA SSD is 3970 CHF in SSCT. My assumption would be that this is a pricing error, but i’m not too sure about that.

Pricing

Pricing hasn’t changed much. DDR3 memory seems to be a bit more expensive than the FB-DIMMs were, but that’ll pass as soon as DDR3 volume ramps up. I’ve created a sample config that is probably valid for most SMB deployments that clocks in at around 10k CHF. This is roughly the same as it was before with the standard x3650, except that the x3650 M2 will deliver a lot more performance for the money.

With yesterdays official announcement of the Nehalem Xeons, all major server manufacturers announced their new product lineup. HPs new announcements are equally interesting as what IBM has announced.

HP DL380 G6 Quickspecs
HP DL360 G6 Quickspecs

There are some interesting things to see there – unlike IBM, HP chose to go with 18 memory slots, allowing each CPU’s three channels to drive three memory modules. HP also offers a flash-based Memory configuration tool, that alleviates having to learn complicated placement rules.

HP also offers more disks per system, allowing a DL 380 G6 configuration with 16 2.5″ drives in 2U server. The DL360 G6 can have up to 8 2.5″ drives – that’s the same number of drives that previous generations had in a 2U form factor!

Having the ability to add lots of local storage is a good thing in smaller businesses. Larger companies probably won’t need that as much. Interestingly, HP choose to stay with a traditional BIOS with their G6 series. IBM is already offering UEFI on their x3550 M2 / x3650 M2.

Another new thing that i gleaned from those documents is the ability to choose a power supply based on the expected load of the server – making it possible to favor higher efficiency (for example, if you do not intend to have local disks in the server, as you’re planning to use your SAN).

A few days ago i bought an i7 920 for my computer at home. The performance is absolutely astonishing, compared to the AMD X2 CPU i had before. Especially VMs perform very well on it.

The first Nehalem-based servers have already been released, and as always IBM is a quiet compared to other server manufacturers. However, today while looking at some support documents i saw that IBM finally offers downloads of IBM x3650 M2 and IBM x3550 M2 documentation:

x3650 M2 Installation Guide
x3650 M2 Service Guide
x3550 M2 Installation Guide
x3550 M2 Service Guide
IMM User Guide (RSA II Successor)

All Documents on Scribd

I’ve now skimmed through the documents a few times and found several things worth mentioning – please note that i haven’t seen the actual hardware, so there may be mistakes here. If i’m aware of the, i’ll correct them of course.

RSA II Replaced by IMM

The RSA II adapter has been succeded by the IMM. The IMM is integrated into both the x3650 M2 and the x3550 M2, but certain features need to be enabled by purchasing a “Media Key”. This is similar to how HP handles this issue, except with a physical key. The Media key enables additional features, as described in the IMM User Guide:

In addition, IMM Premium has the following features:
- Remote presence, including the remote control of a server
- Operating-system failure screen capture and display through the Web interface
- Remote disk, which enables the attachment of a diskette drive, CD/DVD drive, USB flash drive, or disk image to a server

More Disks

The x3650 M2 will support up to 12 hard disks. The x3550 M2 will support up to 6 hard disks. Only 2.5″ of course. I currently didn’t see much mention of SSD, but i’m sure both these machines will see SSDs during their lifecycle.

There are also several changes regarding the hard disk controller. The Onboard ServeRAID 8k made by Adaptec is gone. It will be replaced by MR10i and BR10i ServeRAID Adapters, made by LSI Logic. For more details about those controllers, see the ServeRAID Quick Reference. These adapters will fit into a special slot, and are no longer fully integrated into the system board.

This is a good thing – RAID controller troubles in the x3550 and x3650 always meant replacement of the system board, which could be problematic because it changed the various IDs of the system (most notably the MAC adresses).

More Memory

Nehalem brings CPUs with On-Die memory controllers to Intel-based servers. If you’ve never handled AMD servers before, this means a few changes.

* Memory is local to the CPU, meaning you will need both CPUs if you want to use the full 16 memory slots.
* Nehalem features a triple-channel memory subsystem, making memory placement interesting
* If you’ve wondered how triple-channel and 16 slots works, see the graphic to the right

More Ethernet ports

The x3650 M2 comes with 4 Gigabit ethernet ports. The x3550 M2 with two, but an option for an additional two without requiring a full pci-e slot. This is especially good when you’re using virtualization and need to segregate hosts or need the extra bandwidth.

UEFI & other features

UEFI is the successor to the common BIOS. This will get interesting – i have no real idea what to expect from this. I only know that WS08 supports UEFI boot directly, and that IBM will provide emulation for legacy operating systems.

There’s also an in-system USB port for running a Hypervisor directly from flash. The system also features an integrated TPM module, which can be used together with Bitlocker to provide full disk encryption in physically insecure servers.

An important Whitepaper from IBM about Transitioning to UEFI is also available.

NOS support

Both new systems have also appeared in IBMs NOS support list. While full Windows support is offered, it looks like VMware will not be supported, at least initially.

Support for high performance graphics cards

The system board lists an 8 pin power connector for high performance graphics cards. The parts list also lists NVIDIA FX1700 and NVIDIA FX570. This might be needed for CUDA or similar applications.

Conclusions

I want one. According to my sources, they should be available for sale starting on the 21st of April. As of today, the System x New Generation information page is finally online.