Date: Fri, 02 Mar 2007 16:38:04 -0800 From: "Marc D. Brooks" <marc@paralan.com> To: Baurzhan Ismagulov <ibr@radix50.net> Cc: aic7xxx@freebsd.org Subject: Re: Adaptec 2940UW with Fujitsu MAN3184M Message-ID: <45E8C36C.1020407@paralan.com> In-Reply-To: <20070302220950.GD7331@radix50.net> References: <20070228204034.GA2379@radix50.net> <45E602B3.2020306@paralan.com> <20070228233314.GA7348@radix50.net> <45E77B30.2090505@paralan.com> <20070302220950.GD7331@radix50.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello Baurzhan,
I think the following are the two major questions you are asking:
1. Why is my setup running at Fast instead of Ultra?
From the reading you supplied, It shows the initial speed to be 40
Megabytes (20 megatransfers of a wide bus), which is Ultra speed. It
then in the next two lines shows 20 Megabytes (10 megatransfers of a
wide bus), which is Fast speed. This looks to have happened about the
time that the Adaptec controller would have completed its Domain
Validation. During Domain Validation, the Adaptec controller tries
communications at various speeds to determine if the initial negotiated
speed is stable for use, or if the bus has to go slower due to unstable
communications while testing at the negotiated speed. Domain Validation
was developed due to the problems the industry was having implementing
stable Single Ended SCSI based systems that worked at Ultra speed.
Most likely, your system is running at fast instead of Ultra because
during Domain Validation the SCSI bus communications tested out that it
would not be stable running at Ultra, so the Adaptec controller
negotiated communications down to Fast. (This is if its not due to an
OEM configuration built into that particular HBA you are using.) There
are a lot of factors that can cause this to happen: Cables, Terminators,
and Drives and/or SCA adapters that were made for LVD, which can switch
down to Single Ended, but the REQ/ACK timing may be less than optimal
when running Single Ended, etc... Basically, if you want to try
different cables, terminators, or one or more of the major components to
find the problem, this can be done. OEMs spent many hours working on
combinations of hardware configurations to standardize on that would be
consistently stable at Ultra Speed with Single Ended devices. The SCSI
Standards Committee shortened the cable lengths, Recommended using
Domain Validation (to go slower if it doesn't seem stable during startup
at Ultra) and specified that Active Single Ended Terminations should be
used at Ultra Speed in order to pass Single Ended SCSI communications at
Ultra Speed. Add in the factor that SCSI went LVD for faster speeds than
Ultra, which use different (LVD) signal transceivers than Single Ended,
with Multimode versions of those transceivers able to switch down to
Single Ended but optimized for LVD, and the possibility of having a
system switching down to Fast for stable communications increases. What
I am trying to say is that with Single Ended, you can run into exactly
the problem you have experienced going at Ultra, and you may have to try
changing out some or all of the components of your setup to get it going
at Ultra with Single Ended.
2. What can I do to make my system run faster?
I would recommend that since all of the hardware (except for the
controller) is LVD, capable of running at Ultra160 (which is stable) to
change the 2940 controller out for an Ultra160 capable LVD controller.
You will definitely get a faster speed than the SCSI Fast you are
currently experiencing, or the Ultra you ar trying to get, and not have
the struggles in trying to find what part(s) in your current
configuration are causing the system to run slower than Ultra.
I'm sorry I cannot supply a better answer which would not have the cost
of lots of time trying different components, or the cost of a different
controller, but when mixing and matching Single Ended SCSI hardware at
Ultra speeds, you can get a string of 1000 good setups, then hit on a
string of 100 problematical ones.
Best Regards,
Marc
Single configurations for systems
Baurzhan Ismagulov wrote:
>Hello Marc,
>
>On Thu, Mar 01, 2007 at 05:17:36PM -0800, Marc D. Brooks wrote:
>
>
>>>>It is possible that even though the 2940UW initially negotiates for
>>>>Ultra, the domain validation forces it to the slower rate due to
>>>>issues seen with the communications.
>>>>
>>>>
>>>Any way to see those at the Linux administrator level?
>>>
>>>
>>The type of SCSI the device and HBA supports is usually in the
>>specifications for the device and the HBA. Some BIOS implementations may
>>report the Single Ended, LVD or HVD bus type somewhere in its menus. For
>>Adaptec, in their model number they have the bus type encoded. 2940 =
>>Single Ended, 2944 = HVD, 29160 = LVD Ultra160, 29320 = LVD Ultra320.
>>For the Device you are using, I did a search for the data sheet for the
>>device and found out that it claimed to be Ultra160 device (It
>>automatically has to be LVD capable to do Ultra160).
>>
>>
>
>Thanks for the explanation! But how can I find out what factors drive
>the bus speed to 10 MHz?
>
>
>
>
>>>The card is branded Siemens-Nixdorf, I don't see the usual Ctrl-A
>>>prompt. I've seen older SNI controllers, all settings were accessible
>>>
>>>
>>>from the main BIOS. So, I guess I can't check this.
>>
>>
>...
>
>
>>The card you have may be set to max out at Fast operation in the
>>firmware for that OEM card.
>>
>>
>...
>
>
>>The OEM "Siemens" Factory Settings may be causing the negotiations
>>with the drive to not go at the Ultra speed, or in performng domain
>>validation, it had set to go slower in order to operate without
>>transmission errors.
>>
>>
>
>I see User set to 40 MB/s in /proc/scsi/aic7xxx/0, so I think BIOS
>settings are not the limiting factor.
>
>
>
>
>>>The "when" part above is strange. Now I don't have the drive connected,
>>>aic7xxx is loaded, but I see neither kernel messages in log, nor
>>>/proc/aic7xxx.
>>>
>>>
>...
>
>
>>Was the cable disconnected at the Adaptec, or at the drive package?
>>
>>
>
>Sorry, the file name is /proc/scsi/aic7xxx, I hadn't looked good enough.
>
>
>
>
>>>1.2 m, four 68-pin connectors for drives, black box at the end saying
>>>"Foxconn" and "Ultra 320" (terminator?), the drive connected to the last
>>>entry before the terminator ("far" from the controller).
>>>
>>>
>>I do not have any direct experience with "Foxconn". It sounds like you
>>have a raid package or JBOD with a backplane. The terminator is
>>installed at the end of the backplane?
>>
>>
>
>That isn't a backplane, that's a round cable with a black box at the
>end.
>
>
>
>
>>The Adaptec can have termination set or unset through the BIOS menus,
>>but I do not know what it would take to do that or if its possible with
>>the Siemens OEM version card you have.
>>
>>
>
>Ah. Again.
>
>But if the controller side is not automatically terminated, would the
>drive function reliably? I've built OpenEmbedded on this drive without
>problems (4 GB after compilation, many hours on my 1.5-GHz system). A
>similar configuration with a bad adapter didn't last 10 min. Can I
>somehow measure whether the controller has the termination enabled?
>
>
>
>
>>The similar Adapte you mention is sill a Single Ended one. The stability
>>problems are most likely associated more with Ultra Speed Single Ended
>>SCSI than with the card itself. Ultra speed communications with Single
>>Ended devices were at the limit of the physical bus capabilities and a
>>lot of things could go wrong with it.
>>
>>
>...
>
>
>>The only thing you really have to watch out for is improperly made
>>SCSI round cables. I would recommend that you swtch to an LVD version
>>(the 29160 should be fine) of a generic Adaptec controller, since the
>>SCA adapter, the Box that the adapter is in, and the Drive are made to
>>be able to support it.
>>
>>
>
>That is exactly the problem: Everything is made for at least SE Ultra,
>but the drive operates at Fast. So my concern was why it should be
>better with a 29160 if the current setup doesn't do Ultra in the first
>place. Is there a way to check the cable and the adapter? I trust the
>controller, and the drive works at Ultra in a Sun workstation.
>
>
>With kind regards,
>
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45E8C36C.1020407>