toasters March 1999

toasters@lists.teaparty.net

138 participants
192 discussions

RE: Future Newbie Needs Advice
by Clarke, Bruce 05 Mar '99

05 Mar '99

One clarification: NFS and CIFS behave quite differently. We commonly demo Oracle/Unix (NFS) running against a cluster and routinely do "unexpected" things to the target filer making it fail (pulling both plugs is my favorite). It seamlessly fails over to the second node and Oracle, after a hesitation for a minute or so, takes off where it left off. The giveback is equally transparent. Thank Sun (circa 1983) for the statelessness of NFS. You are right; NT connections are not quite so robust. Bruce Clarke - Database Marketing Director Network Appliance, Inc. 2770 San Tomas Expressway Santa Clara, CA 95051 Phone: 1-408-367-3156 Fax: 1-408-367-3670 Email: bruce.clarke(a)netapp.com

1 0

RE: Future Newbie Needs Advice
by Sam Schorr 04 Mar '99

04 Mar '99

Russ, We have just installed clustering using 2 F740s. We had one F740, bought a second with clustering, and it was installed Monday night. A couple of things have happened: 1. We had to replace the disk array shelves onthe existing filer last night because the cluster partner (the new F740) could not see all of the disks on the "older" F740. Also, we had to upgrade the disk firmware on the "older" F740. 2. The existing filer has now started to crash. We had one crash Tuesday before correcting the disk problem above, so the cluster partner did not take over. We had a second one today on the "older" filer, and this time clustering worked fine. 3. However, it is not seemless. The failover breaks file shares (we are a CIFS-only environment, but I understand it would do the same to NFS mounts). So, users/application may get file errors if they are in the middle of performing a file operation. When you do a "giveback", you either have to manually break file shares again, or do a "cf giveback -f" which forces the share break. Only then, after the shares have been broken, does the filer you are do the "giveback" to reboot. This means, once again, file errors to clients. 4. Netapp believes the crashes may be the result of a bad ServerNet card (the card that connects the two filers together for clustering). If true, that means that we will have suffered 3 bad/faulty parts in one install/purchase. If you want, I'll let you know the end of this soap opera. Sam Schorr -----Original Message----- From: Russ Amidon [mailto:amidon@pico.sps.mot.com] Sent: Thursday, March 04, 1999 1:56 PM To: toasters(a)mathworks.com Subject: Future Newbie Needs Advice Hello Everyone, I am about to purchase two F760's w/ Cluster Failover Software and Hardware because we need a system that is 100% available and rock solid. I was convinced after talking to other Net App users that this was the way to go. After a week of reading this mailing list, I am no longer convinced. Does anyone out there have the system I described? Are you happy with it? I am particularly concerned about the fiber channel problems and all of the OS upgrades and patches. We are also considering the new Auspex NS2000. Does anyone know of an Auspex mailing list like this one? Thank you in advance for any advise you can offer to the "future newbie". Russ Amidon UNIX Sys Admin

1 0

Re: Future Newbie Needs Advice
by sirbruce＠ix.netcom.com 04 Mar '99

04 Mar '99

On 03/04/99 13:56:29 you wrote: > >Hello Everyone, > >I am about to purchase two F760's w/ Cluster Failover Software and Hardware >because we need a system that is 100% available and rock solid. I was convinced >after talking to other Net App users that this was the way to go. After a week >of reading this mailing list, I am no longer convinced. Does anyone out there >have the system I described? Are you happy with it? I am particularly concerned >about the fiber channel problems and all of the OS upgrades and patches. > >We are also considering the new Auspex NS2000. Does anyone know of an Auspex >mailing list like this one? Data seems to suggest that despite the problems described, your Netapp is going to be far more reliable than an Auspex or virtually any other system. *Nothing* is going to be 100% available. Clustered filers are close; for more reliability you're going to have to pay a LOT more for really big iron which may not have the software capabilities you want. The other important thing is that the filer, when it does crash, generally results in a minimal outage. Other serves may take a half-hour or more to recover. Bruce

1 0

Re: Gb-ether and Extreme: closure! :-)
by Devi Nagaraj 04 Mar '99

04 Mar '99

At 10:38 AM 3/4/99 -0800, Dane Jasper wrote: >On Thu, Mar 04, 1999 at 09:19:36AM -0000, Garrett Burke wrote: >> As an aside, have you run a filer into a 100Mb/s port on a S48? I'm seeing >> CRC errors on the port that my F230 is on. Netapp are investigating, just >> wondering if anyone else is seeing/has seen this. > >We've got a F230 running 100-FD with a Summit 48. We found that we had >better results if we hard set both devices to 100-FD and turned off >negotiation. The F200/F300 series on-board ethernet doesnot autonegotiate. So if you had the filer's ethernet interface set to operate in 100BaseTx-FD, the corresponding switch port should also be set to operate in 100BaseTX-FD. When speed/duplex mismatches occur, depending on the switch to which the filer's interface is connected, you will see CRC errors and collisions. You will definitely see poor network throughput. Here is a table that gives info on the board types and the supported speed and duplex (mediatype) Interface Sysconfig name Mkt P/N Mediatype --------- -------------- ------- --------- Single 10/100 - ZX342 - X1001 - 10BaseT(half/full), 100BaseTx(half/full) Quad 10 BaseT - ZX314 - X1002 - 10BaseT(half/full) Quad 10 BaseT - ZX314 - X1003 - 10BaseT(half/full) Quad 10/100 - ZX346 - X1012 - 10BaseT(half/full), 100BaseTx(half/full), auto Single 10/100 - ZX345Q - X1001B - 10BaseT(half/full), 100BaseTx(half/full), auto Quad 10/100 - ZX346Q - X1012B - 10BaseT(half/full), 100BaseTx(half/full), auto F200/300 series onboard - 10BaseT(half/full), 100BaseTx(half/full) F700 series onboard - 10BaseT(half/full), 100BaseTx(half/full), auto Here is a small note on autonegotiation in the ethernet world: Q. What is Auto-negotiation? A. Auto-negotiation is an optional function of the IEEE 802.3u Fast Ethernet standard that enables devices to automatically exchange information over a link about their speed and duplex abilities. This allows devices to perform automatic configuration to achieve the maximum common level of operation over a link. Q. How does Auto-Negotiation work at Both Ends? A. When two auto-negotiation devices with multiple capabilities are linked together, they find their maximum common mode of operation based on a priority table. The following table lists priorities from the highest (1) to the lowest (5). The full-duplex mode of operation is given higher priority over half-duplex because, at the same speed, more data can traverse the full-duplex link. ------------------------------------------------------- Priority Capabilities ------------------------------------------------------- 1 100BaseT4 (not supported on NetApp filers) 2 100BaseTX Full Duplex 3 100BaseTX 4 10BaseT Full Duplex 5 10BaseT -------------------------------------------------------- Q. How does Auto-Negotiation work at One End? A. If one link partner is set to auto-negotiate and the other link partner is manually set, the manually set partner must always be half-duplex.The IEEE spec does not allow the auto-negotiating partner to detect full-duplex on a manually set link partner. auto ----X--> 100BaseT-FD auto -------> 100BaseT-HD If one end is configured to operate in 100BaseTx fullduplex make sure that the other end is also manually configured to operate in the same mode. If there is a speed/duplex mismatch, you might see CRC errors and poor network throughput. Hope this helps. -Devi Network Appliance

4 3

RE: UNCORR PROC MCHK 98
by Bulfer, David 04 Mar '99

04 Mar '99

This is not a memory error on and F630. For an F630, an MCHK 98 means "Processor Hard Error". It is a SW problem associated with SNMP. Access NOW and check out bug 11610. David Bulfer -----Original Message----- From: G D Geen [mailto:geen@msp.sc.ti.com] Sent: Wednesday, March 03, 1999 1:46 PM To: 'toasters(a)mathworks.com' Cc: Peter Pross; Mike Ball Subject: Re: UNCORR PROC MCHK 98 We were told that this was a memory issue as well. This was not so in our case. Ours occurred when a network device discovery was run from the Spectrum server. This did not affect the system running version below 5.1.1. If you are running some sort of network monitoring tools such as Cabletron's Spectrum server, the SNMP call made by the discovery may be causing your problem as it did ours. NetApp took ownership of this issue and created a patch to fix our problem. Though this problem had been reported by other vendors, Cabletron does not consider it a bug in *their* code. -gdg Peter Pross wrote: > > Hi, > > I believe this may be memory errors. I think you aloud 1 in a blue moon, if > more replace mem. Check with netapps. > > Cheers, > > Peter > > __^__ > ( ___ > )--------------------------------------------------------------------------- > ------ > | / | Peter Pross Title: Unix Systems > Administrator > | / | Nortel Networks Montreal Email: pross(a)nortel.ca > | / | 16 Place du Commerce Voice: 514-765-7973 > | / | Nun's Island, Quebec, Fax : 514-761-8710 > > | / | Canada H3E 1H6 ESN : 852-7973 > > |___| > > (_____)--------------------------------------------------------------------- > ----------- > > > -----Original Message----- > > From: Mike Ball [SMTP:mball@rtp.opensystems.com] > > Sent: March 3, 1999 2:06 PM > > To: 'toasters(a)mathworks.com' > > Subject: UNCORR PROC MCHK 98 > > > > Does anyone have an idea of what the below error is pointing to on my F630 > > running 5.21? > > > > "Wed Mar 3 09:26:49 EST [rc]: relog syslog UNCORR PROC MCHK 98 > > pc:fffffc0000482260 cia:0 mm:4090 eia:1820f550 fs:ff00 eis:945000000 isr: > > 0 pci0:3010106 pci1:4000" > > > > I discovered my system was locked up this morning when I came in. After > > rebooting i discovered the following error in my messages file. > > > > -- --------------------------------------------------------------- G D Geen mailto:geen@ti.com Texas Instruments Phone : (972)480.7896 System Administrator FAX : (972)480.7676 --------------------------------------------------------------- Life is what happens while you're busy making other plans. -J. Lennon

3 2

RE: Gb-ether and Extreme: closure! :-)
by Garrett Burke 04 Mar '99

04 Mar '99

Hi Bruce, You'll find the same thing happens if you setup port mirroring. When you turn mirroring off the port isn't in any VLAN and hence doesn't pass any packets. Couldn't find it mentioned in any of the Extreme docs...must doc it for my colleagues..now where's my clone and a 48hr day As an aside, have you run a filer into a 100Mb/s port on a S48? I'm seeing CRC errors on the port that my F230 is on. Netapp are investigating, just wondering if anyone else is seeing/has seen this. Regards, GB > ---------- > From: Bruce Walker[SMTP:bmw@visgen.com] > Sent: 04 March 1999 09:02 > To: toasters(a)mathworks.com > Subject: Re: Gb-ether and Extreme: closure! :-) > > > But what happens when you unconfigure a trunked port on a Summit, > apparently, is that the port ends up not being on *any* VLAN. The > port is "alive" but nothing is routed to or from it. > >

3 2

Re: newbie asking about autoloaders
by rbutler＠pdc.com 04 Mar '99

04 Mar '99

> From bmw(a)visgen.com Wed Mar 3 19:26:09 1999 > From: Bruce Walker <bmw(a)visgen.com> > Subject: Re: newbie asking about autoloaders > To: kfischer(a)rsinc.com (Kathy Fischer) > Date: Wed, 3 Mar 1999 20:20:43 -0500 (EST) > Cc: toasters(a)mathworks.com > X-Mailer: ELM [version 2.4 PL24 ME5a] > Content-Type: text > Sender: owner-toasters(a)mathworks.com > Content-Length: 1536 > > Kathy Fischer writes: > > > > Can anyone out there recommend an autoloader (DLT type), or warn against any > > for use with the toaster? > > If it doesn't have to be DLT, the Exabyte Mammoth tapes get excellent > performance and are cheaper. I am getting better than 5 Mbytes/sec > continuous xfer, which lets me backup 65 Gigs in about 4 hours > using dump on the filer. This is to an Exabyte F220 autoloader > (10 tapes) running in dumb mode, attached directly to an F720. > > Well almost directly; I have a Rancho differential SCSI to single-ended > converter mashed in between the Exabyte and the F720. I ordered > both the F720 and the Exabyte at the same time and failed to notice > that the 720's come with differential SCSI! I sure croaked that > filer bad when I connected a plain 68pin to 50pin HD SCSI kluge > adapter to it. Kernel fault and core dump. > > Supposedly, BudTool works with the Exabyte too, but I may never be > able to find that out :-) > > On the performance thing: it's important that the filer be able to > get the data rate up high enough to prevent the DLT tapes from > shoe-shining which brings down their effective xfer rate a lot. > I had a DLT4000 drive for a while before I bought the F220 and it > shoe-shined a lot when driven from an F330 filer. It only managed > to do about as well as the 4mm DAT drive it was temporarily replacing. > > I kinda like helical scan technology for that start/stop ability. > > Cheers! > -- > -bmw | Double helix in the sky tonight > | Throw out the hardware > | Let's do it right -- Steely Dan; Aja > Hi, BudTool supports most all libraries that are manufactured. This includes the Exabyte 220. There are a number of DLT libraries with DLT7000 and DLT4000 tape drives in them. In my opinion, DLT is the defacto standard in the industry. Mammoth may be a little cheaper, but was never really accepted widely by the market. The top three Tape Library manufacturers are StorageTek, ATL, and Breece Hill. Any tape library from these manufacturers will give you very good service for the long term. If you go with 8mm technology such as the Mammoth or Sony AIT, then Exabyte or Spectrologics makes excellent tape libraries to fill this nitch. As an aside, the top three Tape Library manufacturers I mentioned above (STK, ATK and BH) decided to only support DLT tape technology for OpenSystems (UNIX) and passed on Mammoth and AIT. Randy Butler System Engineering Supervisor - Western Region PDC Solutions, Inc. phone: 972-446-2644 x15 Systems Engineering Solutions FAX: 972-245-9320 1445 MacArthur Drive, Bldg. 1, Suite 116 Carrollton, TX 75007 E-mail: rbutler(a)pdc.com //////////// //////////// ////////// // /// // // /// // // //// // /// // // /// // /// // /////////// // /// // // // /// // // // /// /// /// //// ///////////// /////////////

1 0

Sun Trunking
by Mohler, Jeff 04 Mar '99

04 Mar '99

Im looking for a reference to the Sun Trunking software..and Im unable to locate so far. Sorry for the off topic, but if anyone has a reference...thanks.

1 0

RE: VIF command
by Mohler, Jeff 04 Mar '99

04 Mar '99

> What switch are you using (mfg/model)? There are many differences in how > load > balancing is implemented. I have even found differences in different > model > switches from the same manufacturer. All that I have seen are based on > MAC > addresses - Either source, destination or a combination. Some mfg's let > you > choose which to use. If your client base is reasonably large with random > MAC > address and equal loads, then the traffic will be balanced. Small client > pools > or most traffic being generated/requested by a few clients can skew the > load. > Some implementations have standby ports in the trunk, which seems like > what you > are seeing. I believe Extreme allocates one port out of the trunk as a > standby. > Perhaps this is configurable. --- Its a FEC blade in a Cisco 5505.

1 0

RE: VIF command
by Naden, Barbara 04 Mar '99

04 Mar '99

What switch are you using (mfg/model)? There are many differences in how load balancing is implemented. I have even found differences in different model switches from the same manufacturer. All that I have seen are based on MAC addresses - Either source, destination or a combination. Some mfg's let you choose which to use. If your client base is reasonably large with random MAC address and equal loads, then the traffic will be balanced. Small client pools or most traffic being generated/requested by a few clients can skew the load. Some implementations have standby ports in the trunk, which seems like what you are seeing. I believe Extreme allocates one port out of the trunk as a standby. Perhaps this is configurable. Barbara -----Original Message----- From: Mohler, Jeff [mailto:jeff.mohler@wilcom.com] Sent: Thursday, March 04, 1999 7:01 AM To: 'Sam Cramer'; Mohler, Jeff Cc: 'toasters(a)mathworks.com' Subject: RE: VIF command > If you unplug one of the busy links, does one of the idle links pick up > the > slack? > > Sam --- Yes, a spare link picks up the dead one, and holds onto it when you restore the dead link.

1 0

← Newer
1
...
12
13
14
15
16
17
18
19
20
Older →

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

toasters March 1999