On 12/12/2013 2:42 AM, Scott Ferguson wrote:
> Perhaps because 64-bit gives their use case brings disadvantage but no
> advantages? Perhaps for other reasons. To assume that you *should* use
> 64-bit in all cases is incorrect.
There are old 32 bit PAE only machines around with plenty of capability
today, albeit with a lot more space and power consumption for the
performance. Take the Unisys ES7000 Orion 230 for example. It's a 32
processor Foster Xeon 72" mainframe, 32 bit PAE only CPUs. Original
price when new in 2001 was ~$300,000 USD. Today? A few thousand, if
you could find a complete working unit at a surplus equipment dealer or
on Ebay.
CPU count: 32
CPU type: Intel Xeon MP Foster, first gen NetBurst
Specs: 1.4-1.6 GHz, 256KB L2, 1MB L3
System cache: 256MB static RAM, 32MB per 4 CPU module, 8 modules
System RAM: 64GB ECC SDRAM, 128x 512 MB DIMMs, 32-way interleaved
RAM bandwidth: 20 GB/s sustained, 25.6 GB/s peak
IO Slots: 64 PCI 2.1 66 MHz
32 PCI 2.1 33 MHz
IO Bandwidth: 5 GB/s sustained
There are a number of workloads at which this 13 year old PAE only
system would offer excellent performance today. It would make one
heckuva server for web, mail, database, etc. It would have decent SETI
or Folding throughput though individual work unit processing would be
pretty slow compared to today's CPUs. You'd be hard pressed to find a
32 bit PCI FC400/800 or SAS controller, but Intel still sells a PCI GbE
card. This allows for MPIO iSCSI over multiple HBAs and GbE links to
modern iSCSI SAN RAID arrays. Say 16 HBAs, 4 links to each of 4 arrays
with 24x 2.5" SAS drives, 96 drives total, 3.2 GB/s throughput.
This is obviously an obscure and unlikely scenario, but it is a good
example of why one would choose to run a PAE kernel.
--
Stan
--
To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/52a9a5c3.2090...@hardwarefreak.com
