[nylug-talk] Slim home server for samba and subversion ( and possibly IMAP)

[Bryan J. Smith]

Chris Knadle wrote:
> I had half written a post back concerning the poor GigE NAS device
> performance.  [In practice I see limits of 14 MB/sec on an embedded
> GigE NAS vs 40+MB/sec GigE transfer between a laptop + desktop.]

The Intel X-Scale ARM-based IXP embedded microcontrollers are often used
without their proprietary, Network Processing Engine (NPE) firmware in
many devices.  If they were just doing network, with the NPE, then
they'd be okay.  But they are also doing storage and application-level
serving.  Not good.  ;)

Chris Knadle wrote:
> I hadn't even been aware about the USB + FireWire timeout issues
> until you mentioned those here.  Ugh.

I solved major server downtime issues in various, separate labs (i.e.,
couldn't be connected to IT's networks for various reasons you can
assume ;) at a huge defense contractor by doing the simple favor of
removing all of their external USB and FireWire devices.  ;)

I was also involved with early Apple Server testing of their external
FireWire arrays.  Although Apple wouldn't admit it at the time, they
would later finally do what I told them to do -- admit that FireWire has
disconnect issues.  These have yet to be completely solved.

As far as eSATA, Seagate _refuses_ to certify (let alone include) any
eSATA cables for their own Free Agent Pro line.  I have hit their
support up multiple times, and each time, they fully admit they have "no
certified cables" for them.

So, until then, only SAS continues to be a viable, external storage
technology for 24x7 connection for myself in my professional endeavors.
For home usage, I don't attempt 24x7 with FireWire, USB and eSATA.

Although I hear newer TiVO Series 3 Firmwares are more tolerant with
eSATA disconnections.  Most OSes are not.


Chris Knadle wrote:
> Yeah the Yonah Celeron core seems to have 512k L2 cache (or so says
> Wikipedia) whereas the AMD Geode NX doesn't have any L2 cache at all,
> and L2 cache makes a huge performance improvement.
> The Geode NX runs at lower power at 6W compared to about 19W, but I'm 
> guessing that the power savings likely isn't worth the performance hit.

You're confusing the Geode LX with the NX, the NX _is_ an "Athlon-M".  ;)

  http://en.wikipedia.org/wiki/Geode_%28processor%29#Geode_NX  

All Geode products from AMD have L2 cache -- 256KiB typical for NX.
That's in addition to the split 4x L1 cache AMD has over Intel.

People forget the major differences between AMD and Intel.  Yes, Intel
uses big L2 caches, but AMD uses 4x the L1 cache, plus AMD can load
directly into L1 and L2 separately (although the new L3 is different).

Chris Knadle wrote:
> Seems like you've done plenty of research into this already, certainly
> more so than I have on this particular niche.

I've done embedded hardware (both x86 and non-x86) and OSes throughout
various points in my career.  The more recent commodization of Mini-ITX
(6.7" x 6.7") really challenged traditional 3.5" SBC (4" x 6") and 5.25"
SBC (6" x 8") prototype boards.

Chris Knadle wrote:
> I've focused more on Geode LX based products, i.e. more embedded
> x86-compatible hardware, for use as low-powered Linux boxes as firewalls
> or other uses, and that kind of hardware doesn't cover 500 GB storage
> that performs well.

The LX is really an old core.  Nothing like the NX.

Chris Knadle wrote:
> Is the Atom licensing restrictive or something?  WTH?  I wonder why it's
> not getting more widespread adoption.

First off, Atom was just _announced_, 0 products yet.  ;)

Secondly, Atom is a 2-prong approach ...

1.  A "cookie cutter" core, so for custom designs with Intel as a
Foundry (they sold their X-Scale to Marvell)

2.  A "Centrino" product -- which isn't out yet

Third, Atom is a power-performance pig versus non-x86, like IBM Power
and licensees (including the Power5-based PowerPC fabless design firm
that Apple just snatched up)

Fourth, Atom is a 2-issue, in-order design, so it sucks compared to
other x86 (sans maybe ViA)

Intel has long contended that people wanted an embedded x86 rather than
non-x86.  This may be true for server hardware, like HBAs for network
and storage.  But for portables, it's more debatable for things not
running Windows.

Chris Knadle wrote:
> It also seems to me I've experimentally gotten more reliability out of 3.5" 
> hard disks than 2.5" hard disks -- however this statement is completely 
> subjective, only has a tiny sample set (less than 5  2.5" drives) and I 
> haven't (yet) read any hard data comparing 3.5" to 2.5" disk reliability.

Let me guess ... you're comparing 2.5" disks in notebooks?

I'm talking about 2.5" v. 3.5" in the _same_ environment.  ;)

A lot of people abuse 2.5" discs unlike 3.5" discs.  Yes, 2.5" discs can
take 10x the vibration and abuse as 3.5", but they are still more
reliable under the same environment than 3.5".

I had a colleague complain about his 2.5" disc reliability versus 3.5",
and I fully witnessed him drop his 2.5" external drive on the floor.
After he did so, and I challenged him, he responded, "oh, it's small so
it can take it."  I then took his 3.5" external disk and motioned that I
was going to do the same and he said, "no, you'll break it!"

I rest my case.  ;)

Chris Knadle wrote:
> I don't even know how to compare these, beacause practially speaking
> it's just not possible to use a 3.5" disk exactly like a 2.5" disk.

Ah ha!  So you are _not_ comparing apples and apples, eh?
Both 2.5" and 3.5" have the _exact_same_ connectors for SATA.  ;)

Chris Knadle wrote:
> Yes I've moved around 3.5" disks in external drive bays, but IMHO
> that's just not the same thing.

Huh?  Why not compare 2.5" and 3.5" in external enclosures?
That was my _exact_ context under discussion in my original response.  ;)

Chris Knadle wrote:
> That's interesting and I hadn't heard about it.

Yes, most people not deploying new server designs in the last year have
never heard of them.  Right now, 25 of 28 of my new servers are 2.5"
15Krpm.  They've been around almost 2 years now.  ;)

You can get a crapload of performance out of 3-4 times the number 2.5"
drives for a single 3.5" disc when striped.  They run cooler.  They take
far more vibration, especially at 10-15Krpm.

In fact, it's never been recommended that a 15Krpm drives be
manufactured with larger than 3" (not 3.5") platters, hence why the
first 15Krpm invented last decade had only 3" platters (not 3.5"), even
though they were in a 3.5" HD enclosure.

Chris Knadle wrote:  
> I'll put that on the list of things to look into.

The key thing is that with 7200, 10,000 and 15,000rpm 2.5" drive
SATA/SAS drives becoming popular in the data center, the fabrication of
2.5" drives -- even at 5400rpm -- have become more commodity priced.

That's why 250GB 5400rpm 2.5" drives are now only a little over
$0.30/GB, basically less than double the most cost-effective, 7200rpm
3.5" design (640GB at 2x320GB platters).  That gap will likely close to
less than 50% soon.

So within the next few years, it's extremely likely that 2.5" will
_overtake_ 3.5" drive shipments.  There's going to be a shift from 3.5"
to 2.5" as the primary drive, and 1.8" will likely be the next 2.5" --
especially since EEPROM drives are often sold in 1.8".

The 3.5" drive is going the way of the 5.25" disc.  In all honesty, from
the standpoint of reliability, it's about time -- especially its utter
lack of ability to survive G forces and vibration like 2.5".


-- 
Bryan J  Smith              Professional, Technical Annoyance
mailto:b.j.smith at ieee.org  http://www.linkedin.com/in/bjsmith
-------------------------------------------------------------
           Fission Power:  An Inconvenient Solution