Subject:
|
New LUGNET server hardware details
|
Newsgroups:
|
lugnet.admin.general, lugnet.admin.finance, lugnet.general, lugnet.lego.direct, lugnet.robotics, lugnet.trains, lugnet.technic, lugnet.space, lugnet.castle, lugnet.org, lugnet.announce
|
Followup-To:
|
lugnet.admin.general
|
Date:
|
Sat, 28 Feb 2004 21:29:01 GMT
|
Reply-To:
|
MATTDM@ihatespamMATTDM.ORG
|
Highlighted:
|
!!
(details)
|
Viewed:
|
117 times
|
| |
|
|
So, we've purchased our cool new servers, from Cambridge, MA vendor "PCs for
Everyone". Todd and I and several other friends have gotten systems from
them before, and have been extremely pleased -- they know their stuff,
including Linux, which is key, because Linux is what we'll be running. (More
on that below.)
Both systems are identical, which will both make testing more realistic and
make it easier to switch roles if something happens to the primary server.
The specs:
SuperMicro 1U Rackmount System -- Chassis and Motherboard
2x Intel Xeon 2.66Ghz Processor w/ 512MB Cache *
1Gb PC2100 DDR266 ECC REG Memory (2x512MB Modules)
2x Seagate 120GB SATA 7200RPM SerialATA HDD
Intel 10/100 and Intel 10/100/1000 Ethernet Adapters
Integrated Video Controller
Floppy disk, CD ROM, and so on, of course
Red Hat Linux 9 installed, but will be replaced with BU Linux.
Also, we got an extra hard drive, memory module, and power supply -- the
three most likely things to go horribly wrong in the middle of the night.
(These will be stored on site.)
The systems are scheduled for delivery this coming Friday. And actually,
I've had stuff from PCsFE show up before the scheduled date fairly often, so
we'll see. In any case, I'm out of town for that weekend, so it'll be a
little while before we can get really started with the setup. (And
lugnet.com itself will continue to run at Pair for the unspecified future,
until we're convinced our new setup is solid.)
* Actually purchased with only once CPU now; we will get the second
processors shortly. See below.
We're thinking calling the systems "brick" and "beam" (the devel server is
technic, of course, and the primary is a trusty 2x4). These are just
internal names that will only be seen inside the servers, so it's not really
a big deal.
This message gets increasingly geeky from here on out. Feel free to stop
reading if your eyes are glazing over. Really geeky replies should be
directed to .off-topic.geek, since there is no .on-topic.geek.
You'll notice that we have almost half a terabyte of storage. This is hard
to avoid with today's hard drive prices, and is actually a good thing. We'll
use all that extra space to provide redundancy and staged backups -- giving
us safety nets for both hardware failure and for data corruption or
accidental deletion. Plus it does give us plenty of room to grow -- by the
time we're running out of space, it'll be time to get new drives anyway. In
the meantime, we'll use the disks to make sure the chance of downtime is
very small -- let alone data loss.
The server motherboards used support dual Xeon processors, but we only
ordered one each for now. It'll be an easy and cheap upgrade when the time
comes. "Xeon" is basically Intel's marketing term for the current
server-targeted version of their Pentium line -- at this point, of course,
that's Pentium 4. The primary difference is that the Xeons are certified to
run in a dual-CPU configuration, and the performance benefits for that will
be significant.
Xeons also have a feature Intel calls "hyperthreading" -- something just now
coming to the desktop-oriented Pentium 4 line. Modern processors are divided
up into specific sections which each handle different kinds of instructions.
Hyperthreading attempts to keep each of these subsections busy more of the
time by telling the operating system that it its *two* processors, and that
it can do two things at once. This works great if the two things use
different parts of the CPU, but is a problem if they both need to be handled
in the same place. So, some types of programs end up with a speed increase,
while others are dragged down. We'll have to do some testing to see what's
best for LUGNET.
The motherboards also support a maximum of 8GB of memory. We're using up two
of the available slots with half-a-gigabyte DIMMs -- that way, if one goes
bad, we can test more easily. When it comes time to upgrade, the memory
isn't required to be in pairs, so we could add another 1GB easily and still
have room for 2GB more after that. And, as with the hard drive, by the time
we're running against that limit, time for new hardware anyway.
The infrastructure where we're putting the machines is 100Mbit. That's
plenty, plenty fast. The systems have two network interfaces, though -- one
100Mbit maximum, which we'll plug into the wall, and one 1000Mbit, which
we'll use to hook the two systems together directly, creating a private
high-speed link which we'll use to send data back and forth like crazy as
part of the above-mentioned insanely redundant backup scheme.
I would have taken the floppy drive out of the config if that were an
option. I hate floppy drives. But I guess if something were to happen to the
CD and we needed to boot to a rescue disk, it could save us.
And on the non-hardware side: we're going to install Boston University Linux
<http://linux.bu.edu/> on the system, and then remove a lot of BU custom
configuration. This is actually easier and nicer than basing it on the
default Red Hat Linux install and customizing from there, since a majority
of the stuff in BU Linux is changes we want anyway.
--
Matthew Miller mattdm@mattdm.org <http://www.mattdm.org/>
Boston University Linux ------> <http://linux.bu.edu/>
|
|
Message has 4 Replies:
 | | RE: New LUGNET server hardware details
|
| Ok, first time I've posted to off-topic, so hopefully I put in the TO header properly :) Matthew, I just want to point out one thing to make sure of with the two processors you are getting now. Make sure the stepping is same so when you go to dual (...) (21 years ago, 28-Feb-04, to lugnet.off-topic.geek)
| | | Re: New LUGNET server hardware details
|
| (...) Interesting that the HDDs are serialATA rather than SCSI. However from my experiences this week no bad thing. One server with dual HDDs in a mirrored RAID config had the SCSI controller backplane fail. However replacing it lost all the data on (...) (21 years ago, 29-Feb-04, to lugnet.off-topic.geek)
| | | Re: New LUGNET server hardware details
|
| (...) I hope that you are planning to use this extra storage for mirroring (I forget the numbers... RAID-1 ?). Mirroring will reduce your critical path failure from a HD, but will (remember statistics) halve the MTBF. With one drive at 50k hours (...) (21 years ago, 1-Mar-04, to lugnet.admin.general)
|
17 Messages in This Thread:
  
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
