SBS 2008 and Dell Rack-mount Servers
ByYesterday I finally ran into my first failed SBS 2008 install, not one I had picked up from someone else, but one of my own where I’ve managed the source server for years and did all the prep work myself. It was both a good thing and a bad thing: good because we got to finally troubleshoot one where we knew the entire environment up front so we could immediately eliminate several potential causes, bad because we were on a really limited migration timetable for this particular customer and we’ve ended up losing more than just a day on the project.
But the interesting thing about this migration is why it failed, and the blame lies right at the feet of Dell. Short story: the Dell PowerEdge R310 server does not allow you to individually disable the on-board NICs in BIOS. In BIOS, you can choose to disable both NICs, but the individual integrated NIC options include “Enabled,” “Enabled with PXE,” and “Enabled with iSCSI.” No “Disabled” option at all.
I’ll be perfectly honest, this was not a scenario I tested when working on the migration documentation with Microsoft and researching for the SBS 2008 Unleashed book. Since the SBS 2008 requirements list a single NIC only, that’s all the testing that I did. So when I went through with the migration install yesterday, I knew I was taking a bit of a chance, but hoped that since I only had one NIC connected to the network it wouldn’t be a problem.
Well, it was.
During the setup, the SBS portion of the install tried to do a WMI query against DNS for the existing domain. The query succeeded on the connected NIC, but the installer performed the query again on the disconnected NIC and failed. That failure led to the dreaded cascading failure of Exchange and everything else that followed. We were able to get the source server back online quickly since we followed best practices and took a System State Backup immediately prior to launching the SBS 2008 installer on the destination server, but then we faced the dilemma of how to proceed. After discussing the issue with a Microsoft contact, we thought the setup might complete if we connected both NICs to the network, but the better option is to disable the unused NIC in BIOS and do the migration setup the way it’s supposed to be done.
After a 3.5 hour call with Dell, it basically cannot be done on the R310. Apparently someone decided that an all or nothing configuration on the NICs on that particular server was the better solution than letting each NIC be individually disabled as done on every other Dell server I’ve worked with for a decade. The issue has been escalated with the engineering team, but we don’t yet know if there will be a fix or how long it will take to get one if it can be done.
Bottom line, I cannot recommend installation of SBS 2008 on a Dell PowerEdge R310 server until (or if) Dell resolves the issue of disabling the NICs individually in BIOS. I can almost guarantee that a migration setup will fail on this platform, but I don’t know if a clean install will have a similar issue or not. If anyone has successfully installed SBS 2008 on an R310 server, I’d love to hear from you. Since this is a relatively new model from Dell, however, I may well be one of the first to attempt this particular configuration. And I hope that our pain can save someone else from the same…
6 Comments
May 22nd, 2010 at 1:46 pm
Eriq,
The workaround is straightforward: disable both onboard nics, and install a PCI[eX] nic.
I’d bill Dell for said extra nic, of course (or try and get them to provide it FOC).
Steve
May 22nd, 2010 at 2:59 pm
Steve – Problem with that approach (which I do on every other Dell server we’ve deployed) is that the R310 has exactly one PCI slot. So you can have hardware RAID or a NIC. Not knowing this would be an issue, it seemed an easy option to do the PERC controller over an Intel NIC.
I’ll be posting an update with the workaround we used that will now be part of our standard Migration process. We were able to get the migration completed on the second pass with little incident.
September 30th, 2010 at 10:34 am
Hi Eriq,
We have just completed a new install on a Dell R510.
We are migrating the client from a non-sbs 2003 domain to SBS 2008, and while the server itself is ok, and had no real problems (apart from the migration being more intricate than planned ).
However, while the server is generally ok, I have NEVER heard a server so loud under so little stress or temperature! The fans just kick in randomly and spin for about 5 mins then slow down. When we are working in there is drives me nuts. Even the 10 year old HP’s I’m taking out down run that way.
I think it’ll be the last Rxxx I’ll out in for a while as we never had any issues with any other Dell PE server we have put in
January 5th, 2011 at 1:23 am
A small update on the Dell Server issue: the problem is compounded in the 710 series, which ships native with two pairs of Broadcom Gb dual port NICs embedded [i.e. 4 NICs in total]. These are paired on 2 discrete embedded chipsets and neither element of any pair can be disabled individually; both ports, or none. So far rather similar to the 310.
However it gets slightly worse: there is some dynamic allocation going on in firmware.
The ports are labelled externally from 1-4.
If one is doing an alternate deployment strategy using SBS Premium, so as to create a Hyper-V host, with SBS itself running as a virtual, it’s “logical” [if one hasn't been through the loop of discovering that things screw up as the 310 series does] to connect only NIC port 1 and complete the base install; then connect additional NICs and run the Hyper-V installation as normal: that’s the normal Hyper-V install routine. So far so good: in theory.
In practice what you are quite likely to discover is that the NIC ordering has been dynamically re-allocated in the OS. What was seen by the SBS host as NIC 1 in Network Management console prior to installing Hyper-V is quite likely not to be NIC 1 afterwards; so far I’ve seen it presented as NIC 2 and NIC 4. This doesn’t affect any actual operations of Hyper-V, or of the host system, the OS seems to be “smart” enough to know which NIC it needs to talk to, and all NIC configs that were set on the “original” NIC 1 will be unaffected, but it can make a real mess of any logical PS scripting or logging one might be doing to monitor/manage NICs by name, and can cause significant confusion when attempting to work through any MS Virtual Switch issues if one hasn’t previously encountered the issue and isn’t aware that the NIC numbering may bear no logical relationship to the port numbering on the back of the machine.
There is of course no guarantee that any subsequent changes to the OS, or to bios settings will not trigger another dynamic reconfiguration. Which is slightly worrying.
After several hours on the phone to Dell they are intransigent – “It’s not a problem!” – clearly they’ve never had to deal with virtualisation issues on SBS [it's not _that_ common as a deployment scenario]. Fortunately, unlike the 310, the 710 series does have several PCI-X ports, even with a pair of SAS controllers installed, and Intel dual port server cards are relatively low cost.
Be warned guys: Dell mid-range n10 servers aren’t quite as straightforward to deal with as one might hope, or as straightforward as their predecessors once were.
May 5th, 2011 at 12:03 am
Oh how I wish I had seen this. I have two sites where we have Dell T310′s and I had to install SBS2008 from scratch. Its a pain – The most recent was two weeks ago, on a brand new box. Called Dell to find out how to disable second nic – “no can do – it will be ok – just disable on OS post install” was their reply. It isnt. I have had problems with various aspects of the boxes since
May 11th, 2011 at 1:12 pm
Q,
We’ve encountered exactly the same issue, although prior to failure of migration, whilst working through an sbs201 split migration. You mentioned a work around above, would you be able to point me in the right direction for this?
thanks
W