toasters

toasters@lists.teaparty.net

1 participants
13532 discussions

RE: Performance Indicators
by Glenn Walker 15 Sep '06

15 Sep '06

I believe that this is more along the lines of QoS (though not true QoS and not guaranteed). What Jochen is looking for is a bit difficult: you can't really guarantee performance on the head due to workload differences. Each of the machines Jochen listed is a very capable machine - but selling IOPS to a user is not easy. Each workload is different, each application would require a different amount of IOPS in order to facilitate functionality (hard to define these sometimes), you can't easily segregate workloads and adding workload from one user\system would definitely affect existing workloads. For example: CIFS typically requires about 5 IOPS to open a file as a minimum - if you add some newer functionality from 2k3\XP and MS Office, you could be talking about hundreds of IOPS to open a file (pulling file metadata). Perhaps the best thing to do is to talk to sales to get actual numbers, but simply list the capabilities of the filer and manage how the workload is spread among them. One last point: when you speak of performance, you mentioned IOPS - this is workload, not performance. Performance is more a measure of latency at a particular level of work: ie, how much latency for a given amount of IOPS. If I stated that the R200 was capable of about 12000 IOPS (CIFS) per second, that would be true - but I could also position it as being capable of higher if the latencies of the workload were to increase (still the same amount of work, but response time would decrease). You could speak to the user-base about performance in terms of latency and IOPS. Glenn -----Original Message----- From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Wischer, Sebastian Sent: Friday, September 15, 2006 5:18 AM To: Willeke, Jochen; NetApp Toasters List Subject: RE: Performance Indicators HI Jochen, have you think about FlexCache?? There you give volumes priorities. I know it is not exactly what you are looking for but perhaps with the help of FlexCache you can sell categories to your departments. Here is a small summary from a NetApp slideset regarding FlexCache: FlexShare(tm) is enabled when there are system bottlenecks CPU NVRAM Disk Control of priority by volume Five priority levels Sebastian -----Original Message----- From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Willeke, Jochen Sent: Friday, September 15, 2006 10:32 AM To: NetApp Toasters List Subject: Performance Indicators Hi Toasters, right now i am thinking about any kind of performance indicator for my fileservice. We do supply storage to other departments of our company. Of course one thing i can sell is space. But there is another indicator which is performance. Has anybody good ideas how to define the performance i can grant to a special service? I was thinking about I/O's per second but do not know what a good number would be. Has anybody ever had to work out those values and can give me something to think about? I am interesting in any experience you had with this topic... I/O's, data-throughput, responsetime....And what are good values which i can grant with netapp systems? We use R200, FAS960 and FAS3050. Thanks in advance Jochen

2 1

Need Advice for changing qtree from Unix to NTFS security style
by Stephen C. Losen 15 Sep '06

15 Sep '06

We have a filer that serves files in a qtree with both NFS and CIFS. The file security style is currently Unix. The qtree has about 7000 files using about 150G. CIFS authentication is done via a Windows DC and we have a Unix passwd NIS map for converting the Windows users to Unix users. There are over 100 folks using this share. Each one owns a directory (not a home directory -- just storage) and each one wants to be able to grant some of the 100 other users access to their folder. Plus they have an admin who needs to be able to manipulate access rights on everything. Obviously Unix security style isn't going to cut it. So we want to switch the qtree to NTFS security style. We switched a test qtree from Unix to NTFS style and discovered that no one (not even a Windows user with admin privs in the domain) has "modify" or "full control" rights on any file or folder that already exists. So it looks like we can't change the ACLs on anything that already exists. If a Windows user creates a new folder or file, then he has the ability to change the ACL. Has anyone else switched a qtree from Unix to NTFS style? Is there anything we can do either before or after we switch the security style to make it possible to set ACLs on files that already exist? I didn't try logging in to CIFS as a local filer user. Is that the secret? I suppose we could create a new empty qtree with NTFS style and copy the files from the old qtree to the new one. But I would really like to avoid that if possible. Steve Losen scl(a)virginia.edu phone: 434-924-0640 University of Virginia ITC Unix Support

2 2

Re: Need Advice for changing qtree from Unix to NTFS security style
by Stephen C. Losen 15 Sep '06

15 Sep '06

> > We switched a test qtree from Unix to NTFS style and discovered that no > one (not even a Windows user with admin privs in the domain) has "modify" > or "full control" rights on any file or folder that already exists. So > it looks like we can't change the ACLs on anything that already exists. > If a Windows user creates a new folder or file, then he has the ability > to change the ACL. > Ooops. Looks like I was wrong here. The owner of the file or folder does have full control. So I guess I can chown everything to me and then from my Windows account I can set all the ACLs. I'll check and see if the local filer "administrator" user can do it. Steve Losen scl(a)virginia.edu phone: 434-924-0640 University of Virginia ITC Unix Support

1 0

Performance Indicators
by Willeke, Jochen 15 Sep '06

15 Sep '06

Hi Toasters, right now i am thinking about any kind of performance indicator for my fileservice. We do supply storage to other departments of our company. Of course one thing i can sell is space. But there is another indicator which is performance. Has anybody good ideas how to define the performance i can grant to a special service? I was thinking about I/O's per second but do not know what a good number would be. Has anybody ever had to work out those values and can give me something to think about? I am interesting in any experience you had with this topic... I/O's, data-throughput, responsetime....And what are good values which i can grant with netapp systems? We use R200, FAS960 and FAS3050. Thanks in advance Jochen

2 2

the Toasters list is now restricting posts!
by Kevin Davis 12 Sep '06

12 Sep '06

Folks, Just an FYI to all: we've had to take the measure of restricting posting to the 'toasters' list, meaning that the list will only accept posts from an address which is subscribed. Sorry for taking this measure, but the nature of email today has left us no alternative (not one that will still allow us to get our jobs done, that is!); it's a way to keep the spam to the list to a minimum. And for what it's worth, the spam to the list is almost nil now. (knock wood and cross your fingers). Hopefully, this measure along with the four (yep...four!) separate facilities we have for filtering the gooey pink stuff will be sufficient to make the list a friendlier place again. For most, this will have no effect. But some may be subscribed via list addresses which will be expanded on the destination domain, and for these folks the ability to reply to or start new threads will be foiled, unless the "From:" address matches the address which is subscribed to the list. In other words: Admin1 is one of 4 admins at XYZ Corp. who are all part of the alias 'toasters(a)xyzcorp.com'. The admins at XYZ subscribe 'toasters(a)xyzcorp.com' to 'toasters(a)mathworks.com'. An email to the 'toasters' list goes to XYZ and is expanded to all 4 admins. Admin1 then they won't be able to reply or create new posts because 'admin1(a)xyzcorp.com' is not subscribed to the 'toasters' list. So...sorry for any inconvenience to individuals subscribed in this manner, but you'll have to subscribe individually in order to post to the list. -- *------------------------------------------*-----------------------* | Kevin Davis (UNIX Sysadmin) | Natick, Massachusetts | | 508.647.7660 | 01760-2098 | | mailto:kevin.davis@mathworks.com *-----------------------* | http://www.mathworks.com | | *------------------------------------------*-----------------------*

2 2

RE: Snap status and wafl scan
by Glenn Walker 08 Sep '06

08 Sep '06

You are correct - the reclamation is a new feature - introduced around 7.0 I believe. Unsure what version of ONTAP the issue was being seen on, but there were some bugs related to that feature-set that were fixed in later versions - some bugs were related to its efficiency, while some were related to the timing (when the freed space was showing back up). Glenn -----Original Message----- From: Chris Thompson [mailto:cet1@cus.cam.ac.uk] Sent: Friday, September 08, 2006 12:07 PM To: toasters(a)mathworks.com Cc: Glenn Walker Subject: Re: Snap status and wafl scan ggwalker(a)mindspring.com (Glenn Walker) writes: > Ok - those are other things than what I previously mentioned. > > Active Bitmap rearrangement is normal - I believe it has been around > since 6.5 (pretty sure it wasn't a flexvol thing, that was 'deswizzler' > scan). It's to keep memory 'pretty', if memory serves. > > The container block reclamation is a scanner responsible for ferreting > out blocks that have been freed and marking them as usable to the system > (delayed free scanner\delayed freer working here). This too is normal. Just in case it wasn't clear, I did not mean to suggest that these things were not "normal", or that they had any implications for filing system integrity. It was the performance aspect of the "container block reclamation" scan that originally brought it to my attention: specifically that after our dumps had finished (and the associated snapshots thus deleted) disc I/O remained high and cache age very low for some time. I don't think ONTAP used to behave like that in 6.5. By contrast the "active bitmap rearrangement" scan (which is always running: it just cycles around the block numbers indefinitely) has no bad effect on performance that I have ever noticed. -- Chris Thompson Email: cet1(a)cam.ac.uk

1 0

RE: jumbo frames support
by Fox, Adam 08 Sep '06

08 Sep '06

Yes, Jumbo frames are supported via the ifconfig command. The command allows you to set the mtusize. Although it's probably worth opening a support case and have them run a perfstat on your system while you are experiencing slow performance. If it's really as slow as you say, it's probably not the mtusize. It might be, but there are probably other things you can adjust in your environment first. But NGS has the resources to work with you to figure those out. Just a thought. -- Adam Fox adamfox(a)netapp.com -----Original Message----- From: kunle windapo [mailto:kwindapo@yahoo.com] Sent: Friday, September 08, 2006 11:26 AM To: toasters(a)mathworks.com Subject: jumbo frames support hello, i have a FAS 270 with volumes nfs mounted on a Solaris 9 sparc server. there is a direct connection from the solaris server to ethernet interface on the FAS 270 running at 1Gb/s. problem is i experience very slow performance on the system, (io related) and i'm troubleshooting. i want to know if NetApp supports Jumbo frames on the Network interface. thank you __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

1 0

jumbo frames support
by kunle windapo 08 Sep '06

08 Sep '06

hello, i have a FAS 270 with volumes nfs mounted on a Solaris 9 sparc server. there is a direct connection from the solaris server to ethernet interface on the FAS 270 running at 1Gb/s. problem is i experience very slow performance on the system, (io related) and i'm troubleshooting. i want to know if NetApp supports Jumbo frames on the Network interface. thank you __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com

2 1

RE: Snap status and wafl scan
by Glenn Walker 08 Sep '06

08 Sep '06

Ok - those are other things than what I previously mentioned. Active Bitmap rearrangement is normal - I believe it has been around since 6.5 (pretty sure it wasn't a flexvol thing, that was 'deswizzler' scan). It's to keep memory 'pretty', if memory serves. The container block reclamation is a scanner responsible for ferreting out blocks that have been freed and marking them as usable to the system (delayed free scanner\delayed freer working here). This too is normal. Glenn -----Original Message----- From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com] On Behalf Of Chris Thompson Sent: Thursday, September 07, 2006 4:01 PM To: toasters(a)mathworks.com Cc: ecoli82(a)msn.com; etraitel(a)gmail.com Subject: Re: Snap status and wafl scan Eyal Traitel writes: > Note that snap status is an advanced command so its output shouldn't > necessarily cause you concerns... Well, while we're talking about things wot mere mortals were not meant to wot of... ec0li <ecoli82(a)msn.com> wrote: > When I check snap status, I see next to recent snapshots : (xxxx/xxxx > remaining). > Also, It seems that this is in fact the wafl scan process "snap create > summary update" The effect worrying me is somewhat different. After a snapshot is deleted the above state lasts only seconds, but there is then a much more extended period where "snap status" looks normal but "wafl scan status" shows e.g. carina*> wafl scan status Volume CUS: Scan id Type of scan progress 1 active bitmap rearrangement fbn 3340 of 4461 w/ max_chain_len 3 925 container block reclamation block 382 of 4461 The "container block reclamation" scan lasts several minutes (and this is a volume & aggregate of only a few hundred GB), and while it's going on disc read activity is high and severely impacts the cache - at least, if you trust the "cache age" shown by sysstat, which I am not sure I do. This is with ONTAP 7.1.1P1 (but similar effects observed in earlier 7.x's). -- Chris Thompson Email: cet1(a)cam.ac.uk

2 1

NDMP backups, timeouts
by Scott T. Mikusko 08 Sep '06

08 Sep '06

I'm running into a recurring issue with backing up my NearStore and a long F760 filer. In the procoess of backing some large filesystems (qtrees), it appears that NDMP is basically timing out and fails. I'm running Veritas NetBackup 6.0 (finally upgrading from 4.5 recently). I think the issue has to do with NDMP starting the dump commands on the filer, which then does a three-way backup to another filer with local tape drives. The debug from ndmp logs shows for a manual full backup of a filer: Sep 05 21:05:57 EDT [ndmpd:58]: Log message: DUMP: creating "/vol/vol0/../snapshot_for_backup.370" snapshot. Sep 05 21:06:05 EDT [ndmpd:58]: Log message: DUMP: Using Partial Volume Dump with Exclude Lists Sep 05 21:07:31 EDT [ndmpd:58]: Log message: DUMP: Date of this level 1 dump: Tue Sep 5 21:05:57 2006. Sep 05 21:07:31 EDT [ndmpd:58]: Log message: DUMP: Date of last level 0 dump: Fri Aug 18 19:20:57 2006. Sep 05 21:07:31 EDT [ndmpd:58]: Log message: DUMP: Dumping /vol/vol0/files3.rt to NDMP connection Sep 05 21:07:31 EDT [ndmpd:58]: Log message: DUMP: mapping (Pass I)[regular files] It never completes the mapping of the files in Pass I. It just sits there. The filesystems I'm backing up are between 250-650GB with a LOT of small files (millions) and many subdirs. Tons of mail and html files. My guess is that the amount of files and dirs are getting large that NDMP can't map them fully and is timing out after 8 hours. It shouldn't take that long to map files for the size of data it's doing; I've seem mapping times be much less for larger sets of data. I know that the filesystem topology can cause NDMP backups to be slow, depending on file sizes, data layout, filer load, network etc, but this is happening on moderately busy filers or bone idle NearStores. Anyone run across lengthy DUMP times? Chewing through NOW and Veritas support site hasn't turned up anything obvious. Just curious if others have run into NDMP/dump issues like this. -Scott

4 3

Jump to page:

2026

2025

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