Al Mulnick wrote:
There would be a little more to gain than that but often that's the
reason. joe might point out that a two mirror configuration is not his
optimal configuration. I'm pretty sure he'd also point out that compared
with software raid, that he'd take that option. :)
I can honestly say I'd agree with him on this one. Software mirroring
for this type of application is never a good idea. The slower spindle
speeds likely won't be enough of an issue to matter in your
configuration. Unless you have a very large DIT <queue jokes hereor
applications that pound the snot out of the individual servers spindle
speed won't be nearly as important. Since it's 64 bit you're after,
spend some money on the memory and take advantage of the cache as much
as you can.
Al
6/22/06, *Noah Eiger* <neiger (AT) prbo (DOT) org <mailto:neiger (AT) prbo (DOT) org>wrote:
What would the partitions on the first configuration gain you (over
just a
single C:)? I thought the idea behind placing NTDS, etc on something
_besides_ C: was to get the performance benefits of extra spindles
(as in
#2).
The mirrors would be in hardware. Software raid - only time I've ever
lost customer data was due to software raid. Never again.
Splitting a large volume into two partitions gains nothing IM
Personally I like my databases on different spindles than the S.
We have some Unix based apps that hit the DC's pretty hard. Their use is
only going up. We should be able to fit the DIT in memory so I think
I'll push for that. Not sure it's a battle I can win but it's always fun
to try.
thanks, (the other) al
al.lilianstrom (AT) fnal (DOT) gov
-- nme
Message
From: Al Lilianstrom [mailto:al.lilianstrom (AT) fnal (DOT) gov
<mailto:al.lilianstrom (AT) fnal (DOT) gov>]
Sent: Thursday, June 22, 2006 1:24 PM
To: ActiveDir (AT) mail (DOT) activedir.org <mailto:ActiveDir (AT) mail (DOT) activedir.org>
Subject: [ActiveDir] DC Configuration
We have some budget money to replace domain controllers this year. Not
all of them but probably half of them. We've pretty much decided on 64
bit Dell PowerEdge servers. Most of the discussion is about disk
configuration. Two schools of thought exist here.
1) 2x73GB 15K drives in RAID1. Carve up the volume at the S level with
20GB or so for the S and the remainder for NTDS, Sysvol, and system
state backups
2) Two sets of 2x73 10K drives in RAID1. The first set is for the S,
the second is for NTDS, Sysvol, and system state backups.
I've always liked physically separating the S from the application
data. here like carving up the volume at the S.
Any thoughts, opinions, suggestions?
tia, al

There would be a little more to gain than that but often that's the reason.
joe might point out that a two mirror configuration is not his optimal
configuration. I'm pretty sure he'd also point out that compared with
software raid, that he'd take that option. :)

I can honestly say I'd agree with him on this one. Software mirroring for
this type of application is never a good idea. The slower spindle speeds
likely won't be enough of an issue to matter in your configuration. Unless
you have a very large DIT <queue jokes hereor applications that pound the
snot out of the individual servers spindle speed won't be nearly as
important. Since it's 64 bit you're after, spend some money on the memory
and take advantage of the cache as much as you can.

Al

6/22/06, Noah Eiger <neiger (AT) prbo (DOT) orgwrote:

What would the partitions on the first configuration gain you (over just a
single C:)? I thought the idea behind placing NTDS, etc on something
_besides_ C: was to get the performance benefits of extra spindles (as in
#2).

-- nme

Message
From: Al Lilianstrom [mailto:al.lilianstrom (AT) fnal (DOT) gov]
Sent: Thursday, June 22, 2006 1:24 PM
To: ActiveDir (AT) mail (DOT) activedir.org
Subject: [ActiveDir] DC Configuration

We have some budget money to replace domain controllers this year. Not
all of them but probably half of them. We've pretty much decided on 64
bit Dell PowerEdge servers. Most of the discussion is about disk
configuration. Two schools of thought exist here.

1) 2x73GB 15K drives in RAID1. Carve up the volume at the S level with
20GB or so for the S and the remainder for NTDS, Sysvol, and system
state backups

2) Two sets of 2x73 10K drives in RAID1. The first set is for the S,
the second is for NTDS, Sysvol, and system state backups.

I've always liked physically separating the S from the application
data. here like carving up the volume at the S.

Any thoughts, opinions, suggestions?

tia, al
--

Al Lilianstrom
CD/CSS/CSI
Al.Lilianstrom (AT) fnal (DOT) gov

List archive:
--

We have some budget money to replace domain controllers this year. Not
all of them but probably half of them. We've pretty much decided on 64
bit Dell PowerEdge servers. Most of the discussion is about disk
configuration. Two schools of thought exist here.

1) 2x73GB 15K drives in RAID1. Carve up the volume at the S level with
20GB or so for the S and the remainder for NTDS, Sysvol, and system
state backups

2) Two sets of 2x73 10K drives in RAID1. The first set is for the S,
the second is for NTDS, Sysvol, and system state backups.

I've always liked physically separating the S from the application
data. here like carving up the volume at the S.

Any thoughts, opinions, suggestions?

tia, al

Yeah, I threw that in there mostly for your benefit [1]
That's correct, you cannot repeatedly build a successful system based on a
single datapoint. Go big and you might get lucky, but that's not the science
part of computer science.

The biggest issue I tend to see is echoed across many many folks that design
everything from datacenters to assembly lines. Spend the most time defining
your requirements. Typically that means figuring out the usage scenarios
and defining the load on the components of the system and the system as a
whole. Inevetibly, you come to a point where you're guesstimating the amount
becuase there is just a certain amount of historical data and just so much
estimation you can do on a variable such as the future usage. You try and
you do the due diligence, but eventually you estimate high.

Load is far more important than number of users in determining a proper
installation. I get that. I was just questioning why anyone would suggest a
particular layout before hearing that information? I'll be more direct next
time :)

[1] and Deji. But he acts like he's been busy lately.

6/23/06, joe <listmail (AT) joeware (DOT) netwrote:

Number of users isn't critical, it is how the system is used. While it
would be odd for a 500 user system to take a beating, I don't think we could
rule it out until you understand how the system is used. Any designs that go
off of user count and nothing else is going to be flawed. Without the
details, the recommend from me is to go as big as you can. If that doesn't
end up being big enough, at least you tried and now you don't have as much
more to buy now. :)

So why not a RAID 1 partition that holds all the S, binaries, log
files, file and print facilities etc?

For a low level use, I was right there with you until you said file and
print my friend. ;)
>
>
>