toasters

toasters@lists.teaparty.net

13531 discussions

ndmp restores
by Ian Dotson 21 Dec '11

21 Dec '11

We've just started using our first Netapp here. One surprise involves backups (we're using NetVault's NDMP plugin). I'll let our local backup guy explain: "We're setting up our new netapp, have the ndmp plugin in place and can back up ok. But I don't see how to set up backups of netapp filer volumes so that I can do restores in the same way we can do restores from windows file servers. Specifically, the ndmp restore window only shows the files backed up in that specific dump, so if I want to restore all files that existed in a directory up to that date, it appears I have to separately do a restore of the level 0 dump, then all pertinent level 1 dumps, then all pertinent level 2 dumps, etc. With windows file restores, I could restore the whole directory with one job. How do I set up the netapp ndmp to work like that? " The NetVault folks reply that "the NDMP protocol doesn't allow for that. Each restore builds on the previous restore, so you have to do Level 0, Level 1, etc.. It would be the same with any filer that utilizes the NDMP protocol" -- which doesn't seem right at all. So, is this indeed the same for those using NetBackup, etc.? Are you forced to restore each dump level individually? Thanks, Ian Dotson

2 2

Flex_scale licensing question.
by Jeff Cleverley 20 Dec '11

20 Dec '11

Greetings, I'm confused about the flex_scale licensing. We have installed PAMII modules in 2 of our clusters running 7.3.x. We have a license for pamii, but not for flex_scale. To manage the extended cache, it has us set flexscale.xxx options. Since we don't have flex_scale licensed, are these options doing any good at all? I didn't know if flex_scale was tied to the old flashcache modules and pamii now incorporates the same thing for the new PAMII cards. Thanks, Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

3 6

Disk space increasing after snapshots are deleted.
by Jeff Cleverley 18 Dec '11

18 Dec '11

Greetings, We have a 8.0.1 filer that runs as a NearStore for SV backups. The group using this normally kept 120 nightly backups. The aggregates were getting full so they decided to delete all snapshots older than 90 days. This was done with a script using snap delete <vol> <snapshot> with a list of snapshots fed into the command. Some of the lists were not in reverse order so it deleted nightly.91 before nightly.119. I know deleting them in this type of order will work anyway, but it causes the system to do more processing. The problem is the disk space is continuing to grow on all volumes in the aggregates. An aggr show_space of the aggregates shows everything increasing in used space. The volumes have no space reservation. I'm trying to figure out how this mass deletion of snapshots (several hundred) is causing an increase in space usage. Any ideas? Thanks, Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

4 6

Installing & Testing Flash Cache cards
by Fletcher Cocquyt 15 Dec '11

15 Dec '11

We are taking delivery of a couple flash cache cards (512Gb demo units) soon. I plan to install them in our standby 3270 cluster then present several cache enabled volumes for various benchmark us cases (Oracle, VMware, Web etc) Qs: 1 - how easy/hard is the physical installation? Any gotchas to avoid? 2 - can I enable / disable caching on the volumes to help define the benefit flash cache is providing? 3 - how do I best monitor the flash cache metrics (I use cacti to graph "Cache Age" currently) thanks -- Fletcher Cocquyt Principal Engineer Information Resources and Technology (IRT) Stanford University School of Medicine Email: fcocquyt(a)stanford.edu Phone: (650) 724-7485

5 6

flexcache and cold caches
by Peter D. Gray 15 Dec '11

15 Dec '11

Hi toasters, we have just recently "upgraded" our netapp infrastructure and moved from lots of small FC drives, to a much smaller number of SATA drives. We knew this would cause a performance decrease and compensated by installing 512GB flexcache cards. Its a HA 3170A. So far it all kind of works. However, its clear that the caches take a long time to heat up, several hours in fact. During that time, performance is "average" to say the least. So, I guess I have 2 questions for people who are doing similar things: 1) are there any tips on ways to improve the cold start performance (I cannot think of any). 2) if I have a HA failover, I imagine my performance will be even worse since the second heads cache will be completely cold for the data from the failed head and to make things worse, the second head will have to eject data from the cache for its own local volumes. Anybody with experience of this? 3) Any bright ideas other than more spindles? It would be nice if the caches were nvram, to avoid the cold start problem. I really netapp should use SSD's for the cache for exactly this reason. Or am I missing something? Regards, pdg

4 5

Re: 64 bit root volume conversion.
by Jeff Cleverley 15 Dec '11

15 Dec '11

>From what I can see from the audit logs they ran "ndmpcopy /vol/old_vol0 /vol/vol0" after they had renamed the original vol0 to old_vol0. I checked the audit logs on the test system and they did the ndmpcopy from vol0 to new_vol0, then did the rename. Would running the ndmpcopy when the original vol0 had been renamed first affect any of the files or the copy somehow? I wouldn't expect so, but I'll have him try it again in the morning. Jeff > > On Dec 14, 2011, at 2:10 PM, Jeff Cleverley <jeff.cleverley(a)avagotech.com> wrote: > >> Greetings, >> >> We have a filer (3140) running 8.0.1 and the group wanted to convert >> it to a 64 bit root aggregate and vol0. I let them practice on my 8.1 >> test box and everything went as planned. When they did it on their >> 8.0.1 box, all went as plan except for the assigning the root volume. >> >> They had moved the original vol0 out of the way and renamed the new 64 >> bit volume to vol0. When they ran the "vol options vol0 root" command >> it gave the following output: >> >>>> >> Necessary system files are not present on the volume 'vol0'. It is >> recommended that you copy system files from the current root volume to >> the new root volume using the steps outlined in System Administration >> Guide, chapter 'Understanding Root Volumes', section 'Changing Root >> Volume'. If you do not, then after reboot, the appliance will enter >> setup and service will be disrupted until all the setup steps >> (including a reinstall of system software) have been completed. >> Continue with set root operation (y/n)? >>>> >> >> He answered no. We did file listings and verified all the files are >> in both places. There were no obvious ndmpcopy errors. The size of >> space used differs a little bit, but I assume the 32 vs 64 bit >> aggregate layouts would cause some fluff. I even had him destroy the >> 64 bit vol0 volume and run everything again. Same results. >> >> Does anyone know what this is complaining about? >> >> Thanks, >> >> Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

3 7

System Manager for 8.1
by Jeff Cleverley 12 Dec '11

12 Dec '11

Greetings, I've got 8.1RC1 installed on a test filer. FilerView went away with 8.1 7-mode and we now have to use System Manager. I have 2 questions about this: 1. Has anyone found a hack to make FilerView work such as copying over files or directories from an 8.x system? 2. If I only want to set up a centralized configuration that everyone could see, would I have to set up a OnCommand Core package on a server? Right now I have to install the rpm on my linux box and a different setup on my PC. I have to tell it to go discover all of the filers for each installation. This makes supporting this a pain for our support team. Everyone has to install it on their PC and desktops in order to be able to manage the 8.1 filers. It really seems a waste to have to do so many installs, and then tell it to go find all the needed filers. I hope there is a better way :-) I'm only looking for FilerView capabilities to create aggregates/volumes/shares, etc. I'm not needing monitoring or other advanced tools right now. Thanks, Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

11 20

Re: limit i/o for lun or volume
by Alon Zeltser 06 Dec '11

06 Dec '11

I run sysstat perfstat cmpg and other performance tools the problem here is not in the writes but in the reads while the user run his stuff I get 1500 iops most of them read which is kind of awsome for only 5 data disks and 2 parity disks so back to back cp is not the issue here But I will try to set the priority for this volume to low maybe it will help Any other ideas on how to limit io ? Fred Grieco <fredgrieco(a)yahoo.com> wrote: Have you run sysstat on the command line of the filer? (sysstat -x 1) Are you getting a lot of "b" deferred back to back type CPs? My understanding is that the filer starts rejecting new scsi requests when there are back-to-back deferred CPs (because there's no where to put them in memory locally). That's when you get resets and errors on the hosts. Flexshare will limit the IOs to that volume and will reduce the deferred CPs, which may prevent the resets and errors. The host itself may not like that though-- there may be a timeout or some other tuning in iscsi that will help with this. ________________________________ From: Alon Zeltser <alonz(a)emet.co.il> To: Fred Grieco <fredgrieco(a)yahoo.com>; "toasters(a)teaparty.net" <toasters(a)teaparty.net> Sent: Tuesday, December 6, 2011 9:43 AM Subject: Re: limit i/o for lun or volume Thank you for replay but as I wrote im aware of flexshare but I don't think it will help since this system is not impacting other systems and not compeating with other vms on resources Its only hurting itself it doing so much iops and getting such bad latency that after few minutes its doing lun reset and the vm freezes and the user procces exit abnormally I'm trying to limit the iops from this system so it wont load the Netapp so much and get better latency I don't care if it will take much longer to run Thanks again Fred Grieco <fredgrieco(a)yahoo.com> wrote: Do you mean that there is a performance issue with that particular VM, or with other systems on the filer that are caused by I/O on that VM? You should be able to use flex share to set a lower priority for a "cache-hogging" and reduce its impact on the other volumes. On the command line, run "priority on." Then run "priority set volume $volumename level=Low" for the problem volume. This will give less priority to that volume in the write cache, free up the cache for other things, and (hopefully) reduce latency on those other volumes. In your case, you'll need to have that RDM in its own volume. ________________________________ From: Alon Zeltser <alonz(a)emet.co.il> To: toasters(a)teaparty.net Sent: Tuesday, December 6, 2011 6:00 AM Subject: limit i/o for lun or volume Hi all i have a situation of a virtual machine running on esx 4.0 over iscsi this machine (while the user is running his stuff) is using a lot of i/o on a very small netapp controller (2020) with only 5 data disks aggr and getting very bad latency and eventually getting a lun reset errors and the machine freeze for a few minutes i have increased the timeouts from the windows side but it doesn't seems to help this application don't need so much i/o and can be run a 5.4k disk on an old laptop but given the resources of netapp it uses them to the fullest it is important to mention that this application is running over MSSQL db and the i/o problem goes to rdm lun my question: is there a way to limit the i/o of this machine from either netapp side /vmware side / windows side / network side i'm aweare of vmware storage i/o control but this is only supported in vsphere 4.1 and an upgrade is not an options right now i'm also aware of flexshare from the netapp side but i don't think giving low priority to this volume will help in this case is there other way you can think of to limit the i/o goes from vmware virtual machine through iscsi to an rdm lun or his hosting volume? thank you _______________________________________________ Toasters mailing list Toasters(a)teaparty.net http://www.teaparty.net/mailman/listinfo/toasters

1 0

Re: limit i/o for lun or volume
by Fred Grieco 06 Dec '11

06 Dec '11

Have you run sysstat on the command line of the filer? (sysstat -x 1) Are you getting a lot of "b" deferred back to back type CPs? My understanding is that the filer starts rejecting new scsi requests when there are back-to-back deferred CPs (because there's no where to put them in memory locally). That's when you get resets and errors on the hosts. Flexshare will limit the IOs to that volume and will reduce the deferred CPs, which may prevent the resets and errors. The host itself may not like that though-- there may be a timeout or some other tuning in iscsi that will help with this. ________________________________ From: Alon Zeltser <alonz(a)emet.co.il> To: Fred Grieco <fredgrieco(a)yahoo.com>; "toasters(a)teaparty.net" <toasters(a)teaparty.net> Sent: Tuesday, December 6, 2011 9:43 AM Subject: Re: limit i/o for lun or volume Thank you for replay but as I wrote im aware of flexshare but I don't think it will help since this system is not impacting other systems and not compeating with other vms on resources Its only hurting itself it doing so much iops and getting such bad latency that after few minutes its doing lun reset and the vm freezes and the user procces exit abnormally I'm trying to limit the iops from this system so it wont load the Netapp so much and get better latency I don't care if it will take much longer to run Thanks again Fred Grieco <fredgrieco(a)yahoo.com> wrote: Do you mean that there is a performance issue with that particular VM, or with other systems on the filer that are caused by I/O on that VM? You should be able to use flex share to set a lower priority for a "cache-hogging" and reduce its impact on the other volumes. On the command line, run "priority on." Then run "priority set volume $volumename level=Low" for the problem volume. This will give less priority to that volume in the write cache, free up the cache for other things, and (hopefully) reduce latency on those other volumes. In your case, you'll need to have that RDM in its own volume. ________________________________ From: Alon Zeltser <alonz(a)emet.co.il> To: toasters(a)teaparty.net Sent: Tuesday, December 6, 2011 6:00 AM Subject: limit i/o for lun or volume Hi all i have a situation of a virtual machine running on esx 4.0 over iscsi this machine (while the user is running his stuff) is using a lot of i/o on a very small netapp controller (2020) with only 5 data disks aggr and getting very bad latency and eventually getting a lun reset errors and the machine freeze for a few minutes i have increased the timeouts from the windows side but it doesn't seems to help this application don't need so much i/o and can be run a 5.4k disk on an old laptop but given the resources of netapp it uses them to the fullest it is important to mention that this application is running over MSSQL db and the i/o problem goes to rdm lun my question: is there a way to limit the i/o of this machine from either netapp side /vmware side / windows side / network side i'm aweare of vmware storage i/o control but this is only supported in vsphere 4.1 and an upgrade is not an options right now i'm also aware of flexshare from the netapp side but i don't think giving low priority to this volume will help in this case is there other way you can think of to limit the i/o goes from vmware virtual machine through iscsi to an rdm lun or his hosting volume? thank you _______________________________________________ Toasters mailing list Toasters(a)teaparty.net http://www.teaparty.net/mailman/listinfo/toasters

1 0

Re: limit i/o for lun or volume
by Alon Zeltser 06 Dec '11

06 Dec '11

Thank you for replay but as I wrote im aware of flexshare but I don't think it will help since this system is not impacting other systems and not compeating with other vms on resources Its only hurting itself it doing so much iops and getting such bad latency that after few minutes its doing lun reset and the vm freezes and the user procces exit abnormally I'm trying to limit the iops from this system so it wont load the Netapp so much and get better latency I don't care if it will take much longer to run Thanks again Fred Grieco <fredgrieco(a)yahoo.com> wrote: Do you mean that there is a performance issue with that particular VM, or with other systems on the filer that are caused by I/O on that VM? You should be able to use flex share to set a lower priority for a "cache-hogging" and reduce its impact on the other volumes. On the command line, run "priority on." Then run "priority set volume $volumename level=Low" for the problem volume. This will give less priority to that volume in the write cache, free up the cache for other things, and (hopefully) reduce latency on those other volumes. In your case, you'll need to have that RDM in its own volume. ________________________________ From: Alon Zeltser <alonz(a)emet.co.il> To: toasters(a)teaparty.net Sent: Tuesday, December 6, 2011 6:00 AM Subject: limit i/o for lun or volume Hi all i have a situation of a virtual machine running on esx 4.0 over iscsi this machine (while the user is running his stuff) is using a lot of i/o on a very small netapp controller (2020) with only 5 data disks aggr and getting very bad latency and eventually getting a lun reset errors and the machine freeze for a few minutes i have increased the timeouts from the windows side but it doesn't seems to help this application don't need so much i/o and can be run a 5.4k disk on an old laptop but given the resources of netapp it uses them to the fullest it is important to mention that this application is running over MSSQL db and the i/o problem goes to rdm lun my question: is there a way to limit the i/o of this machine from either netapp side /vmware side / windows side / network side i'm aweare of vmware storage i/o control but this is only supported in vsphere 4.1 and an upgrade is not an options right now i'm also aware of flexshare from the netapp side but i don't think giving low priority to this volume will help in this case is there other way you can think of to limit the i/o goes from vmware virtual machine through iscsi to an rdm lun or his hosting volume? thank you _______________________________________________ Toasters mailing list Toasters(a)teaparty.net http://www.teaparty.net/mailman/listinfo/toasters

1 0

limit i/o for lun or volume
by Alon Zeltser 06 Dec '11

06 Dec '11

Hi all i have a situation of a virtual machine running on esx 4.0 over iscsi this machine (while the user is running his stuff) is using a lot of i/o on a very small netapp controller (2020) with only 5 data disks aggr and getting very bad latency and eventually getting a lun reset errors and the machine freeze for a few minutes i have increased the timeouts from the windows side but it doesn't seems to help this application don't need so much i/o and can be run a 5.4k disk on an old laptop but given the resources of netapp it uses them to the fullest it is important to mention that this application is running over MSSQL db and the i/o problem goes to rdm lun my question: is there a way to limit the i/o of this machine from either netapp side /vmware side / windows side / network side i'm aweare of vmware storage i/o control but this is only supported in vsphere 4.1 and an upgrade is not an options right now i'm also aware of flexshare from the netapp side but i don't think giving low priority to this volume will help in this case is there other way you can think of to limit the i/o goes from vmware virtual machine through iscsi to an rdm lun or his hosting volume? thank you

3 3

3170 NFS weirdness
by Randy Rue 01 Dec '11

01 Dec '11

Hello All, We run a VMWare farm that mounts its datastores via NFS on two different NetApp clusters, one a 3020 HA pair running 7.2.4 and the other a 3170 running 7.3.5P5. We also mount those datastore exports to two linux (CentOS 5.x) hosts, which have the CommVault/Simpana backup agents installed. We effectively back up a snapshot of those datastores. We've been testing restores of files back to the datastores on the 3170 and they've been failing. We get a linux Error 13, basically "permission denied," in spite of the fact that the exports are configured for root access to both the ESX hosts and the backup hosts. We've confirmed that the backup hosts are mounting as root. This works on the 3020 pair. Can't spot any meaningful differences in "options NFS" between the two NetApp systems, or in any other setting. Thought it might be a posix permissions issue as a CV restore runs under the local linux group "simpana." The files are written with ownership of root:simpana and then permissions are changed when the data is all there. But if we try to rsync any test data to the 3170 datastores we get the same error, even if we're logged in as root and copying from a local directory on the backup host. Weirder, we CAN write to the datastore folder using cp, rm, mkdir, rmdir, but NOT using rsync. Help?

4 3

8.1 32->64bit volume convert
by Fletcher Cocquyt 30 Nov '11

30 Nov '11

Hi, We're testing out Ontap 8.1RC2 Can someone point out/describe the procedure for converting volumes from 32bit to 64bit? thanks Fletcher

11 23

moving vol0 from one aggregate to another with zero downtime using failover?
by Cotton, Randall 28 Nov '11

28 Nov '11

On an N3400 active/active w/DOT 7.3.5.1P1: I need to move vol0 from aggr0 to aggr1 on node A (so I can nuke its aggr0). I've done this kind of thing on a single independent node - it's fairly well-documented in the "Storage Management Guide". Basically: 1. Create a newvol0 on the destination aggregate of adequate size 2. ndmpcopy /vol/vol0/etc over to newvol0 3. vol options newvol0 root 4. reboot But in this situation, there's stuff on aggr1 I'd like to keep serving during this transition. I'd like it very much if I could use failover to do that. Is it possible to pull this off with a takeover/giveback? It seems it might be problematic what with the takeover node presumably using /vol/vol0 as the root vol and the other node booting up using /vol/newvol0 as the root vol. Would the giveback work OK? It seems there's room here to really hose things up. Any advice? Anyone seen documentation on such a procedure? R

5 4

tracking pBLKs
by Jon Hill 28 Nov '11

28 Nov '11

Is there any way to monitor and alert on pblock utilization via DFM or another monitoring tool? We had an issue recently where CIFS operations ground to a halt that was not resolved until we disabled fpolicy. We believe the entire pblock pool wound up in use and CIFS operations ground to a halt. We're considering running something like this to ensure that CIFS never runs out of pBLKs in the future: setflag fp_maxcifsreqs_pblkpercent X But we want to determine what the current range is before we make any changes. Cifs stat shows the value under Max pBLKs = xxx, but XXX is a number without context, like 1700, and doesn't help much. This e-mail, including any attachment(s), is intended solely for use by the named addressee(s). If you are not the intended recipient, you are not authorized to disclose, copy, distribute or retain this message, without written authorization from Jennison Associates LLC ("Jennison"). This e-mail may contain proprietary, confidential or privileged information. If you have received this message in error, please notify the sender immediately. If you have requested Jennison to e-mail any account information to you or your designee, you are deemed to have consented to electronic delivery of information from Jennison. E-mail messages are not secure and may contain computer viruses or other defects, may not be accurately replicated on other systems, or may be intercepted, deleted or interfered with without the knowledge of the sender or the intended recipient.

2 1

Cifs user name mapping.
by Jeff Cleverley 22 Nov '11

22 Nov '11

Greetings, We have a cifs share that gets worked on by various people in the company in different countries and business groups. We have access opened up to users with a valid Windows 2003 Server domain authentication. Since we don't know everyone who will have access to this, we don't have unix accounts for them, or entries in the usermap.cfg file. This share is also nfs accessed and has unix security style. The problem is when we look in the messages files these users seem to show up with their workstation name followed by the $ sign. It's difficult to track who owns the machine and who logged into it. Here is an trimmed example entry: auth.trace.authenticateUser.loginAccepted:info]: AUTH: Login by dwc46q1$ from xx.xx.xx.xx accepted. I'd like a better way to figure out who logged in from dwc46q1$ . The cifs.trace_login is on. Thanks, Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

2 2

shelf id mismatch
by Chaim Rieger 19 Nov '11

19 Nov '11

I have a shelf that displays ID 4 however when i do show disk they come up as belonging to ID 0. This probably happened because the ID was changed, then powered off, then connected to the loop, then powered on. Just a guess... I am not comfortable with removing a full shelf from a loop (even though all these disks are marked as spares currently) Therefore my question is, if i remove one disk at a time, and reseat it. will it come up as ID 4, or ID 0 ?

3 6

mpha confusion
by Chaim Rieger 18 Nov '11

18 Nov '11

assuming i have two loops (id 00-04, and 10-18) 1 4 port sas card (0a-0d) loop 1 00A connects to 001a 0a 00B connects to 001b 0a ... 04A connects to 001b 0b 04B connects to 001a 0b loop2 10A connects to 001a 0c 10B connects to001b 0c ... 18A connects to 001b 0d 18B connects to 001a 0d KEY = 00A = Shelf ID 00 Module A 001A = Filer 001a/b 0a = SAS port A Its been a long three days of no sleep, but this is mpha, correct ?

6 8

How do I find the driver version for the FC HBA's on my 3170?
by Randy Rue 15 Nov '11

15 Nov '11

Upgrading the attached SAN and the vendor wants to know this...

3 2

Snapshot reserve
by Vincent Liggio 11 Nov '11

11 Nov '11

We tend to have very large file systems (largest NetApp one is > 40tb), and the default 20% snapshot reserve starts to be a huge waste of space. Is there any good reason not to just set it to 0%? The snapshotted files aren't limited by the reserve, so what's the real point of it? A lot of times we end up with reserves that are only a few % in use, so it's truly wasted space. I've been setting those ones to have small reserves, and the discussion ensued whether it's worth having any reserve at all. Thanks. Vince Blue Sky Studios

4 3

Clarify ndmpcopy behavior.
by Jeff Cleverley 04 Nov '11

04 Nov '11

Greetings, I was experimenting with ndmpcopy and got an unexpected result. I was wondering if anyone can clarify if this is correct behavior or not. I've tried it on both 7.3.3P5 and 8.1RC1. I got slightly different error messages, but both failed. I set up a source directory of foo and populated it with some data. I ran the following command which worked: ndmpcopy /vol/perf_test/foo /vol/perf_test/foo2 The part that broke was I deleted the zoneinfo directory from foo2, then ran the same command with the "-l 1" option. I got the following error: Ndmpcopy: easystreet: Log: RESTORE: Invalid operation: ./zoneinfo I was expecting it to put zoneinfo back into the destination. I started over again and used the "-d" option for the level 1 incremental. The log file had no additional information. Should an incremental put this directory back? Thanks, Jeff -- Jeff Cleverley Unix Systems Administrator 4380 Ziegler Road Fort Collins, Colorado 80525 970-288-4611

2 2

from performance vs rg size to new topic: rg size uniformity within the aggregate
by Cotton, Randall 04 Nov '11

04 Nov '11

Unfortunately, I've not been enlightened as to performance numbers for varying small aggregate sizes with one raid group, so I guess I'll have to be content with shooting in the dark by starting out small and growing from there if the performance is too lousy. Regardless, may comments were still helpful and they are much appreciated. But now to a tangentially related subject: >From recent posts, one might be led to believe that it really doesn't matter much at all what the sizes of your raid groups are within an aggregate. One might get the idea that an aggregate with, say, 3 raid groups of size 6, 12 and 18 wouldn't perform much differently than one with 3 identical raid groups of size 12 because what matters far more than anything else is the data disk count of an aggregate: For instance: "...there is no reason one 4+2 and two 16+2s cant perform quite well. There is no unreasonable penalty to doing so ... if the right tools are used to manage it." "Performance of an aggregate is the total sum of spindles across the entire aggregate." And "Forget parity as a limitation to system performance" However, NetApp's recommendations on raid group sizing within the aggregate might lead one to believe somewhat the opposite: >From http://communities.netapp.com/thread/1587 : "the recommended sizing approach is to establish a RAID group size that ... achieves an even RAID group layout (all RAID groups contain the same number of drives). If multiple RAID group sizes achieve an even RAID group layout, NetApp recommends using the higher RAID group size value within the range." and "Drive deficiencies should be distributed across RAID groups so that no single RAID group is deficient more than a single drive." That is, NetApp took the time and trouble to say: 1. RAID group sizes within an aggregate shouldn't vary by more than 1 disk 2. Larger RAID group sizes are better (up to a point) Which runs rather strongly against the grain of what's been said recently in this forum. Can anyone shed more light on this topic? For instance, what are the drawbacks (and how bad are they) to, say, adding a 6-disk raid group to an aggregate consisting of one 16-disk raid group (instead of spilling for way more disk space than you might need by forking over for the full 16 disks necessary to even things out?). Thanks Randall

9 13

mixing disks in aggr
by Joosep Meus 04 Nov '11

04 Nov '11

Hi list! Currently we have 3 DS14mk4 shelves full of 300GB 3,5" 15K FC drives. All drives are in a single aggregate in three raid groups with two spare disks. Time has come to upgrade capacity in the aggregate and i'm thinking of 3 possible ways: [a] buy another DS14mk4 shelf and continue with 300GB drives - safest way, but most expensive per GB [b] buy another DS14mk4 but use 600GB drives - probably good idea, not so expensive as [a] [c] buy DS2246 with 600GB SAS drives - Probably not a good idea, since drives are different type(in this case 2,5" 10K SAS), but it seems that future will be SAS', it's cheaper per GB and takes less space in rack. As i understand, it is possible to use different drive types in the single aggregate(allthough not recommended), but probably will have some impact on performance. Right now my first choice would be [b] followed by [a]. If anyone would share his/her thoughts on the matter I would be most grateful. Thanks, Joosep

4 5

One large ~40Tb 8 x DS4243 aggr too unbalanced for 3270 ?
by Fletcher Cocquyt 03 Nov '11

03 Nov '11

Fellow toasters, we are trading up some old trays for new DS4243 trays and we are trying to decide if we allocate all 8 of our DS4243 trays to one ~40Tb aggregate (we plan to run ONTAP 8.1 for data motion) assigned to one of the cluster's heads will the advantage of all those spindles in one aggr be killed by the head being overloaded? In which case would we be better served by 2 x 4 DS4243 AGGRs - one assigned to each 3270 head? We also have 7 DS14-MK4 shelves to help balance out the workloads All other things equal - I'm inclined to go for the one big 40Tb aggr vs 2 x 20Tb aggrs to reap the biggest IO capacity and performance Currently our CPU weekly average workloads on the two 3270 heads are only 20% and 16% My impression is a single 3270s head should be have enough headroom to push the large aggr to its full IO capacity. But I'd be very interested to hear the advice others who have tested both options and any unforeseen issues with an aggr this large (our previous largest is 16Tb under 7.3.5.1P2) thanks Fletcher.

4 3

unsubscribe
by Bryan West 01 Nov '11

01 Nov '11

unsubscribe

9 9

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