[KLUG Hardware] Re: Tech-Report's IDE RAID comparison -- good, but needs better explainations ...

[Bryan J. Smith]

Quoting David Simmons <dsimmons@powersmiths.com>:
> Now I can't believe this....an IDE RAID, very in-depth, card comparison
> and no posts, comments, "Yes, but..."
> Either way, here's the link to this great article:
> http://tech-report.com/reviews/2002q4/ideraid/index.x?pg=1
> While it's very indepth...one of the BIG points they missed and didn't
> address in their performance column - was the fact that the 3Ware card
> is 64-Bit PCI...they only ran it on 32-bit pci system...(which the other
> cards were 32-bit PCI cards)....BUT you truely miss the REAL performance of
> that card

Actually, the review was good.  The scores are what I figured they would be too.
 I'm going to "fill in" some info and analysis that explains a lot:


- PAGE 2:  COMPARING THE SPECS ...
[ http://tech-report.com/reviews/2002q4/ideraid/index.x?pg=2 ]

                     Design**   Controller    Memory
                     ---------  ------------  ----------
3Ware                ASIC +     64-bit ASIC   2MB
Escalade 7500-4      SRAM       33MHz         Static RAM

Adaptec              uC +       32-bit i960   to 128MB
2400A                SDRAM      33MHz         SDRAM
      
HighPoint            Software   none (uses    none
RocketRAID133        RAID**     host CPU)

Promise              uC +       32-bit i960   to 128MB
SuperTrak SX4000     SDRAM      66MHz??       SDRAM
^^^^^^^^^

[ **NOTE:  For "design" types, see this unpublished draft:  
http://www.smithconcepts.com/files/articles/ata_raid.pdf ]


- PAGE 38:  CONCLUSIONS
[ http://tech-report.com/reviews/2002q4/ideraid/index.x?pg=38 ]

o  3Ware Escalade 7500

PROS:  This card is most optimal when doing lots of reads, and short, seqential
write bursts.  The ASIC plus 0-latency Static RAM (SRAM -- same logic as CPU
cache) is the only non-blocking I/O controller of the three cards that have
on-board controllers, which results in the high read and high IOMeter scores. 
Although the ASIC, in essence, is little more than a 64-bit PCI bus arbitrator
for the card, it still queues up commands in its SRAM so it does off-load a lot
of load from the host CPU and further increasing IOMeter scores -- so it can be
considered an "intelligent" RAID controller.  If you run Linux, 3Ware has had a
full GPL driver in the stock Linux kernel since version 2.2.15, and they
officially support Linux.

CONS:  Unfortunately, SRAM logic is very large (not simplistic like DRAM cells),
so their is only 2MB SRAM on-board (4MB SRAM on-board the 12-channel models). 
So write performance suffers as not much can be buffered, especially in RAID-5
where XOR calculations must be made during writes.

o  Adaptec 2400A

PROS:  Originally born out of a DPT I2O microcontroller-based design, this is an
ATA adaption of a traditional, intelligent SCSI RAID controller.  That means an
on-board i960 is controlling all transfers, with upto 128MB memory, providing
excellent RAID-5 performance as it can buffer writes well while executing XOR
operations.

CONS:  Unfortunately, unlike most DPT/Adapatec SCSI RAID controllers, the i960
on the 2400A is a very old model.  It's performance becomes a bottleneck for
JBOD (just-a-bunch-of-disks) as well as in RAID-0/1 in comparison to most of
their SCSI RAID cards.  Standalone ATA itself is normally non-blocking I/O,
unlike SCSI which is command-queued by default, which hurts JBOS and RAID-0/1
performance.  And since they purchased DPT, Adaptec no longer officially
supports the DPT I2O driver on Linux (although they indirectly continue its
development).

o  HighPoint RocketRAID 133

PROS:  This is just an ATA controller with a 16-bit boot-time BIOS.  That means
the card is driven by the host CPU of your PC.  So you get non-blocking I/O as
you normally would with standard ATA channels.  You also get caching equivalent
to whatever your OS is using for file/filesystem buffering/caching.

CONS:  As with software RAID, your main CPU is now tasked with all operations,
increasing CPU utilitization.  Unlike software RAID, instead of an OS-based
organization, disk organization is tied to the vendor's implementation.  If your
OS has software RAID, I recommend it over these BIOS-only ATA RAID cards instead.

o Promise SuperTrak SX4000
[ NOTE:  I think the author meant to say "SuperTrak," not "FastTrak" ]

PROS:  Like the Adaptec 2400A, an intelligent, on-board controller and lots of
SDRAM cache.

CONS:  Unlike DPT/Adaptec, Promise does not have a history of SCSI RAID card
design and the firmware in the SuperTrak series shows.  The code is horrendously
slow at many operations, and can taxi your main CPU, defeating the purposes of a
dedicated, intelligent ATA RAID controller.  Although Promise officially
supports Linux on the SuperTrak series (unlike the "dumb" FastTrak series), its
drivers are not in the stock kernel, and I have found patching to be non-trival.


-- 
Bryan J. Smith, E.I. (BSECE)       Contact Info:  http://thebs.org
[ http://thebs.org/files/resume/BryanJonSmith_certifications.pdf ]
------------------------------------------------------------------
  The more government chooses for you, the less freedom you have.