Two workarounds to "fix" (read postpone) this:
Compaq has a set of Gigabit Ethernet NICs which use the problematic driver.
The NIC models are: NC6132, NC6133, NC6134, NC6136, NC7131, NC7132
Workaround 1 - Tune the NIC Resources (This did it for me!)
Using the GUI, increase the available NIC resources by increasing the
Transmit Descriptors, Receive Descriptors and Coalesce Buffers to their
maximum allowed values. These increases will cause the driver to utilize
more memory, but the increase is usually negligible given the amount of
memory that servers in these configurations usually have. This should allow
the NIC to handle the higher volume of traffic without any problems.
Workaround 2 - Use Intel's driver
The bare bones driver comes from Intel, who has
already fixed the problem in their version of the driver (3.63.363.0). This
is available to download from Intel's support Web site:http://www.intel.com
-------------------------------------------------
> -----Original Message-----
> From: scott.kendall AT abbott DOT com [mailto:scott.kendall AT abbott DOT com]
> Sent: 18. juni 2002 00:39
> To: Anderson, David
> Cc: veritas-bu AT mailman.eng.auburn DOT edu
> Subject: Re: [Veritas-bu] Interesting Things w/ SSO on a
> Fibre Tape SAN
> (Win2k)
>
>
>
> We ran into a similar problem with Compaq's GbE card, NC6136.
> It would die
> (but still had link light) after a large number of
> connections from clients
> were established, wasn't necessarily related to throughput...
> more number os
> sessions. This would KILL SSO. Working with Compaq on it
> and am told a new
> driver is in the works. In the meantime, it works great with
> the Intel driver
> (actually an Intel card underneath all the logo'ing).
>
> As for the SCSI reserve/release... fyi, it is available with
> 3.4.1 in patch
> _3. In either one, 3.4.1_3 or 4.5, I'm being told that it is
> off by default.
>
>
> - Scott
>
>
>
>
>
> "Anderson, David"
>
> <anderson.david@scrippshealth To:
> "'veritas-bu AT mailman.eng.auburn DOT edu'"
> .org>
> <veritas-bu AT mailman.eng.auburn DOT edu>
>
> Sent by: cc:
> "Adams, Tim" <Adams.Tim AT scrippshealth DOT org>, "Morris, Brett"
> veritas-bu-admin AT mailman DOT eng.
> <Morris.Brett AT scrippshealth DOT org>
>
> auburn.edu
> Subject: [Veritas-bu] Interesting Things w/ SSO on a
> Fibre Tape SAN
>
> (Win2k)
>
>
>
> 06/17/2002 10:48 AM
>
>
>
>
>
>
>
>
>
> First off, I'll apologize for being long winded, but this is
> too good NOT to
> share.
>
> Our environment:
> All brand new hardware
> - two Compaq ProLiant ML530's, dual CPU, 1.1 G memory
> - Windows 2000
> - Master Server also acting as a Media Server
> - StorageTek L180 (4 Drives, also tried a Compaq MSL5026 (2
> drives) all SDLT
> - Compaq 8EL Fibre Switch, Compaq Modular Data Router (SCSI bridge)
> - We had started with NBU 3.4.1 and in desperation, installed
> 4.5 (the one
> with SCSI Reserve/Release).
> (by the way, if you're a Windows shop, do the 4.5, and to
> heck with Motif!)
>
> We have been having a lot of problems with drives randomly
> dropping off
> line. The libraries all saw a "SCSI Bus Rewind" request, so dropped
> whatever it was doing and waited for the actual command to
> rewind (which
> would never come). There was literally nothing that we
> didn't try. With
> marginal support from any of the vendors by the way, except
> StorageTek.
>
> There seemed to be no correlation between events in NetBackup
> or issues
> directly related to the hardware. Over the past (many)
> weeks, we did notice
> that we had occasional network problems. Specifically, the
> Master and Media
> servers deciding to not communication on the Gigabit Ethernet LAN. We
> originally had Compaq Gig-NIC's, but they seemed to be having a lot of
> packet over-run problems, among others. We have installed
> the 3Com card
> instead. This helped, but did not resolve the issues.
>
> Finally, one of our guys started thinking about SSO being able to
> communicate with itself, server to server to coordinate the SCSI
> Reserve/Release function. With this in mind, he installed a
> second NIC (yes
> the original Compaq card) and set the hosts tables so that
> the Master and
> Media servers would use only the dedicated channel to communicate.
>
> We have now run for three days, including a full weekend
> running over 1500
> separate small jobs. No a single problem. I'll admit that
> it is still too
> early to put this matter to bed just yet, but this solution looks very
> promising at this point.
>
> Has anyone else seen problems of this sort?
>
> David Anderson
> ScrippsHealth Information Services
>
> _______________________________________________
> Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>
>
>
>
> _______________________________________________
> Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>
|
