|
Anyone have any thoughts on NetApp's new offerings? The FAS6200, but in particular ONTAP 8.0.1. I'm thinking of going to 8 just for the larger aggregates.
|
#
?
Nov 11, 2010 15:58
|
|
|
|
| # ? Apr 23, 2024 22:23 |
|
|
i do know i would like to have the inline compression on my boxes. looks sweet in the propaganda.
|
|
#
?
Nov 11, 2010 22:13
|
|
|
Erwin posted:Anybody have experience with Promise VessRAID units? I know it's super low-end, but it's just going to be our backup-to-disk target/old files archive. My concern with the unit is that it's been synchronizing the 6TB array for almost 23 hours now, and it's at 79%. If I add another array down the road and it takes a day to synchronize, the existing array better be usable. That's what I was shoehorning just yesterday! We had to synch 8 1 TB disks. That doesn't sound unreasonable given my experience. So far the Promise isn't bad. Pretty quick, easy to set up, easy to get talking and easy to administer. We're using an LSI card too, those support guys are awesome. Good luck fellow cheap array goon. SmellsOfFriendship fucked around with this message at 22:21 on Nov 11, 2010 |
|
#
?
Nov 11, 2010 22:16
|
|
|
Does anyone have any experience with NexSan? We are looking at their SATABoy product for our hosting our initial round of VDIs. Our users only really use Word, Excel, and a couple of web apps and nearly bottom out when it comes to perfmon across the board. We wanted something inexpensive, but decent. It is looking to go between NexSan or a MDI3200.
|
|
#
?
Nov 15, 2010 18:49
|
|
|
egoslicer posted:Does anyone have any experience with NexSan? We are looking at their SATABoy product for our hosting our initial round of VDIs. Our users only really use Word, Excel, and a couple of web apps and nearly bottom out when it comes to perfmon across the board. We wanted something inexpensive, but decent. It is looking to go between NexSan or a MDI3200. I'm using 2 of their SASBoys w/dual controllers over FC (they are pretty much the same as the SATABoy). The SASBoys are the core of our ESX cluster and are rock solid. I think I had to use support once and they were very responsive.
|
|
#
?
Nov 15, 2010 19:04
|
|
|
Any Compellent users? Opinions? edit: it's for a lab environment, our dev's want complete control over drive assignment, something our LeftHand P4300 boxes don't let them do. skipdogg fucked around with this message at 19:28 on Nov 15, 2010 |
|
#
?
Nov 15, 2010 19:25
|
|
|
skipdogg posted:Any Compellent users? Opinions? The software still has some kinks in 'bizarre' failure scenarios (like if a controller loses both connections to a loop because they were both on a failing ASIC on your FC card, it does NOT fail over to the other controller) but overall it works as advertised.
|
|
#
?
Nov 15, 2010 19:56
|
|
|
skipdogg posted:Any Compellent users? Opinions?
|
|
#
?
Nov 15, 2010 23:50
|
|
|
adorai posted:The sales engineer i spoke to about a year ago could not have possibly more condescending. My VAR just complained about not being able to get ahold of his rep today. He bitched all the way up to the regional manager. I'm sure he ruffled some feathers in the process.
|
|
#
?
Nov 16, 2010 00:26
|
|
|
Does anyone know how to make a RAID 10 in Freenas? I can't seem to find it anywhere, all the guides point me to raid 5 or 1 or 0 but not 1+0? Am I missing something or does freenas not do this. Yes this is software emulation
|
|
#
?
Nov 16, 2010 05:25
|
|
|
Corvettefisher posted:Does anyone know how to make a RAID 10 in Freenas? I can't seem to find it anywhere, all the guides point me to raid 5 or 1 or 0 but not 1+0? Am I missing something or does freenas not do this. Yes this is software emulation Maybe it's like some really terrible LSI firmware versions which are out there. Do you have to make a bunch of raid 1's, then start over and make a raid 0 out of all of your mirrors?
|
|
#
?
Nov 16, 2010 06:01
|
|
|
Jadus posted:Somewhat related, we're looking at an MD3220i with two hosts, and since we don't expect to grow beyond two hosts for a while, we're thinking about cutting out redundant switches between the servers and the MD3220i. Sorry for the slightly delayed response, haven't been checking in much the last week. Yeah, if you've only got 2 systems, then there's no reason not to direct connect, you've got 4 ports on each controller, you can use all of em, and with enough connected it will load balance fine as long as you've got MPIO sorted properly. If you've got 2 ports available on server, connect one to each controller, this will give some load balancing, but as each virtual disk is bound to a controller you'll only get 1GB through put for each VD, but you'll get more spread across all the VD's Best case is 4 ports on each server, with 2 connected to each controller, that will give full load balancing and a nice perf boost.
|
|
#
?
Nov 16, 2010 10:37
|
|
|
Semi-related to Enterprise Storage (and I don't want to start a new topic): For a file server, what is the best way to set up the storage? We plan on virtualizing the host, and connecting it directly to an iSCSI SAN (through the iSCSI initiator, instead of creating VMDKs for the drives). We have a 10TB unit dedicated to this, but we're currently only using 2.75TB of space on our existing file server. Should we immediately allocate a giant 10TB partition or create a more reasonable size (5TB) and then add in more partitions/volumes if needed? What is the best way to add in volumes after-the-fact? Add the volumes then symlink them in? Add volumes and move shares to them?
|
|
#
?
Nov 16, 2010 21:06
|
|
|
three posted:We plan on virtualizing the host, and connecting it directly to an iSCSI SAN (through the iSCSI initiator, instead of creating VMDKs for the drives).
|
|
#
?
Nov 16, 2010 21:14
|
|
|
Misogynist posted:Is there any particular reason you're avoiding VMDKs? Storage vMotion would basically eliminate all of your problems here. From my understanding, the largest LUN size that can be presented to ESX is 2TB. I believe there are ways to get around this by having multiple 2TB LUNs and using extents, but I was under the impression extents were bad. Also, if it's presented just as a straight LUN, we could map it to a physical server if need be.
|
|
#
?
Nov 16, 2010 21:21
|
|
|
three posted:From my understanding, the largest LUN size that can be presented to ESX is 2TB. I believe there are ways to get around this by having multiple 2TB LUNs and using extents, but I was under the impression extents were bad.
|
|
#
?
Nov 16, 2010 21:36
|
|
|
I would guest side iscsi and allocate 30% more space than you are currently using. Then I would resize volume (hopefully your storage supports this) as needed
|
|
#
?
Nov 16, 2010 22:03
|
|
|
adorai posted:I would guest side iscsi and allocate 30% more space than you are currently using. Then I would resize volume (hopefully your storage supports this) as needed I feel retarded for not thinking of this. I overthink things way too much. 
|
|
#
?
Nov 16, 2010 22:20
|
|
|
H110Hawk posted:Maybe it's like some really terrible LSI firmware versions which are out there. Do you have to make a bunch of raid 1's, then start over and make a raid 0 out of all of your mirrors? I thought I would make 2 raid 1 arrays, then go to raid 0 and add both arrays but to no avail
|
|
#
?
Nov 16, 2010 22:41
|
|
|
Corvettefisher posted:I thought I would make 2 raid 1 arrays, then go to raid 0 and add both arrays but to no avail Could it be like some of the HP onboards where you just select RAID1, then select more than 2 disks, and it automagically changes itself to 10/1+0?
|
|
#
?
Nov 17, 2010 05:35
|
|
|
are there many emc users here? we just had a nightmare day at the office - emc engineer was in to install 3 new dae's to our cx3-40. first tray went in with no problems but when the second was plugged in, the bus faulted and took the entire array down. the "non-disruptive" upgrade pretty much brought down our entire call centre after the emc engineering lab did all their research it looks like a dodgy lcc in the dae was the cause. has anyone else seen anything like this happen before?
|
|
#
?
Nov 18, 2010 11:24
|
|
|
It's the tradeoff you make for running midrange hardware. You often get fine performance but the uptime is not guaranteed to a gazzilion nines. But all is not dandy in the enterprise world either. Today we got word from our HP support contact that they are holding us hostage for $700 before giving us any future support. Support under a contract we pay $100k per year for. gently caress HP in the rear end, i will suck a bag of soggy dicks before i buy a single new HP storage product.
|
|
#
?
Nov 18, 2010 13:25
|
|
|
GrandMaster posted:are there many emc users here? yes. Once a few years ago saw a bad lcc in a clariion when adding DAEs. Don't think it took down the whole array though.
|
|
#
?
Nov 18, 2010 15:42
|
|
|
Is there any way to increase the write performance on a NetApp FAS3140 running OnTap 7.2.6.1? It appears that our options, according to NetApp support, are: 1) Increase the number of spindles (we've got 4 fully populated ds14mk4 disk shelves attached to this head with a mix of 150 and 450 GB 15K FCAL drives) so that the NVRAM is able to flush the writes to disk faster without hitting back-to-back consistency points. This may be difficult as we might run into power issues with adding another shelf. 2) Setup a proper reallocate schedule for all volumes to ensure that the volumes aren't "fragmented" and that we're not running into a hot disk scenario. We've tried this and although it appears to help somewhat, there are several times when we still see latency rise due to the back-to-back consistency points. 3) Stop pushing so much drat data to the filers so drat quickly - this might not be achievable, as it's an Oracle database that the DBA's insist must be able to handle this type of load. 4) Buy a bigger filer head that has a larger NVRAM size to help mitigate the instances when it is pushing alot of data to the filer. Are there any other bits of performance tuning that can be done? Have there been any significant changes in OnTap 7.3 as far as performance tuning is concerned? We're looking specifically for write performance, so I'm not sure if a PAM module would help us out (I had believed that they were meant for reads more than writes). We had recommended they go for a FAS3170 when it was specced out a couple of years ago, but they saw the cost and backed off. Thanks!
|
|
#
?
Nov 18, 2010 15:49
|
|
|
What does your IOPS profile look like, read vs. write? How about just pure total IOPS? NFS, ISCSI or block? At peak load what does cache age look like? We have a FAS3140 with 7 full DS14MK4s and 2 full DS4243s emitting block storage to a vSphere 4.1 installation and I have looked at a lot of performance numbers. We are running 7.3.3.
|
|
#
?
Nov 18, 2010 16:37
|
|
|
ferrit posted:Is there any way to increase the write performance on a NetApp FAS3140 running OnTap 7.2.6.1? It appears that our options, according to NetApp support, are: I don't know the model number translation from IBM to NetApp but this doc has a pretty good explaination about the PAM module for you to decide yourself. http://www.redbooks.ibm.com/abstracts/sg247129.html The IBM redbooks are a really good source of netapp info.
|
|
#
?
Nov 18, 2010 16:37
|
|
|
ferrit posted:2) Setup a proper reallocate schedule for all volumes to ensure that the volumes aren't "fragmented" and that we're not running into a hot disk scenario. We've tried this and although it appears to help somewhat, there are several times when we still see latency rise due to the back-to-back consistency points. I'm a bit rusty on this, but you can see if you have a hot disk yourself to see exactly how much performance you're gaining from reallocation. I haven't sat at a NetApp console in a year+ so verify these commands before running them. During your worst IO performance times, where you are getting back to back CP's do: # priv set advanced # statit -b (wait several minutes) # statit -e # priv set This should give you a whole pile of output. Look through the disk utilization numbers and see how you are doing. Netapp still does consolidated parity, right? This means 1 or 2 of your disks per raid group will show some piddling amount of utilization and that is normal. Also look through sysconfig -r, wafl scan status, and options to make sure you aren't doing some kind of constant scrubbing or other high impact job during peak hours. Any scrub jobs should be paused during times of extremely high utilization. Sometimes you can just slap a bigger NVRAM card into your existing netapp. This might get into warranty voiding territory. I've never used Oracle, but make sure you are doing aligned writes to your netapp. Ask Netapp support how you can verify that. One block write should net one data block write (and one or two parity blocks) on your netapp. At a certain point management is going to have to bite the bullet on extra disks. Depending on how risk adverse you are you can spend as much or as little as you want. PM me for details. This is also a pretty good source of diagnostic commands. Temporarily bypass the ssl warnings. Do not under any circumstances run a command which you do not understand. Do not argue with the filer. The filer will win all arguments. The filer knows best. https://wiki.smallroom.net/doku.php?id=storage:netapp:general H110Hawk fucked around with this message at 16:58 on Nov 18, 2010 |
|
#
?
Nov 18, 2010 16:55
|
|
|
Thanks to all of you for the responses - I really appreciate it.complex posted:What does your IOPS profile look like, read vs. write? How about just pure total IOPS? NFS, ISCSI or block? At peak load what does cache age look like? I should have mentioned this in my first post but we're presenting every volume as NFS. As I only really have a very amateurish knowledge of NetApp performance (by using 'sysstat -u' and 'statit' mainly), I only really know how to get the total number of operations. For example (a snippet of stuff from running 'sysstat -u 1'): CPU Total Net kB/s Disk kB/s Tape kB/s Cache Cache CP CP Disk ops/s in out read write read write age hit time ty util 48% 2998 57149 20355 107524 72588 0 0 35s 93% 99% M 70% 47% 3214 59451 20107 105572 67364 0 0 33s 94% 100% : 76% 33% 1417 9045 21801 91588 51480 0 0 33s 97% 100% # 60% 27% 1077 318 20653 100956 66712 0 0 33s 95% 100% # 70% 47% 3504 59658 20528 65120 48220 0 0 33s 97% 100% b 85% 28% 1815 14856 28860 66544 50176 0 0 33s 96% 100% : 70% 26% 1376 3569 24536 69968 54756 0 0 33s 95% 100% : 68% 28% 2098 11302 23890 71072 52128 0 0 34s 96% 80% : 73% So just in that small period where I see the back to back CP, we're seeing anywhere from 1000 ops to 3500 ops. To me, the disk utilisation figures look quite high, though obviously they are lower (around 20-40%) when there is not a CP occurring at that moment. Our big problem is that the incidents of back-to-back CP's are really peaky - we might hit a 5 or 6 second period of high latency and then not see it again for hours. My statit's are definitely showing that the last rg (the shelf with the 450 GB disks) have a higher utilisation figure than the other 3 rg's with the 150 GB disks - but could that just be because they are bigger disks and thus utilised more? I thought that the volume reallocate would have assisted us in levelling this out. ghostinmyshell posted:I don't know the model number translation from IBM to NetApp but this doc has a pretty good explaination about the PAM module for you to decide yourself. I'll have to check this out - thanks for the link. I believe there is some way that you can actually simulate what a PAM module would do for you if you had it by running some advanced commands, but I can't find it in the NOW site right now. H110Hawk posted:Also look through sysconfig -r, wafl scan status, and options to make sure you aren't doing some kind of constant scrubbing or other high impact job during peak hours. Any scrub jobs should be paused during times of extremely high utilization. We have the automatic WAFL scrub scheduled for Sunday Mornings, so we tend to discount those stats if we see the high latency during that period, but that's a good point. We also regularly check 'wafl scan status' to be sure that the customer has not decided to delete a few TB of data and thus cause a big block reclamation scan to take place (although we have that one option 'wafl.trunc.throttle.hipri.enable' set to 'off' so it shouldn't take the filer to its knees if they do decide to do it). Do NetApp even sell NVRAM upgrades anymore? I had a quick look around the site but couldn't find anything - otherwise, this would probably be ideal for us, unless it were to invalidate our support contract. And that link looks good too - though the java commands, frankly, look frightening. Another quick question to throw in the mix, not necessarily related to this issue - aggregate snapshots. Do you have them enabled or not? Is it worth having them? I've spoken to several people who've worked with NetApp for years and they've never had to use an aggregate snapshot to recover. Our biggest issue is the amount of reserve it takes up - the customer is angry about that space that should be his (his words, not mine). Also, if he ends up deleting an arseload of data and it's bigger than the aggregate snap reserve, then we have seen the filer go loopy trying to clear that aggregate snapshot (we have aggregate snapshot autodelete turned on) - and several times it has gone over 100% snap usage and eaten into the normal aggregate.
|
|
#
?
Nov 18, 2010 18:39
|
|
|
ferrit posted:I'll have to check this out - thanks for the link. I believe there is some way that you can actually simulate what a PAM module would do for you if you had it by running some advanced commands, but I can't find it in the NOW site right now. We've done this. PCS. Because we are dealing with very large files in VMware, Predictive Cache Statistics indicated that an increase in FlexCache/PAM would not significantly increase cache hit ratios for us, and this not be worth it. Instead we decided to simply increase spindles.
|
|
#
?
Nov 18, 2010 20:01
|
|
|
ferrit posted:Are there any other bits of performance tuning that can be done? Have there been any significant changes in OnTap 7.3 as far as performance tuning is concerned? We're looking specifically for write performance, so I'm not sure if a PAM module would help us out (I had believed that they were meant for reads more than writes). We had recommended they go for a FAS3170 when it was specced out a couple of years ago, but they saw the cost and backed off. Same basic question I always ask, are your SQL data files and transaction logs on the same spindles? If they share spindles with an active db then you will end up killing your log performance. After the last SQL thread I decided to test this best practice to satisfy my curiosity. HP P2000 G3 6Gig SAS DA. 2 Minute runs over a 22 Gig file. With a few 64k rand read going on to simulate data file activity. Logs separated to own 4xDisk 15k RAID10, seq write 8k blocks: 111596 IOs/s 90.59 MBs/s Logs mixed with Data on a 16xDisk 15k RAID10, seq write 8k blocks: 3297 IOs/s 25.75 MBs/s Edit: Not sure what block size oracle uses for it's logs, but 64k was similarly affected. 412MB/s vs 149.8 MB/s Nukelear v.2 fucked around with this message at 20:56 on Nov 18, 2010 |
|
#
?
Nov 18, 2010 20:54
|
|
|
Check your alignment priv set diag; stats show lun; priv set admin Look for any writes that are on .1 through .7, they should all be on .0 or partial if you are aligned.
|
|
#
?
Nov 18, 2010 21:47
|
|
|
paperchaseguy posted:yes. Once a few years ago saw a bad lcc in a clariion when adding DAEs. Don't think it took down the whole array though. just heard back from support, they will be replacing the cabling on the SPA side bus 0 as there were some other strange bus errors. it looks like SPA crashed, SPB didnt so i'm not sure why the luns didnt all trespass and stay online
|
|
#
?
Nov 19, 2010 05:01
|
|
|
If im not mistaken the 3140 box has 4 gigs of ram. Is there a way to check the amount of memory involved in the operations stored in nvram? Ie, adding more nvram might be useless if the buffers are full anyway?
|
|
#
?
Nov 19, 2010 11:47
|
|
|
Forgive my ignorance on this topic, but could someone point me to an explanation of how NFS compares to direct block access in terms of performance? How is an Oracle server using NFS for it's data storage? Thanks, found http://media.netapp.com/documents/tr-3496.pdf which is interesting reading, if a few years old. \/ \/ \/ Mausi fucked around with this message at 12:47 on Nov 19, 2010 |
|
#
?
Nov 19, 2010 12:11
|
|
|
Mausi posted:Forgive my ignorance on this topic, but could someone point me to an explanation of how NFS compares to direct block access in terms of performance? Netapp has alot of whitepapers online about oracle and nfs.
|
|
#
?
Nov 19, 2010 12:25
|
|
|
I'm on the verge of signing the order for a HP x9720 in a 164TB configuration with three accelerators for online media archiving and small-time online viewing. Anyone have an informed opinion on those? I haven't been able to find any independent comments since the system is still pretty new, and I'd like to know if I'm buying a piece of crap.
|
|
#
?
Nov 19, 2010 14:16
|
|
|
Crowley posted:I'm on the verge of signing the order for a HP x9720 in a 164TB configuration with three accelerators for online media archiving and small-time online viewing.
|
|
#
?
Nov 19, 2010 14:20
|
|
|
As an HP vendor I'd be interested to know your reasoning behind that.
|
|
#
?
Nov 19, 2010 21:56
|
|
|
I would too. I've been using EVAs for the better part of a decade without any issue at all.
|
|
#
?
Nov 19, 2010 22:25
|
|
|
|
| # ? Apr 23, 2024 22:23 |
|
|
ferrit posted:My statit's are definitely showing that the last rg (the shelf with the 450 GB disks) have a higher utilisation figure than the other 3 rg's with the 150 GB disks - but could that just be because they are bigger disks and thus utilised more? I thought that the volume reallocate would have assisted us in levelling this out. The 450GB disks are going to run hot because new blocks are likely to wind up there as you reach capacity on the 150gb disks. You have roughly the same IOPS performance, I assume, between the 150gb and the 450gb disks. This means you are trying to pull more blocks from the same number of IOPS and creating a bottleneck. If your VAR did not explain this they should be raked over the coals. I have no idea if they ever sold NVRAM upgrades. I do know that if you slapped a bigger NVRAM card into an older box it would use it.  Snapshots have their place. If you have no use for them then just disable them. Snap reserve is a magical thing sometimes, as it can let you squeeze some "whoops!" space out of the device much like ext2's root reserved space. One common thing you can do with them is tie them to your frontend software, issue a coalesce/get ready for backup command, fire a snapshot, then release the lock. This then lets you do a copy of the snapshot somewhere else for backup. If this is not a part of your backup system then don't worry about it. I personally love snapshots.
|
|
#
?
Nov 19, 2010 23:48
|
|