On Feb 6, 2008, at 5:45 PM, Need Coffee wrote:
Does anyone run OpenBSD on blade servers? I don't mean
"Sun Blade 150" kind of hardware, but rather blade chassis
with server blades (a la Sun Blade 8000, HP, Dell, etc.).
I've been running FreeBSD on an Intel blade chassis with varying
amounts of success.
The biggest problem is that the interfaces, bge(4), bce(4), aren't all
that well supported. Just unsupported enough that the Var I went
through had to delay delivery by 3 weeks while a driver was written
from the ground up. Even then, the only supported speed was
1000baseSX, in full duplex. It's not that big of an issue, but under
load I've got worries the interface will drop out.
The next issue is cost. Each blade runs around $4000, USD. A little
more, actually. The chassis needs 208v in the cabinet, and draws a
huge amount of power. It's expensive to run, basically.
So, on to the good.
If you fill out the entire chassis (14 blades, 2 switch units and one
MM), you have a good farm of systems sitting in a small space. It
really does shove a bunch of systems in to 5U. The units themselves
are pretty powerful (2 AMD64 capable systems with twin cores, so 4
available CPUs. 8 gb of RAM, RAID 1 on two 80gb SATA drives, etc).
But, the best aspect is the MM. The MM gives you a fair amount of
control over the chassis itself, and each system hosted on it. A java
based KVM permits you to move a virtual keyboard around between
machines easily. The catch is that you don't know if that will cause a
kernel panic (it seems to happen ever 2 or 3 times on freebsd) or not.
The internal switches do full 802.1Q VLANs, manage through an IOS/
CatOS-like language and a web interface available from the MM. All in
all, the system is very powerful, and you get an okay bang for the buck.
Expect to spend $12k or so, right away, since unless you at least fill
the chassis 1/2 of the way you're taking up more space and power than
you would with 7 1u systems. The blade chassis units themselves are
expensive since the Switch Module and MM don't come standard.
Custom power requirements are expensive: $500/drop for the 208v
install from a DC was the average quote, and we're paying around $300/
mo in power alone. But, we also have a normal 110v drop in to the same
cabinet.
As for running OpenBSD on them, the first question you need to ask
isn't even related to the OS.
Will this thing pay for itself, and save me money in the long run? If
the answer is "no." Don't buy the chassis.
Will you maximize your space usage (fill it out, to at least 7 units)?
If the answer is "no." Don' buy the chassis.
Are you so cramped in space, and will an extra cabinet (or rack in a
cage) cost you MORE than putting 14 servers in to 6 or 7 U, as a one
time cost? If the answer is "yes," go for it.
If the decision were mine to have made when my client went with one
(before I came aboard) they'd have saved themselves around $20k over
the last 3 years.
When looking at the expense of colocation, I've found that adding a
cabinet and adding power, running cable to the new cabinet, is cheaper
than handling a blade system. For that $25k or so building out the
chassis, i can add in a full cabinet and run it for a year or more.
I'd appreciate any details... I'm having a bit of trouble finding
anything conclusive about OpenBSD on blades.
Hope this helps.