>>>>> "hp" == Harry Putnam <rea...@newsguy.com> writes:
hp> I'm thinking of turning to Asus again and making sure there is
hp> onboard SATA with at least 4 prts and preferebly 6.
I would like 64-bit hardware with ECC, 8GB RAM, and a good Ethernet
chip, that can run both Linux and Solaris. I do not plan to use the
onboard SATA. So far I'm having nasty problems with an nForce 750a
board from asus (M3N72-D) under Linux.
* random powerdowns. about once per week.
* freezes setting the rtc clock. as in, locks the whole board hard.
and confuses something deep inside the board, not just Linux. Why
do I think this? I've set it to ``power on after power failure''
which it normally obeys, and I know it normally obeys because I
always do cold-boots of this board because of other strange
intermittent problems below. But after such a clock-set freeze, i
try to cold-boot, and it stays off until I press the front panel
button disrespecting the BIOS setting (and making a Baytech strip
useless. piece of shit!)
* sound card attaches sometimes, not other times. In 'cat
/proc/interrupts' I see the sound card shares an interrupt with USB
(which AIUI it should not need to do with ACPI and ioapic, aren't
there plenty of interrupts now?), and it seems to march through
interrupt wirings after each warm boot, always picking a different
one, so I blame the Asus BIOS. I think it attaches about 1/3 the
time. It does play sound when it attaches.
* a USB key that works with other motherboards does not work with
this one. I am not trying to boot off the key. I boot with
Etherboot PXE. I'm only trying to use the key under the booted OS,
and it does not work, while the same OS on a different motherboard
is able to use the key for months without problem.
* with the Supermicro/Marvell 8-SATA-port PCI card installed, the
Asus board will not enter the Blue Screen of Setup. Sometimes it
does, though, maybe 1 in 10 times. It runs fine with this card
installed, just won't enter setup unless I remove it.
* with ``fill memory holes'' enabled in BSoS, as it is by default,
memtest86+ crashes and reboots the machine. When I turn this off,
memtest86+ runs fine. I'm not sure how I even found this
workaround, but seriously, I would rather waste time on mailing
lists than mess aroudn with such junk.
maybe other things that I'm forgetting.
Under Solaris, and I think also Linux, the nForce 750a's nvidia
ethernet chip is pretty performant. (newegg says realtek ethernet,
but this is wrong, it's an nvidia MAC). And also nvidia AHCI SATA
which is supposedly better than AMD AHCI SATA. I can verify at least
four SATA ports work well, on Linux, but I haven't tried the other two
``RAID'' SATA ports which are rumored to be on a crappy JMicron
controller with weird mandatory fakeRAID or something. Maybe they
work fine, I don't know.
While the ATI/AMD chipset boards actually do have crappy realtek
ethernet (crappy performance on every OS), and there is supposedly a
bug in their AMD AHCI that has persisted through several chip
steppings that makes SATA slow under Linux and buggy under Solaris, and
I've heard no resolution so I'm operating under the assumption the
most recent 790{X,GX,FX}/SB750 still have the bug.
There's dispute about updating the BIOS. in general you have to
update it to work with the latest CPU's. Sometimes you need an old
CPU to run the BIOS updater before you can run the new CPU, even if
you ordered the CPU and the motherboard on the same invoice. That
seems highly bogus to me. I did not update mine at first, then
updated it to try to solve the random powerdowns, which it did not.
Other forum posters say the first BIOS revisions are written by the
BIOS programmers who did well in college, while the later updates for
newer CPUs are written by the ones who barely scraped by and are full
of regressions, and since they're not selling a new motherboard with
the update they don't give a shit and would almost rather brick the
board and force you to upgrade, so these posters say the newest BIOS
builds should be *AVOIDED*. I've no idea!
Maybe you should buy two or three cheap boards and try them all, since
the CPU and RAM make up a bigger fraction of the cost. I've just
bought this one:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813186150
because it has Broadcom ethernet, which is actually a good-performing
chip with a decent driver, but this chip has a lot of errata among
revisions so your Ethernet driver has to be updated more aggressively
than it does with the Intel gigabit chips. I'm hoping it is the older
non-RNIC version, because the newest Broadcom chips have a
(newly-added, <cough> <COUGH>) Proprietary driver in Solaris, while I
think the older chips have a free driver. Broadcom, like AHCI SATA,
is decent and much easier to obtain onboard than on a PCI card, so
it's nice to find a board which has it.
This one is also interesting:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813136058
because it's also an AMD chipset also without a Realtek Ethernet. It
has Marvell Ethernet, which is an uncooperative company so the Linux
driver is probably poor but the Solaris driver may be better or at
least worth a go, since I know the Realtek chip is always bad no
matter what Theo de Raadt says. The newegg comments say it does have
onboard video even though you can't see the jack (DVI only). Also it
has 3 PCI slots while almost every other desktop board has only two,
so you could put cheap 4-port Sil3124 SATA cards in those for 12 SATA
ports that do not have the AMD AHCI bug. It would probably be a
better plan to use the PCIe LSI SATA board with the backwards-bracket
that everyone likes, though, in which case you don't need old PCI
slots:
http://www.thenerds.net/SUPERMICRO.Supermicro_AOC_USAS_L8i_8_Port_SAS_RAID_Controller_1_x_SAS_x4_SAS_300_Serial_Attached_SC.AOCUSASL8I.html?affid=8&ci_src=14110944&ci_sku=AOCUSASL8I%5E~%5ESUPERMICRO
http://www.thenerds.net/3WARE.AMCC_Serial_Attached_SCSI_SAS_Internal_Cable.CBLSFF8087OCF10M.html
x2
I've also bought one of these:
http://www.newegg.com/Product/Product.aspx?Item=N82E16813131363 $200
http://www.shopblt.com/cgi-bin/shop/shop.cgi?action=enter&thispage=011003000502_BT25946P.shtml&order_id=!ORDERID!
$230
http://www.memorydepot.com/details.asp?id=KVR1066D3E7K2%2F4G x2 $180
http://www.newegg.com/Product/Product.aspx?Item=N82E16814141083 $36 <--
nvidia instead?
http://www.newegg.com/Product/Product.aspx?Item=N82E16835185125 $20
http://www.servercase.com/miva/miva?/Merchant2/merchant.mv+Screen=PROD&Store_Code=SC&Product_Code=CK4020&Category_code=MS
which is not assembled yet---I will let you know if it works at all.
There seem to be few choices for AM3 boards (DDR3 RAM), or else I
wouldn't have gotten that one since it has no onboard video, and Asus
seems to have more dissatisfied commenters than Gigabyte, DFI, MSI,
.... However the AMD 790FX northbridge may be a good choice for ZFS
because of SATA port density: It has four 8-lane PCIe slots.
Hopefully you can fill three of them with the LSI Logic card at full
8-lane width, for 24 non-oversubscribed SATA ports (non-oversubscribed
meaning no ``port multipliers'', SAS or SATA). The fastest AM3 CPU
seems to be 2.6GHz while the fastest AM2+ is 3.0GHz, and the overall
system cost is lower with the AM2+ CPU, so this path is probably a
rip-off, but I wanted to try out the latest AMD generation just to see
if it actually works or not.
A big issue for all AMD desktop boards is whether they support ECC or
not. The slots need to be physically wired, and the BIOS needs to
write to a few very simple registers in the CPU at boot time to
request the ECC correction, scrubbing, and chip-kill features. I've
been using memtest86+ from memtest.org (_not_ memtest86) to query the
ECC on/off status, and it DOES seem to be responding sanely on the
Asus M3N72-D board with the random-powerdown problem. For example,
memtest shows that ``chip-kill'' is on when the DIMMs are in ganged
mode, and off when they are unganged, which supposedly corresponds to
the controller's limitation. But the ultimate ECC test is to short
one of the pins with a 10-ohm resistor and verify the machine (a)
keeps running, and (b) increments an ECC error counter. That, I have
not done. Also I have not tested with memtest the Foxconn board that
I haven't received yet.
In the early part of the Newegg comment section for the AM3 DDR3
board, Asus claims it supports ECC, but to me this does not mean it
actually corrects errors. It may mean ``it boots up with ECC memory,
but in non-ECC mode,'' or ``it enables ECC, and then randomly powers
down as soon as you hit the first ECC error because of a bug in some
machine-check exception handler.'' who knows. I'm highly suspicious.
But if you look in that comment section you will see Asus mention some
DDR3 unbuffered ECC sticks compatible with their AM3 board, which they
quietly added to the QVL after I complained. I think Newegg does not
sell any of the sticks they mention---you have to resort to
froogle---but you shuld probably get ECC RAM off Asus's QVL as posted
among the oldest Newegg comments and not the Kingston sticks in my URL
which are not on the QVL because I ordered them before Asus responded
to my comment. Foxconn does not even have a QVL that I can find. :)
I'm still in the process of building so I'll let you know when I have
some more results, but those are the directions I'm looking in---I
hope they work out.
pgpJvgAhYLrz0.pgp
Description: PGP signature
_______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
