toasters September 1999

toasters@lists.teaparty.net

81 participants
95 discussions

Re: moving stuff among qtrees
by sirbruce＠ix.netcom.com 04 Sep '99

04 Sep '99

On 09/03/99 10:55:20 you wrote: >Can anybody suggest a way to schlep a hierarchy around that is not >O(n) in this annoying way? A qtree identifier is essentially just a third ID, just like user and group ID, set on each file and inherited from the parent directory. So to change the ID requires essentially a recursive chown-type operation that is always going to be proportional to the amount of data "schlepped". I suppose one could implement a scheme where the changes were not made right away, and instead upon access each parent was checked back to the root to see if the file has the "wrong" qtree ID for the qtree and if so, correct it and then let you access it, but that would be very slow. Bruce

2 1

Web Hits per NFS ops/s ?
by Steven Clarke 04 Sep '99

04 Sep '99

Not sure about the theoretical case but here's some stats from a system of heavily tuned Apache web servers running Linux 2.0 (NFS V2) mounting a NetApp for all content. It's only serving plain content - no scripts. Average NFS operations per url is 7.96. getattr 31867993 10% lookup 142423109 45% read 145310674 45% readdir 58359 0% statfs 30 0% File size breakdown: Percent File Sizes 0.000 0 0.000 1b- 10b 0.001 11b- 100b 0.172 101b- 1kb 7.633 1kb- 10kb 19.897 10kb-100kb 10.331 100kb- 1Mb 61.822 1Mb- 10Mb 0.145 10Mb-100Mb Steve Clarke

1 0

Windows 2000 domain controllers
by Brian Rice 03 Sep '99

03 Sep '99

Hi, all. One of my clients, for excellent reasons of its own, is migrating its domain controllers to Windows 2000 (which is the artist formerly known as NT 5). This means that their CIFS-running NetApps have to play nice. I had been aware that NetApp had put a fix for this into 5.3D16, and when 5.3.2R1 came out I (naively, as it turned out) assumed that that fix would be included. So I brought up a borrowed test filer on 5.3.2R1, and found that it would happily register itself in an Windows-2000-controlled domain...once. If you shut CIFS down and restarted it, the filer would give up on finding a domain controller, and client accesses to the filer would time out. And then you often would not be able to cleanly terminate CIFS again; after 90 seconds of waiting, the filer would invite you to shut down the system using halt -d and send the resulting core dump to NAC. Heh. A call to my friendly tech-support person revealed that the fix was actually not in 5.3.2R1, but was in 5.3.2D3. Aha. So I moved to 5.3.2D3, and now it's playing nice. The build of W2K in play here was 2072. Hope this helps other pioneers out there. Brian

2 1

Re: Web Hits per NFS ops/s ?
by sirbruce＠ix.netcom.com 03 Sep '99

03 Sep '99

On 09/02/99 23:51:59 you wrote: >(The number just thrown out, with no justification, was > > Each "hit" results in 5 NFS operations. > >I'm sitting here pounding my head trying to understand the "5". > >V3 might yield: > > LOOKUP > ACCESS > READ > >Per page hit. Is average object size 8 - 12KB? So one read on V3? >I assume Apache front ends are not "browsing" they're firing names >at the filer. Etc. It has to be even more on the average. First you'd have to deal with pathname translations, which would be stored in a file somewhere on the filer that would have to be accessed first. I suppose it might keep those in memory, so it doesn't have to read that file with every hit, but you also have other files that govern access control potentially for each directory. Then your web server will probably also log each hit, and since that log file is going to get big you're going to be putting that on the filer as well. And I'm sure there must be more. It all depends on exactly what the web server has to handle when doing a hit. There may even be temp files involved that are created and deleted. Bottom-line is I would expect a hit to yield, say, 10 ops, but remember the average page has many objects all of which have to be open and read, although they may not be all logged. Bruce

1 0

moving stuff among qtrees
by Brian Rice 03 Sep '99

03 Sep '99

Here's something that's puzzled me about NetApps for a long time. I would often like to move entire hierarchies of directories out from under one qtree into another. And I'd like to do it fast. I can see why this should take a long time if the qtrees were in different volumes, but they're not. And I can see why this should take a long time if quotas were being enforced, but I do a quota off before the move and a quota on afterwards. And it still takes forever...a forever, in fact, that is proportional to the quantity of data being relocated, so it looks as if files are being copied on the disk surface. Huh? Can anybody suggest a way to schlep a hierarchy around that is not O(n) in this annoying way? (Incidentally, the last time I tried this was on a 5.1.x). Brian

2 1

Backup F720 with Legato/NFS
by Sanborn, Ed 03 Sep '99

03 Sep '99

Hi all, Does anyone have experience backing up a filer with Legato via NFS? So I need to just backup the snapshots, eh? Legato doesn't backup NFS partitions by default so I'm assuming I'll need a custom directive, right? Does anyone have an example directive? My Legato (version 5.0) server is a Sun running Solaris 2.6. Thanks, Ed Ed Sanborn (978) 691-6496 Northchurch Communications Inc. 5 Corporate Drive Andover, MA. 01810 Fax (978) 691-6300 http://www.northchurch.net <http://www.northchurch.net/>

1 0

Web Hits per NFS ops/s ?
by beepy＠netapp.com 03 Sep '99

03 Sep '99

Hey, hey. Without divulging corporate secrets, does anyone have rules of thumb for some "normal" activity, how much NFS traffic if generated out the back of a beefy Web Client to a filer that is serving up the pages via NFS? Rough feels and spreads is all I was curious about... I'm not sure I have ready access to a Web front end/filer back end setup that would be meaningful... Mostly I'm curious. (The number just thrown out, with no justification, was Each "hit" results in 5 NFS operations. I'm sitting here pounding my head trying to understand the "5". V3 might yield: LOOKUP ACCESS READ Per page hit. Is average object size 8 - 12KB? So one read on V3? I assume Apache front ends are not "browsing" they're firing names at the filer. Etc. I thought it an interesting question. What does the op mix look like for a filer backing a Web host? (I've seen the mix, I guess I was never startled enough to remember, seemed boring with lots of LOOKUPs, 10% READs, 5 - 8% WRITEs?). beepy

1 0

RE: Anyone mounting filer on Linux with smbfs ?
by Sam Schorr 02 Sep '99

02 Sep '99

Stephen, We are mounting multiple filers (2 F740's, 2 F760's), running OT 5.3.2D2, using NFS on Linux running on a Compaq. We are almost exclusively a CIFS environment and had wanted to do exactly what you want. However, there are known problems with OT that prevent using Samba (for example) to do a direct CIFS mount. So, we have a mixed ACL on the filers, run both CIFS and NFS, and mount the filers from Linux as NFS. It does require some tricks with WAFL settings to insure that non-trusted UNIX users are defaulted to the proper NT user domain account, and you have to make sure that the wafl options for unicode are set to both convert and create. If you convert from 5.2.x or prior, and your filer data has been created through CIFS, you will HAVE TO go through some procedure to open every directory to force conversion. If you do not, any directory not converted will be inaccessible in its snapshot, and snapshots cannot be converted. --srs -----Original Message----- From: Stephen C. Losen [mailto:scl@sasha.acc.virginia.edu] Sent: Thursday, September 02, 1999 12:54 PM To: toasters(a)mathworks.com Subject: Anyone mounting filer on Linux with smbfs ? Linux provides a SMB fs type and we are trying to mount a netapp CIFS share on a linux box, but with no luck. We are using unix authentication on the filer from a NIS map. We have no Windows or NT authentication set up at all. Windows and NT attach to CIFS shares on the filer just fine, using cleartext passwords. And they also attach to a linux samba server just fine, with unix authentication and cleartext passwords. But we can't get linux to mount a smb fs from the netapp. We would like to do this because we expect a large number of users to run linux on their desktops and we would love to not have to manage the exporting nightmare that NFS would require -- especially when most folks will be getting IP addresses via DHCP. It would be much more convenient (for us sysadmins) for linux users to just mount their home dir as a smb fs type, and they authenticate with their unix password. Steve Losen scl(a)virginia.edu phone: 804-924-0640 University of Virginia ITC Unix Support

1 0

Anyone mounting filer on Linux with smbfs ?
by Stephen C. Losen 02 Sep '99

02 Sep '99

Linux provides a SMB fs type and we are trying to mount a netapp CIFS share on a linux box, but with no luck. We are using unix authentication on the filer from a NIS map. We have no Windows or NT authentication set up at all. Windows and NT attach to CIFS shares on the filer just fine, using cleartext passwords. And they also attach to a linux samba server just fine, with unix authentication and cleartext passwords. But we can't get linux to mount a smb fs from the netapp. We would like to do this because we expect a large number of users to run linux on their desktops and we would love to not have to manage the exporting nightmare that NFS would require -- especially when most folks will be getting IP addresses via DHCP. It would be much more convenient (for us sysadmins) for linux users to just mount their home dir as a smb fs type, and they authenticate with their unix password. Steve Losen scl(a)virginia.edu phone: 804-924-0640 University of Virginia ITC Unix Support

1 0

Has anyone been able to get FrontPage 98 extensions working with Netapp filers?
by Mike Ball 02 Sep '99

02 Sep '99

I am running Data Ontap version 5.3D20 on F230 authenticating to NT 4.0 domain. We are unable to access web pages created in FrontPage 98 stored on the Netapp. The error I receive when I try to access the pages states "credentials passed by the client do not have access to the particular resource. Unauthorized due to ACL on resource". Has anyone been able to run FrontPage extensions on the Netapp? Thank you, Mike Ball mball(a)opensystems.com

1 0

← Newer
1
2
3
4
5
6
7
8
9
10
Older →

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 September 1999