xfs
[Top] [All Lists]

Re: Stability of the Red Hat 7.1 + XFS system

To: Juan Casero <casero@xxxxxxxxxxxxx>
Subject: Re: Stability of the Red Hat 7.1 + XFS system
From: "Bryan J. Smith" <b.j.smith@xxxxxxxx>
Date: Mon, 11 Jun 2001 23:10:43 -0400
Cc: Juha Saarinen <juha@xxxxxxxxxxxx>, "'Micah Yoder'" <yodermk@xxxxxxxx>, linux-xfs@xxxxxxxxxxx
Organization: SmithConcepts, Inc.
References: <ECEDLMDMPBOMHCMAPIHOGELDCBAA.casero@xxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
[ Okay, I'm _really_ getting off-topic.  @-P ]

Juan Casero wrote:
> Normally I don't bother with these kinds of pointless debates
> but this one touches a cord in me.  I have two deskop workstations
> here (both using MS Windows 2k cuz I need the advanced 3D graphics)
> using the 440BX chipset.  In my experience this is one of the most
> stable chipsets I have ever come across.

And, until just recently, was still being made (possibly still being
made, "unofficially" ;-)!  The i810 and 815 chipsets are limited to
2 DIMMs with only 128MB and 256MB SDRAM modules, respectively --
which canNOT be "registered" or use x4b IC sizes (nor ECC if I
remember correctly?).  Numerous mainboard manufacturer's have
released i440BX chipsets with newer FC-PGA/VRM 8.4 support, usually
with an Ultra100 ATA controller on-board.  So other than the lack of
AGPx4 support and unofficial 133MHz FSB support, they are still
quite a good solution.

E.g.:
http://www.epox.com/html/english/products/motherboard/ep-bx7+100.htm

> My primary workstation is a Tyan Tiger 100 with the i440BX
> chipset and two Pentium III 500Mhz processors and 1 Gigabyte
> of ECC PC100 SDRAM from Crucial.  What these boards lack in
> performance they make up for with rock solid stability.

Actually, it's not surprising to find the i440BX matching even i815E
chipset mainboards, even with PC133 on the latter.

> What good is the fastest computer on the market if it can't stay up
> long enough to get any useful work done?

Exactomundo!  Although if I _really_ need "stability" combined with
massive amounts of RAM (which means no RDRAM-option), I go
RCC/ServerWorks -- even Intel agrees!  Of course, ServerWorks has
AGP and ATA/IDE issues just like ViA did prior to its Intel
cross-license.  But the memory/CPU interconnect and north-to-south
interconnects are just unbelievable!  Hence why "the big boys"
(large PC OEMs) used them for years for servers.

> I also have a Tyan trinity (1590S) board with an AMD K6 that
> uses the VIA MVP3 chip set.  This board has been plagued by
> problems and instability under microsoft's OS's.

VIA had "several revisions" of the mVP3.  ;-PPP

> I think the problem is with the AGP port.

Many of the "revisions" of the mVP3 were completely for AGP fixes. 
Most of the "workarounds" are now available for older versions in
the software drivers -- albeit at a performance hit on those older
chipsets.

ViA's Socket-370 chipsets _after_ the Intel cross-license of 1998
(?) are exponentially better.  I wouldn't use the mVP3 as the
"measuring stick" for ViA.

> I was finally able to get some use out if by installing linux
> on it with no graphics-strictly a server-and using it as a
> firewall/gateway.  For this reason I flat out refuse to buy
> any system that uses a motherboard with any VIA chipset.

Because of the i810/815 stability issues, not to mention the MTH
issues of the i820 as well as various i440GX issues that did not
plague the i440BX before it, I _refuse_ to use _any_ Intel SDRAM
chipset after the i440BX.  As such, it's either a "souped up"
i440BX, select ViA chipsets or ServerWorks -- or an actual AMD
chipset if you can get one.  ;-PPP

I usually stick with ViA because of the "flexibility" of the
chipset.  Intel's chipsets are far to limited.  This is especially
true in the case of the new ViA Pro266 DDR chipsets.  Get one that
has both SDR and DDR slots -- you can use all kinds of memory types
in them -- including 512Mb SDRAM ICs for upto 1GB DIMM support, just
like ServerWorks').

[ See my post here for more detail on SDRAM IC support:
http://www.zepa.net/hypermail/elug/hardware/2001/02/0008.html ]

I've had few issues with the ViA Apollo Pro/133A (694X), although
the early Pro (693) is a different story (it pre-dates the Intel
cross-license).  The KT133 is fine except for the crappy 686A
southbridge -- but the new KT133A w/686B is fine (and select
Pro/133A and KT133s with the 686B are as well).  The Pro266 is just
as nice for SDR SDRAM (although DDR SDRAM is still "maturing" on
non-AMD760 chipsets).

> My machines have to stay up for weeks and months at a time.

My KT133 stays up as a primary workstation, including 3-D gaming,
for months at a time (Linux, of course ;-).

> Anticipating a long wait before the Itanium systems could
> become affordable I decided to deck out my current system
> and so bought the extra RAM and a nice 3D graphics card
> from 3D Labs with full OpenGL hardware acceleration.

I'm really looking for a good review of the nVidia GeForce and
Quadra series of products versus the 3DLabs/Oxygen ones at
professional GL applications.  I never was a fan of nVidia until
recently (as they released Linux drivers), and have always preferred
the latter.  But the 16-bit color 3-D performance (under both NT and
Linux) of a low-cost (~$130) nVidia GeForce2 MX dual-headed board
changed that.

> Hopefull I won't have to wait too long before I can afford
> the kind of technology presently used with Itaniuum
> workstations.

The new "bang for the buck" seems to lie in the AMD Athlon MP
(although you can use older chips in dual processor mode on the
mainboards too -- but the MP chip version is better).  The EV6 bus
combined with the full-peer cache concurrency of the Alpha/Athlon
design shreds Intel's Pentium/Xeon SMP to pieces.  But, alas, the
new Tyan AMD 760MP mainboard product is just a "preview" and an
"one-time" product.

The "production" chipset is the AMD 760MPX which will sport a
separate 64-bit x 66MHz PCI bus directly on the northbridge (as a
node on the EV6 CPU-memory-I/O crossbar), in addition to the
"legacy" southbridge for PCI/ISA.  AMD 760MPX mainboard will sport
sub-$200 prices because they still use only a single, 2.1GBps PC2100
memory channel (unlike the more expensive i840/860 chipsets with
dual 1.6GBps PC800 DRDRAM channels).

The point-to-point EV6 CPU-memory-I/O crossbar interconnect lets one
Athlon (or Athlon MP) do I/O while the other talks to memory --
which seems to best even dual 1.7GHz P4-Xeons on the, yet
unreleased, i860 chipset, in all but the most memory-intensive
tasks.

[ Enthusiast review (read the whole thing, very interesting -- and
it's not even the 760MPX yet):
http://www.anandtech.com/showdoc.html?i=1483 ]

We'll have to see if ServerWorks' DDR SDRAM P4-Xeon chipset can
"save" Intel in Q4.  Right now, AMD is really starting to pound
Intel at even the high-end for far, far less money.  Since Intel has
denied ViA any licensure of the P4 architecture, it's up to
ServerWorks to save them (not that ViA is in ServerWorks' league
anyway ;-).  Frankly, I probably won't purchase a P4 _until_ there
is a DDR SDRAM chipset anyway (and I hope the Intel-ServerWorks
cross-patent licensing agreement gives ServerWorks better AGP and
ATA/IDE interfaces).

-- TheBS

-- 
Bryan J. Smith   mailto:b.j.smith@xxxxxxxx   chat:thebs413
SmithConcepts, Inc.           http://www.SmithConcepts.com
==========================================================
Linux 'Worms' exploit known security holes that were fixed
3-12 months earlier.  NT/2000 'Worms' exploit unknown se-
curity holes that won't be fixed for another 3-12 months.

<Prev in Thread] Current Thread [Next in Thread>