Re: Interesting SunFire V65x configuration issue

From: UNIX admin (tripivceta_at_hotmail.com)
Date: 10/07/03 

Date: Tue, 7 Oct 2003 19:39:01 +0200

"Dennis Clarke" <dclarke@blastwave.org> schrieb im Newsbeitrag
news:Pine.GSO.4.58.0310011948460.2059@blastwave...
>
> Hello fellow Solaris for Intel folks, I have a good one here for you to
think
> about. I have to come up with an interesting solution to what seems to be
a
> simple problem. Seems to be. I have an E450 running Solaris 7 and it has
two
> external disk arrays attached. One of them is an A1000 so therefore I
need to
> run Raid Manager. The other array is an Integrix unit and it just works
as an
> normal external differential SCSI device should. The performance of the
A1000
> is really bad. I don't think that I ever get more than 12Mb/sec to it.
The
> unit reports a clean healthcheck but the I/O is pathetic. The Integrix
unit
> is only a little better. This server holds the email for about 12,000
users.
> It is time for a better solution. The V65x looks like a real winner as it
is
> three or four times faster than the E450 ( 3 x 300MHz + 2.5Gb RAM ) and it
> will use Ultra320 LVD SCSI disks internally. The idea is to setup a V65x
and
> attach some sort of external storage that will work with either the E450
or on
> the V65x. You may think that is strange but read on for further info.
>
> The current E450 config has the standard internal SCSI as well as two
PCI
> adapters X6541A differential SCSI with the high density SCSI ports on
them.
> That allows for the two arrays to be on separate controllers. A good
idea.
> They are currently plugged into old slow array technology. A bad idea.
Any
> decent SCSI disk will move along at 25Mb/sec without question. We should
get
> more with more spindles in parallel. What I would like to do is upgrade
the
> E450 to Solaris 9. I will install another CPU and beef the RAM up to 4Gb.
> That will make for a fairly maxed out E450. There are faster CPU options
but
> that is a waste of money on an E450 when IO is the issue, not CPU. I can
then
> take a pile of fast SCSI disks and put them into a 6-bay UltraSCSI
multipack
> and use SVM to stripe them. If I use differential disks then I can plug
the
> entire multipack into the X6541A controller. A striped set of SCSI disks
> should allow for about 50Mb/sec of IO on a differential controller. If I
> create another identical multipack and plug it onto the other X6541A
> controller then I can mirror the stripes and I have good redundency
without
> any controller contention. Sounds easy for the E450 and it is simple to
do.
> I may get better performance if I use Ultra320 disks in the multipacks
with
> the LVD X6758 PCI controller.
>
> Now here comes the tough part.
>
> The V65x is new technology and while it has the Sun logo on it there is
still
> the issue of trust. The E450 can be trusted to run and run and run. The
V65x
> needs to be tested for a while. So here is the idea. I don't want the
V65x
> to become a replacement for the E450 but I do want to see if it can do the
job
> better. I want to be able to pull the SCSI cables from off the back of
the
> E450 and plug them into the V65x. If the V65x is running Solaris 9 with
SVM
> setup for striped mirrors in the precisely same config as the E450 then it
> would be an issue of metainit and run with the data intact. I have tested
> this with the Sparc platform and moved entire stripe sets from one machine
to
> another without dropping a bit. Does that still apply for a move from
Sparc
> to Intel architecture? I don't know.
>
> There is a bigger problem still.
>
> What are the PCI adapters going to be on the V65x? Let's take a look at
the
> Sun System Handbook for PCI SCSI Host Adapters :
>
> http://sunsolve.sun.com/handbook_pub/Devices/SCSI/SCSI_TOC.html
>
> The only two PCI cards that I am interested in is the X6758 which is a LOW
> Voltage Differential and the X6541 which is just differential. But look
close
> and you will see that neither are for the V65x. In fact, have a look at
the
> options for the V65x and there are none. Not one PCI adapter for
differential
> SCSI. Zero. Nada. Zilch.
>
> Do I believe that? Do I simply plug the X6541A into the V65x and see if
the
> 53C876 chip is well supported? Do I take the same risk with the X6758 and
its
> QLogic chips? That is too much of a risk with a new V65x.
>
> I want a pure Sun solution. Is that too much to ask or should I simply
> install an adaptec HBA? If I do, will the data that is laid down by the
> X6541A within the E450 be readable by the adaptec? I doubt it.
>
> The Solaris x86 HCL has 12 disk controllers that are "reported to work".
One
> of them is a PCI-X type. The term "reported to work" is not good enough
to
> house the email for 12,000 people. Not good enough at all.
>
> Is the V65x even worth looking at or should I stick with UltraSparc for
triple
> the price and half the performance? I can't.
>
> Maybe I should scrap the idea of trying to make the external disk packs
work
> with the E450 and the V65x at the same time. Maybe I should just work
with
> the V65x and use an external array like the StorEdge S1 which seems to
work
> well with LVD controllers. But wait! What is that SCSI port on the back
of
> the V65x? Is that a LVD SCSI port? Who knows as its not documented much
of
> anywhere.
>
> sigh ....

You definitely bring out some interesting points. I do warmly recommend
talking to a Sun engineer about this, and what your options are, if you have
been assigned one.
If not, it's time to look into that.

One thing baffles me though. The E450 should be a cold blooded killer when
it comes to I/O and performance in general. I don't understand why you're
having such poor performance. Is it possible, that one of the SCSI devices
is slowing the other one down?

Relevant Pages

  • Re: Adaptec Storage Manager reports "No controllers were found in this system"
    ... I'm just a SCSI fan. ... I thought I had done enough research when I purchaed my Dell Precision ... a PCI-X SATA controller I intended to buy, and which would only work at 133 ... performance depends a lot on the number of disks involved, the RAID type, ...
    (comp.os.linux.hardware)
  • Interesting SunFire V65x configuration issue
    ... normal external differential SCSI device should. ... The V65x looks like a real winner as it is ... will use Ultra320 LVD SCSI disks internally. ... should allow for about 50Mb/sec of IO on a differential controller. ...
    (comp.unix.solaris)
  • Re: mainframe performance, was Is a RISC chip more expensive?
    ... all IBM systems have used SCSI disks. ... Yes, there is a controller that translates ECKD into SCSI commands, but that adds latency to the I/O. ... Yes, desktop PCs typically can do only one I/O per disk controller at a time, but PC servers can easily use the same SCSI disks as an IBM mainframe does and even do it without the additional latency. ...
    (comp.arch)
  • Problem with Maxtor Atlas IV drives
    ... It is time for a M/b and CPU upgrade anyway, and as onboard SCSI ... controller I might as well use it. ... jumpers on the disks - I wanted to see if the controller could recognise the ... can't do if no controller can recognise them. ...
    (comp.periphs.scsi)
  • Re: UC07 with BA350/RZ28 disk spinup problem
    ... VMS did not have SCSI drivers (or its SCSI ... command to a SCSI "spin up" command. ... to the controller configuration, at which point you can issue SCSI ... as far as the DSA/RA series disks, ...
    (comp.os.vms)