The Complete Geek

Fist off we ordered a new Dell blade chassis with 8 M605 servers to start off. (Various configurations). We also ordered it with 2CMC’s 6 PSU’s and 4 Cisco 3130G’s

First impressions and cabling considerations

Well t thought out design, it was fairly easy to mount (a bit heavy) But once mounted cabling, fan placement and power supply placement made it easy to cable in. All devices felt good to dock and locked securely into place with various methods that you could feel secure that they are and will remain docked.
We stayed away from a lot of their cable management “gadgets” that shipped with it we simply ran everything vertically. With our current configurations there were only 12 cables from the switches and 2 from the CMC’s (will discuss some of the networking considerations later) but without pass-through modules the cabling was easily done with some standard Velcro straps

Power Redundancy

We chose to use all 6 Power supplies in “AC” redundancy mode with dells power management enabled, this would give us the redundancy that we were looking for while the ability (early on at least) to scale back the power supplies. We expected the Chassis to be full within the next 2-6 months depending on the direction we go with it.
Our cabling choices figured into the redundancy, we alternated all 6 power supplies across 2 power polls (We were running 30 amp 3 phase 208 power) this allows us to have PSU redundancy, PDU redundancy, PDU Breaker Redundancy, Panel Breaker redundancy, Panel Redundancy (2 Panels on our UPS) Right now the only non redundant systems are the UPS and the Generator, we have bigger issues if they drop

Alerting

Right now we are using the built in email alerting capability, which has the ability to relay alerts off of an SMTP server. In its final form we will be using Traps/SNMP to alert off of a Whats Up Gold box but right now because monitoring the environment (Temp Sensors, HVAC, UPS …) is all on that same box we need to get it migrated before alerting through there.
Using standard practices we are also Alerting via SMS as well (Standard email to SMS)
CMC configuration
CMC’s were deployed pretty standard they are going into two different switches on a stack of 3750’s that is used for monitoring. The CMC’s actually both use the same IP address in an “active, passive” redundancy mode and are statically assigned IP addresses

Networking

This is probably the most complex right now. We have 4 Cisco 3130G’s placed in fabrics A and B (C is not utilized at this time)
IO port A1 And B1 are stacked together with 4 cables to our “core” network these are trunked to our core and port channeled. These two switches are also stacked for redundancy. They are also cabled into two different switches of the core network. It makes it a fairly redundant design that allows for a lot of failure including an entire fabric.
IO port A2 and B2 are our two ISCSI SAN networks, they are NOT stacked as they are redundant networks (They are port channeled further up on the network so cross communication is possible) We have 4 cables to each of the 2 Chassis switches
Current configuration gives us plenty of bandwidth to the blades, no need for the C fabric at this point. Though we are concidering using a high end M605 or the new M905 for an Vmware ESX host. If we choose to do that to get 4Gbps (with MPIO) we will need to either go to the C fabric for the SAN network or use the M905’s which allow you to have 2 cards on each fabric.
Our testing suggests that we are close with the current 2Gbps to the performance we will need with the current number of virtuals that can be hosted on a high end M605.

IDrac, IKVM, Virtual Media and Server Control

One of the better features for these servers is the possibility of remote control for the devices. From their web based interface you are able to set everything from boot priority to launching an actual console.
Some of the better features I have found are
Virtual media: Good design able to mount ISO and IMG files as virtual CDroms and Floppies. It makes it simple to administer and get things done without hunting around for a USB Cdrom drive.
Boot Screen recording: The console actually records the last boot sequence and allows you to reply even if the machine is already booted up. Helpful to check things like raid status or other boot errors and allows you to see it both in motion as well as still images.
Console: The server console is of course the best part the ability to manage the server as if you were sitting in front of it is great. The fact that we have the ability to terminal server (through a rather round about method) into a box that has access to the CMC/IDRAC network means you can administer the machines from anywhere in the world, combined with virtual media there is almost no need to go down into the data center.

The Bad

Well it can’t all be perfect there are a few things so far that have annoyed me
1) The M605’s have been a royal pain to build boot disks for. I have tried every copy of both 32 and 16 bit environments that I can get my hands on to get a boot environment that can load ghost. Our current DR scheme requires that we be able to Ghost the blades short term until a new online backup strategy is implemented. The primary problem appears to be hanging on USB driver load, and because of networking configuration choices this is necessary to have.
2) Blade rails, it is rather easy to slide a blade in incorrectly … getting it on the rails is not to hard if you pay attention but it is something that could be improved upon
3) Lack of 4 port Mezzanine cards, this will be corrected soon, they are planned to release in the not so distant future but for heavy SAN connected blades without going to a C fabric the bottle neck is not the switching but the ports to each blade.
Overall a good product but there are things that I am sure we will run across as we continue to deploy