Has anybody here taken a HDD configured with an Opteron system, and then put it into an Athlon/64 and had it work?
Are they interchangeable, like an Athlon/32 and a P3/P4?
-Ben
Benjamin Smith wrote:
Has anybody here taken a HDD configured with an Opteron system, and then put it into an Athlon/64 and had it work?
Are they interchangeable, like an Athlon/32 and a P3/P4?
-Ben
From a processor standpoint, it should just work. You may have driver issues because of the different motherboards, but assuming that both boards are using common components supported in the default kernel, you shouldn't have any problems with a switcheroo.
Cheers,
On Wed, 2005-12-28 at 12:32 -0800, Benjamin Smith wrote:
Has anybody here taken a HDD configured with an Opteron system, and then put it into an Athlon/64 and had it work?
Are they interchangeable, like an Athlon/32 and a P3/P4?
-Ben
I would think that they would be ... within reason. (The first time you run it, there will be kudzu finding all kinds of new hardware).
BUT ... all the software that runs on x86_64 should work exactly the same on Opteron and Athlon64 ... so unless there is detection issues on bootup and some things are not detected correctly, it should work OK.
(This is my opinion ... I could be wrong :)
Johnny Hughes mailing-lists@hughesjr.com wrote:
(This is my opinion ... I could be wrong :)
There was virtually no difference between early Hammer cores of the Opteron and Athlon 64s. In reality, it's really about interconnect and the mainboard.
Opteron tends to lead in new core design changes, then Athlon 64 follows shortly afterwards. AMD is still perfecting optimizations and packaging with each new revision.
You'll always find the Opterons to be of higher quality than the Athlon 64s. And it wouldn't surprise me if some roll off the same fab lines -- even if the codenames for each respective product is different.
Benjamin Smith lists@benjamindsmith.com wrote:
Has anybody here taken a HDD configured with an Opteron system, and then put it into an Athlon/64 and had it work? Are they interchangeable, like an Athlon/32 and a P3/P4?
Yes, to a point. An x86 system will _not_ boot a x86-64 kernel. But yes, a x86-64 system _will_ boot a x86 kernel.
The x86-64 kernel puts the CPU into a 52-bit PAE memory mode. x86 systems only support a 36-bit PAE memory mode. For more, see my blog entry here:
http://thebs413.blogspot.com/2005/10/what-is-x86-64-long-mode-memory-model.h...
The only other issues are boot-time storage support. Most [parallel] ATA devices are no issue, because they are in the stock ATA/IDE kernel support. SATA is a different issue, because many SATA drivers are SCSI block drivers at this point, so they need to be built into the initrd (initial root disk).
Linux isn't like NT 5.x (2000/XP/2003), which sets boot-time information in the registry and will blue screen if you change mainboard/chipsets. Yes, not even changing the ntbootdd.sys file for the appropriate ATA/SCSI works (like it did for NT 4.0 and earlier). The only way to change that is to boot up another OS (e.g., Linux) with a registry editor and manually change the 3 or so keys for the boot-time storage device.
What it comes down to is this: so far, all the servers I've been administering have been 32-bit P3/P4/Athlon alikes, so if a server died and we needed it up NOW we could go to a local computer store here in smalltown USA, buy some desktop machine, swap harddrives, press enter a few times while kudzu does its thing, and have a working machine.
Now, we're moving to Opteron-based servers, and I just was wondering if it's reasonable to expect that, in a worst-case scenario, we could get an Athlon/64 system locally, and have it work, even if not optimally.
Obviously, the Opteron is better and faster, but if the Athlon 64 will run CentOS X86/64, then I can be pretty certain that in the worst case, I can run to the local Performance Leet g4m3rz store, and get an Athlon/64 to get a needed database server back online.
Just checking the accuracy of the data behind my decisions. (I've turned down Xeon servers for this reason)
-Ben
On Wednesday 28 December 2005 12:57, Bryan J. Smith wrote:
Benjamin Smith lists@benjamindsmith.com wrote:
Has anybody here taken a HDD configured with an Opteron system, and then put it into an Athlon/64 and had it work? Are they interchangeable, like an Athlon/32 and a P3/P4?
Yes, to a point. An x86 system will _not_ boot a x86-64 kernel. But yes, a x86-64 system _will_ boot a x86 kernel.
The x86-64 kernel puts the CPU into a 52-bit PAE memory mode. x86 systems only support a 36-bit PAE memory mode. For more, see my blog entry here:
http://thebs413.blogspot.com/2005/10/what-is-x86-64-long-mode-memory-model.h...
The only other issues are boot-time storage support. Most [parallel] ATA devices are no issue, because they are in the stock ATA/IDE kernel support. SATA is a different issue, because many SATA drivers are SCSI block drivers at this point, so they need to be built into the initrd (initial root disk).
Linux isn't like NT 5.x (2000/XP/2003), which sets boot-time information in the registry and will blue screen if you change mainboard/chipsets. Yes, not even changing the ntbootdd.sys file for the appropriate ATA/SCSI works (like it did for NT 4.0 and earlier). The only way to change that is to boot up another OS (e.g., Linux) with a registry editor and manually change the 3 or so keys for the boot-time storage device.
-- Bryan J. Smith Professional, Technical Annoyance
b.j.smith@ieee.org http://thebs413.blogspot.com
*** Speed doesn't kill, difference in speed does *** _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Benjamin Smith wrote:
What it comes down to is this: so far, all the servers I've been administering have been 32-bit P3/P4/Athlon alikes, so if a server died and we needed it up NOW we could go to a local computer store here in smalltown USA, buy some desktop machine, swap harddrives, press enter a few times while kudzu does its thing, and have a working machine.
Now, we're moving to Opteron-based servers, and I just was wondering if it's reasonable to expect that, in a worst-case scenario, we could get an Athlon/64 system locally, and have it work, even if not optimally.
Obviously, the Opteron is better and faster, but if the Athlon 64 will run CentOS X86/64, then I can be pretty certain that in the worst case, I can run to the local Performance Leet g4m3rz store, and get an Athlon/64 to get a needed database server back online.
Just checking the accuracy of the data behind my decisions. (I've turned down Xeon servers for this reason)
I'm sure you already know this, but....
if it's THAT critical of a machine then it would probably be in your best interest to have a spare or two. Whenever I stop using one brand of server and begin using another type (at least in rackmount datacenter situations), I always have at least one spare installed. On more than one occasion, I've had the power supply fail on a Compaq or HP rackmount machine or a motherboard failure. I just popped out the "hot plug" drive(s) on the stricken machine and plugged them into the spare system, flicked a power switch, and voila. At one of our datacenters, I actually have a Cisco 7513 that's been hollowed out and is a stealth storage bin for spare parts (the 7513's internals were immolated in a rather spectacular PS failure/fire). 8-)
Cheers,
Chris Mauritz chrism@imntv.com wrote:
I'm sure you already know this, but.... if it's THAT critical of a machine then it would probably be in your best interest to have a spare or two. Whenever
I
stop using one brand of server and begin using another type (at least in rackmount datacenter situations), I always
have
at least one spare installed.
I have to 2nd Chris' notion here. Anytime you have 3 or more systems, the cost to procure a spare is not much of an additional cost -- especially without fixed disk storage (which is typically the majority cost of any system).
One set of a spare, pre-assembled enclosure (including hot-swap disk), mainboard, CPU(s), memory, NIC(s) and storage controller(s) is always a good move when downtime is to be avoided at all costs (without breaking the bank on a failover implementation).
I just popped out the "hot plug" drive(s) on the stricken machine and plugged them into the spare system, flicked a power switch, and voila.
Yep, make the spare ready-to-use using existing storage. Once you have the system back up on the new "shell," you can always go back and debug the failed "shell." In the worst case, it now becomes your "spare parts" system.
Client is cash-poor, in particular, is really feeling the squeeze from the death of the dial-up industry.
Saving $3,000 while still providing reasonable options for "worst case" can provide a lot of brownie points... in any event, I've done a fairly large number of hardware swaps between P3/P4/Athlon systems, and haven't had much trouble with it. When the next Opteron server comes in, after I've set it up, I'll test it out on an Athlon/64 system I can borrow for a bit and see what issues I run into.
-Ben
On Wednesday 28 December 2005 14:44, Bryan J. Smith wrote:
Chris Mauritz chrism@imntv.com wrote:
I'm sure you already know this, but.... if it's THAT critical of a machine then it would probably be in your best interest to have a spare or two. Whenever
I
stop using one brand of server and begin using another type (at least in rackmount datacenter situations), I always
have
at least one spare installed.
I have to 2nd Chris' notion here. Anytime you have 3 or more systems, the cost to procure a spare is not much of an additional cost -- especially without fixed disk storage (which is typically the majority cost of any system).
One set of a spare, pre-assembled enclosure (including hot-swap disk), mainboard, CPU(s), memory, NIC(s) and storage controller(s) is always a good move when downtime is to be avoided at all costs (without breaking the bank on a failover implementation).
I just popped out the "hot plug" drive(s) on the stricken machine and plugged them into the spare system, flicked a power switch, and voila.
Yep, make the spare ready-to-use using existing storage. Once you have the system back up on the new "shell," you can always go back and debug the failed "shell." In the worst case, it now becomes your "spare parts" system.
-- Bryan J. Smith Professional, Technical Annoyance
b.j.smith@ieee.org http://thebs413.blogspot.com
*** Speed doesn't kill, difference in speed does *** _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Benjamin Smith lists@benjamindsmith.com wrote:
Client is cash-poor, in particular, is really feeling the squeeze from the death of the dial-up industry.
How much does it cost them per hour of downtime?
If it's enough, then spending $200 to build a backup, Socket-754 system of just a case, PS, mainboard, CPU and memory is well worth it.
Saving $3,000 while still providing reasonable options for "worst case" can provide a lot of brownie points...
Saving $2,800, $200 less, and reducing the downtime to minutes, instead of hours or even days in procuring new equipment, is a far better argument IMHO.
in any event, I've done a fairly large number of hardware swaps between P3/P4/Athlon systems, and haven't had much trouble with it.
On Linux, yes, to a point. Especially with the new 2.6 kernel, where the i686 kernel dynamically loads PPro/P2, P3, P4, Athlon/Opteron, etc... optimizations. Otherwise, I used to see people pull their hair out on 2.4, when they switched out a disk installed on an Athlon for a P4. Kernel panic (due to the Athlon kernel ;-).
When the next Opteron server comes in, after I've set it
up,
I'll test it out on an Athlon/64 system I can borrow for a bit and see what issues I run into.
Or you could just spend $200 and give your client a sub-hour recovery time, instead of hours/days.
Points taken.
Where do you recommend I go that would sell a Socket 754 system with case, PS, M/B, CPU, and RAM with SCSI support for $200? (or at least for cheap?)
-Ben
On Thursday 29 December 2005 12:38, Bryan J. Smith wrote:
Benjamin Smith lists@benjamindsmith.com wrote:
Client is cash-poor, in particular, is really feeling the squeeze from the death of the dial-up industry.
How much does it cost them per hour of downtime?
If it's enough, then spending $200 to build a backup, Socket-754 system of just a case, PS, mainboard, CPU and memory is well worth it.
Saving $3,000 while still providing reasonable options for "worst case" can provide a lot of brownie points...
Saving $2,800, $200 less, and reducing the downtime to minutes, instead of hours or even days in procuring new equipment, is a far better argument IMHO.
in any event, I've done a fairly large number of hardware swaps between P3/P4/Athlon systems, and haven't had much trouble with it.
On Linux, yes, to a point. Especially with the new 2.6 kernel, where the i686 kernel dynamically loads PPro/P2, P3, P4, Athlon/Opteron, etc... optimizations. Otherwise, I used to see people pull their hair out on 2.4, when they switched out a disk installed on an Athlon for a P4. Kernel panic (due to the Athlon kernel ;-).
When the next Opteron server comes in, after I've set it
up,
I'll test it out on an Athlon/64 system I can borrow for a bit and see what issues I run into.
Or you could just spend $200 and give your client a sub-hour recovery time, instead of hours/days.
-- Bryan J. Smith Professional, Technical Annoyance
b.j.smith@ieee.org http://thebs413.blogspot.com
*** Speed doesn't kill, difference in speed does *** _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Benjamin Smith lists@benjamindsmith.com wrote:
Points taken. Where do you recommend I go that would sell a Socket 754 system with case, PS, M/B, CPU, and RAM with SCSI support for $200? (or at least for cheap?)
For everything but SCSI, NewEgg.COM. If you want it pre-assembled, MWave.COM. Otherwise, consider a supportive whitebox reseller. If you don't have one, there are several on the Red Hat AMD64 list.
As far as your storage (SCSI), you want to get _verbatim/exactly_ the _same_ make/model controller as a backup. You should keep one of those on-hand anyway.
... and if the SCSI controller is on the Mobo?
-Ben
On Tuesday 03 January 2006 13:50, Bryan J. Smith wrote:
As far as your storage (SCSI), you want to get _verbatim/exactly_ the _same_ make/model controller as a backup.
You're playing with fire. You're not doing your client any favors by "keeping it cheap." Any money you save will be spent by someone like yourself spending hours to make sure this cobbled-together solution actually works. I don't know what you bill for your time, but for me that would easily exceed the cost of an identical server. 8-) You DO plan to actually test this at some point?
Cheers,
Benjamin Smith wrote:
... and if the SCSI controller is on the Mobo?
-Ben
On Tuesday 03 January 2006 13:50, Bryan J. Smith wrote:
As far as your storage (SCSI), you want to get _verbatim/exactly_ the _same_ make/model controller as a backup.
Chris Mauritz chrism@imntv.com wrote:
You're playing with fire. You're not doing your client any favors by "keeping it cheap."
I have to partially agree with Chris here. You're now talking about an integrated SCSI controller on the mainboard, not my idea of "ideal." So now you're putting more constraints on yourself.
Any money you save will be spent by someone like yourself spending hours to make sure this cobbled-together solution actually works.
Agreed. Until the system has an _off-mainboard_ storage controller, then you're not going to be able to replace 1:1.
At first, I thought you were just talking about having a "light mainboard/CPU/memory" as a backup. You would then have an extra storage controller on-hand as well. That would work.
But now you're got hardware dependencies in the mainboard itself, especially the storage controller. _Not_ good. That makes it difficult to bring up a replacement system by merely putting in the exact same storage controller, and moving the disks over.
You DO plan to actually test this at some point?
Agreed. That's why I suggested you have the "emergency hardware" _on-hand_ in the first place, to verify you can replace it out. Again, it's worth the $200 to do it _if_ you have an extra storage controller on-hand (which you should anyway) *AND* you can move over the storage itself.
Benjamin Smith lists@benjamindsmith.com wrote:
... and if the SCSI controller is on the Mobo?
Then get a board with the exact same host controller. Although then you're no longer saving so much on cost.
Furthermore, if you're using the on-board SCSI with an add-on, 0-channel RAID card, then that complicates matters as well. You might find the equivalent whatever-channel RAID card, but then it's costly as well.
What exact mainboard and/or SCSI controller are you using, and why?
Benjamin Smith lists@benjamindsmith.com wrote:
Now, we're moving to Opteron-based servers, and I just was wondering if it's reasonable to expect that, in a
worst-case
scenario, we could get an Athlon/64 system locally, and
have
it work, even if not optimally.
Yes, Athlon 64 and Opteron are virtually identical from that standpoint. The only issues you'd have are chipset/peripheral, just like any other change.
Just checking the accuracy of the data behind my decisions. (I've turned down Xeon servers for this reason)
Yes, there is less difference between Opteron and Athlon 64 than Xeon and Pentium 4.
-- Bryan
P.S. Any reason you don't keep around a spare Socket-754 Sempron 64 system then? I mean, they are _dirt_cheap_ (like $300 for case/MB/CPU/memory).
-
Benjamin Smith -
Bryan J. Smith -
Chris Mauritz -
Johnny Hughes