Matthew,
Oh, this does show for the 4200 the built-in NICs on a separate chipset
from the PCI-X slots (page 8). It does show the NICs share that PCI-X
chip/bus with the internal storage controller (disks). So it
looks like (page 9):
PCI slot 0 100Mhz: PCI bus shared with Gigabit and Disks (don't use)
PCI slot 1 133Mhz: A PCI bus to itself although it uses the same bridge
as slots 2,3,4.
PCI slots 2-4 66Mhz: A separate PCI bus from 0 and 1 although it shares
the same bridge as slot1.
Thanks for finding/posting this. Typical SUN that the tech papers are
not listed with the server documentation.
I suppose for the PCIe X4 slots I would imagine they would need to
replace the AMD8131 PCI-X Tunnel Chip(s) with a faster one(s)? The chip
shows
4.8 GB/s to the Opterton. Contrast this to my V880; the specs shows
4.8GB/s to the CPUS from a "workgroup data switch" which has 1.2GB/sec
going to the
each PCI bus.
Very interesting indeed.
Robert Maiello
Pioneer Data Systems
-----Original Message-----
From: Matthew Huff [mailto:mhuff AT ox DOT com]
Sent: Wednesday, February 15, 2006 4:52 PM
To: Legato NetWorker discussion; Maiello, Robert
Subject: RE: [Networker] sizing a Sun server for a 4xLTO3 tape library
Here is the tech paper:
http://www.sun.com/servers/entry/x4100/arch-wp.pdf
Five internal MD2 lowprofile 64-bit PCI-X slots (one at 100 MHz, one at
133 MHz, three at 66 MHz) which concerns me. When we last met with our
Sun rep we presented our view that the next rev of these systems should
have at least four PCIe X4 slots
----
Matthew Huff | One Manhattanville Rd
Dir of Operations | Purchase, NY 10577
OTA LLC | Phone: 914-460-4039
www.otaotr.com | Fax: 914-460-4139
> -----Original Message-----
> From: Legato NetWorker discussion
> [mailto:NETWORKER AT LISTSERV.TEMPLE DOT EDU] On Behalf Of Robert Maiello
> Sent: Wednesday, February 15, 2006 4:32 PM
> To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
> Subject: Re: [Networker] sizing a Sun server for a 4xLTO3 tape library
>
> Yes, the x86 X4200 sounds intriguing. I could not find the
> backplane speed of the server anywhere. It appears to have
> may have 2 PCI-X buses but the specs simply say expansion bus
> (singular), 5 slots. The hardware design does not seem as
> well documented as the Sparc servers.
>
> If the built-in gigabits are on a seperate bus and one could
> place the disk and tape HBAs across 2 PCI-X buses this may
> make a good server/storage
> node. If it is a single PCI-X bus, it is unclear that it
> will be able
> to handle 5 HBAs.
>
> Also one would assume Networker and the OS could make use of
> the dual cores.
>
> Very interesting.
>
> Robert Maiello
> Pioneer Data Systems
>
>
> On Wed, 15 Feb 2006 10:17:09 -0500, Matthew Huff <mhuff AT OX DOT COM> wrote:
>
> >Since Legato 7.3 is certified for Solaris X86 64-bit for
> Solaris 10, I
> >would imagine a Sun Fire X4200 with 2 x Opteron 275 chips
> would be an
> >ideal server. First, the 4 x AMD cores would be a lot faster
> than the
> >current sparc chipset. Second, it has 4 gigabit ethernet
> ports built in.
> >Third it has 5 PCI-X slots which are a considerable upgrade
> over Sun's
> >PCI slots. Finally, with 4 GB ram and 2x 73GB drives, it lists for
> >$6300...
> >
> >
> >
> >
> >----
> >Matthew Huff | One Manhattanville Rd
> >Dir of Operations | Purchase, NY 10577
> >OTA LLC | Phone: 914-460-4039
> >www.otaotr.com | Fax: 914-460-4139
> >
> >
> >
> >
> >
> >> -----Original Message-----
> >> From: Legato NetWorker discussion
> >> [mailto:NETWORKER AT LISTSERV.TEMPLE DOT EDU] On Behalf Of Jeff Mery
> >> Sent: Wednesday, February 15, 2006 10:10 AM
> >> To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
> >> Subject: Re: [Networker] sizing a Sun server for a 4xLTO3 tape
> >> library
> >>
> >> Solaris 10 has a completely new TCP stack that eliminates a
> >> single-thread for network communication in Solaris 9.
> We're going to
> >> 10 as soon as our OS admins are comfortable with it
> >> (READ: Very Soon!!).
> >>
> >> Jeff Mery - MCSE, MCP
> >> National Instruments
> >>
> >> --------------------------------------------------------------
> >> -----------
> >> "Allow me to extol the virtues of the Net Fairy, and of all the
> >> fantastic dorks that make the nice packets go from here to there.
> >> Amen."
> >> TB - Penny Arcade
> >> --------------------------------------------------------------
> >> -----------
> >>
> >>
> >>
> >> Teresa Biehler <tpbsys AT RIT DOT EDU>
> >> Sent by: Legato NetWorker discussion
> <NETWORKER AT LISTSERV.TEMPLE DOT EDU>
> >> 02/15/2006 08:54 AM
> >> Please respond to
> >> Legato NetWorker discussion
> <NETWORKER AT LISTSERV.TEMPLE DOT EDU>; Please
> >> respond to Teresa Biehler <tpbsys AT RIT DOT EDU>
> >>
> >>
> >> To
> >> NETWORKER AT LISTSERV.TEMPLE DOT EDU
> >> cc
> >>
> >> Subject
> >> Re: [Networker] sizing a Sun server for a 4xLTO3 tape library
> >>
> >>
> >>
> >>
> >>
> >>
> >> Is there any significant difference between Solaris 9 and
> 10 related
> >> to their ability to handle multiple NICs?
> >>
> >> -T
> >>
> >>
> >> -----Original Message-----
> >> From: Legato NetWorker discussion
> >> [mailto:NETWORKER AT LISTSERV.TEMPLE DOT EDU]
> >> On Behalf Of Robert Maiello
> >> Sent: Wednesday, February 15, 2006 9:43 AM
> >> To: NETWORKER AT LISTSERV.TEMPLE DOT EDU
> >> Subject: Re: [Networker] sizing a Sun server for a 4xLTO3 tape
> >> library
> >>
> >> That is well summed up Vernon, the key concept being 2 LTO3 drives
> >> (and even 2 LTO2 drives) can "eat" a gigabit NIC all on there own.
> >>
> >> That said, I'd like to add that looking at PCI buses for the HBAs
> >> and/or
> >>
> >> NICs I'm always hard pressed to pick a particular SUN server up to
> >> the task. Perhaps others can reccommend one? The ideal
> server being
> >> one where every card is connected to a seperate high speed PCI bus.
> >>
> >> Also, it has been seen that Solaris 9 or Solaris 10 is
> needed to get
> >> the throughput out of mulitple NICS.
> >>
> >> Robert Maiello
> >> Pioneer Data Systems
> >>
> >>
> >>
> >> On Tue, 14 Feb 2006 15:34:25 -0800, Vernon Harris
> >> <harriv00 AT YAHOO DOT COM>
> >> wrote:
> >>
> >> >Ty,
> >> >Rule of thumb for sizing a sun server to drive 4 x
> >> >LTO3 drives would be as follows:
> >> >
> >> > For each LTO-3 drive you would need a minimum of approximately
> >> >1.25GHZ of processing power. That would include the
> >> processing power
> >> >necessary to handle 1 gigabit ethernet nic card. But to
> adequately
> >> >drive the 4 LTO-3 drives if you backup methodology is lan based
> >> >backups, you should consider adding a second nic card and
> >> trunking the
> >> >2 nic cards together to create a fat network pipe. Otherwise max
> >> >throughput would be limited to approximately 80-90MB/sec,
> >> which is the
> >> >practical thruput limit of gigabit ethernet. If you add a
> >> second nic,
> >> >you will need 1.5GHZ of processor power per drive.
> >> >
> >> >Practically, most servers can never generate enough i/o to
> >> keep LTO-3
> >> >drives spooling without shoeshining the drives. The
> >> installations that
> >> >I've seen with
> >> >LTO-3 drives configured attached to solaris servers have not
> >> >expererienced performance issues on the servers.
> >> >
> >> >One important problem that I've seen repeatedly on Sun
> >> Servers attached
> >> >to the fabric is with Sun Branded qlogic hba's using the
> leadville
> >> >driver stack. This is manifested with link offline errors in the
> >> >/var/adm/messages file which causes the hba to go offline and the
> >> >connected drives and libraries to disappear from the
> fabric. This
> >> >condition can only be resolved by rebooting the server.
> Stick with
> >> >native emulex or qlogic cards. Otherwise you are asking
> for major
> >> >problems.
> >> >
> >> >--- Ty Young <Phillip_Young AT I2 DOT COM> wrote:
> >> >
> >> >> All,
> >> >>
> >> >> I apologize in advance if this topic has been covered.
> I looked
> >> >> through the archive using a variety of search terms without
> >> >> successful results.
> >> >>
> >> >> We have determined that a 4 x LTO3 tape library will work
> >> well in our
> >> >> environment. Our Sun SEs, however, claim that
> >> >> attempting to drive such a
> >> >> library with one host (i.e. where all four LTO3 drives are
> >> >> fiber-connected through a switch into the server) is asking for
> >> >> trouble and that we really must consider driving it
> with two, in
> >> >> order to split up the gigE network bandwidth requirements
> >> as well as
> >> >> the FC HBA bandwidth requirements.
> >> >> Their argument seems to be based on the theoretical
> >> maximum sustained
> >> >> I/O that a Sun server backplane can handle, at 1.2 GB/sec.
> >> >>
> >> >> What I'm not understanding is how one calculates I/O
> >> across a server.
> >> >> Given that a server takes network traffic (input) and
> routes it to
> >> >> the tape drives (output), is it accurate to basically
> double the
> >> >> aggregate write-rate of a bunch of tape drives (read and
> >> >> write) and then double that
> >> >> number again to factor in performance with drive compression ?
> >> >>
> >> >> My head is so full of numbers and stats at the moment that
> >> I cannot
> >> >> think
> >> >> straight and I need some help. Thanks!
> >> >>
> >> >> -ty
> >> >>
> >> >> To sign off this list, send email to
> listserv AT listserv.temple DOT edu
> >> >> and type "signoff networker"
> >> in the body
> >> >> of the email. Please write to
> >> networker-request AT listserv.temple DOT edu
> >> >> if you have any problems wit this list. You can access the
> >> archives
> >> >> at http://listserv.temple.edu/archives/networker.html
> >> >> or
> >> >> via RSS at
> >> >>
> >> >http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >> >>
> >> >
> >> >To sign off this list, send email to listserv AT listserv.temple DOT edu
> >> >and
> >> type "signoff networker" in the
> >> >body of the email. Please write to
> >> networker-request AT listserv.temple DOT edu
> >> if you have any problems
> >> >wit this list. You can access the archives at
> >> http://listserv.temple.edu/archives/networker.html or
> >> >via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >> >=============================================================
> >> ==========
> >> ==
> >>
> >> To sign off this list, send email to
> >> listserv AT listserv.temple DOT edu and type "signoff networker"
> in the body
> >> of the email. Please write to networker-request AT listserv.temple DOT edu
> >> if you have any problems
> >> wit this list. You can access the archives at
> >> http://listserv.temple.edu/archives/networker.html or via RSS at
> >> http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >>
> >> To sign off this list, send email to
> >> listserv AT listserv.temple DOT edu and type "signoff networker"
> in the body
> >> of the email. Please write to networker-request AT listserv.temple DOT edu
> >> if you have any problems
> >> wit this list. You can access the archives at
> >> http://listserv.temple.edu/archives/networker.html or via RSS at
> >> http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >>
> >>
> >> To sign off this list, send email to
> >> listserv AT listserv.temple DOT edu and type "signoff networker"
> in the body
> >> of the email. Please write to
> networker-request AT listserv.temple DOT edu
> >> if you have any problems wit this list. You can access the
> archives
> >> at http://listserv.temple.edu/archives/networker.html or
> via RSS at
> >> http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >>
> >
> >To sign off this list, send email to listserv AT listserv.temple DOT edu and
> type "signoff networker" in the
> >body of the email. Please write to
> >networker-request AT listserv.temple DOT edu
> if you have any problems
> >wit this list. You can access the archives at
> http://listserv.temple.edu/archives/networker.html or
> >via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
> >=============================================================
> ==========
> >==
>
> To sign off this list, send email to
> listserv AT listserv.temple DOT edu and type "signoff networker" in
> the body of the email. Please write to
> networker-request AT listserv.temple DOT edu if you have any
> problems wit this list. You can access the archives at
> http://listserv.temple.edu/archives/networker.html or via RSS
> at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
>
To sign off this list, send email to listserv AT listserv.temple DOT edu and
type "signoff networker" in the
body of the email. Please write to networker-request AT listserv.temple DOT edu
if you have any problems
wit this list. You can access the archives at
http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|
