SUMMARY: V890 PCI SCSI card maximum data rates

Sorry for the long delay for the summary.

The reason I was asking the question was to help evaluate a V890 against a
Fujitsu PP650.  The question of which would be most suitable for a high
activity backup server boiled down to which had the best PCI bus data rates.

I was unable to find any definitive Sun document anywhere, but a discussion
with Sun(!) confirmed that the PP650 has a much higher bus data capacity
(one controller per pair of PCI slots.)

Personally, I doubt that the data rate to the tape drives will be the
limiting factor, but I'm not the one who has to make the system work :-)

I do agree with Alex Madden below, there is an obvious and fatal gap in Suns
products starting with the V880.  The V880 is a jumped up V440 and is fine
for computation but lacks the raw data rate through the PCI buses for
applications that need to shift lots of data.  The trouble is, you have to
skip a number of Sun server models before the IO power starts to pick up
again and by that time you are paying a 30% premium for hardware domains
that you may not have any use for... The Fujitsu PP650 fills the gap and
that's what we will probably buy.

The answers I received follow...

From: Tim Chipman 

by "backup server" - you mean something like netbackup, dumping data to 
tape drives (?)

I ask b/c ... typical data throughput to a tape drive is 20megs/second, 
if you are lucky (depends heavily on what kind of data you are backing 
up in my experience, ie, large chunks of data facilitate "streaming" 
backups better than "many small files" backup jobs).

That being said,

we had a temporary netbackup / file server running on a blade100 (32-bit 
PCI of lowest calibre) and it had no trouble feeding (albeit rather 
dated) DLT 40/80gig tape drives in an L1000 robot.

so, it may be the case .. that the actual demands you are coping with 
are not quite as .. demanding .. as suspected (?)


of course, trying to feed .. 4 LTO drives .. simultaneously .. 
20-30megs/second of data per drive .. will be quite a pile of data .. 
but then, to get this kind of inbound data stream suggests multiple 
gig-ether nics (or multiple local storage arrays/SAN paths) also ...


From: Alex Madden 

>From the look of this (from 
http://sunsolve6.sun.com/search/document.do?assetkey=1-9-21216-1) each 
890 has 2 PCI controllers in - each one has a 66Mhz bus (,600000) + a 
33Mhz bus (,700000) - from what I can remember the maximum throughput of 
a 33Mhz PCI bus is 264 Mb/s - so thats shared between all 4 cards on the 
bus.

pci@8

Onboard FCAL-loopA              /pci@8,600000/SUNW,qlc@2/fp@0,0/ssd@<WWN>,0
Network 1 (ge)                  /pci@8,600000/network@1

PCI Slot 0                      /pci@8,700000/<device>@5 
33mhz-64bit-5V
PCI Slot 1                      /pci@8,700000/<device>@4 
33mhz-64bit-5V
PCI Slot 2                      /pci@8,700000/<device>@3 
33mhz-64bit-5V
PCI Slot 3                      /pci@8,700000/<device>@2 
33mhz-64bit-5V
DVD-ROM Drive (IDE)             /pci@8,700000/ide@1

pci@9
 

PCI Slot 7                      /pci@9,600000/<device>@2 
33/66mhz-64bit-3.3V
PCI Slot 8                      /pci@9,600000/<device>@1 
33/66mhz-64bit-3.3V

EBUS                            /pci@9,700000/ebus@1<device>
Network 2 (eri)                 /pci@9,700000/network@1,1
USB                             /pci@9,700000/usb@1,3
EBUS                            /pci@9,700000/ebus@1<device>
PCI Slot 4                      /pci@9,700000/<device>@4 
33mhz-64bit-5V
PCI Slot 5                      /pci@9,700000/<device>@3 
33mhz-64bit-5V
PCI Slot 6                      /pci@9,700000/<device>@2 
33mhz-64bit-5V

Couldn't you use a few smaller boxes instead ? Alternatively the v40z 
has a bit more PCI bandwidth or if you want to stick with SPARC the low 
end Fujitsu PrimePowers seem to have more bandwidth than the Sun's (and 
you can buy PP's via Sun's channel partners now).

Personally I'm of the opinion that the lowend Sun gear (440 -> 2900) is 
crippled by insufficient PCI bandwidth. For instance the 2900 is a joke 
- with one 66Mhz PCI slot.

On Wed, Mar 02, 2005 at 06:24:55PM +0000, Geoff Lane wrote:
> I've just spent a couple of hours searching through Sun web pages for an
> answer to the following...
> 
> We are attempting to evaluate a V890 as a backup server and need to estimate
> the maximum data throughput rates on the PCI slots.  From a block diagram of
> the V890 we can see that four of the 33MHz PCI slots share the same bus.
> What does this imply about the data rates we could see from SCSI cards on
> that bus?  Could we drive each of the cards to the limit of it's ability or
> would each card run at a reduced rate because the bus is shared?
> 
> Ideally I need a reference to a Sun document, but any opinion would be
> useful.

-- 
Geoff Lane
I'm not a Windows user, consequently I'm not 
afraid of receiving email from total strangers.
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers