|
Here's the situation. We have an HP MSA2000fc serving as the storage for our database cluster. It was setup for us by an outside party. I don't have much knowledge on how to setup a fibre channel storage network since I've never gotten to screw around with it when it WASN'T hosting our database backend. However, we are going to be setting up a new database cluster soon and the outside party is no longer with us (for reasons I think will become apparent.) So, I'm getting the lay of the land and trying to get up to speed on what I need to know. Some more background. We have two Brocade Silkworm 200e switches with 12 licensed ports each. Each server has two HBAs, the SAN has two controllers with two connections each, and we will eventually have 4 servers connected to this SAN. What we asked for: Redundant switching fabric. What we got: One switch not plugged in at all. A0 and B0 ports on the SAN connected to one switch. One HBA in each server plugged into the switch, the 2nd one not connected to anything. I start digging into it a bit deeper. I fire up HBAnyware on the database server and take a look at the connections. A0 and B0 of the SAN are listed with both presenting drives. I go into server manager (2008 R2 machine) and take a look at the disks. I see all the disks that are being used online and mapped with a drive letter. But then I see them all listed again (only offline.) Hmm. I'm suspecting at this point if I went in an yanked the A0 connection, everything would go offline. So that's when I started trying to learn about the proper setup and this is where I need guidance from what I figured out today. I'm just trying to see if I'm on the right path and to confirm just how hilariously they hosed up this setup: First off, ports A1 and B1 should be connected to the other switch. 2nd HBA in each server should also be connected to the other switch. At that point, we'll have the physical redundancy down. Then, if I fired up HBAnywhere, I should see each set of drives 4 times under each HBA on the server (one for each of the connections the SAN has). After I had that figured out, I started thinking about the duplicate disks showing up in server manager. That just didn't seem right to me so I dug deeper. I'm guessing the issue is that they never installed the Multipath I/O feature in Windows Server 2008 R2. Because of that, the server has no idea that it's seeing multiple paths to each of the disks rather than different disks. So, does that sound about right? 1) Ensure all physical connections are in place 2) Setup multipath I/O so that all those extra connections are utilized properly This would be the absolute first time I've ever touched configuration of a SAN, so I'm just double checking myself. bull3964 fucked around with this message at 06:11 on May 11, 2012 |
#
¿
May 11, 2012 06:08
|
|
|
|
| # ¿ Apr 19, 2024 13:46 |
|
|
Thanks for the feedback everyone. It's nice knowing that you're going down the right path and even nicer to have your suspicions about how incompetently it was setup before confirmed.
|
|
#
¿
May 11, 2012 15:51
|
|
|
Nukelear v.2 posted:
Are you getting the pricing straight from Dell or after some back and forth? I ask because dual controller MD3620i start at $18k on their site and MD1220s are $3500 a piece. That comes out to $25k and not $15k
|
|
#
¿
May 23, 2012 23:25
|
|
|
somecallmetim posted:Anyone here have any experience with Dot Hill? We have a MSA2000 HP SAN which is just a rebranded Dot Hill SAN. It hasn't given us really any trouble in 3 years though we are only using it as the storage for our two node MSSQL cluster, nothing fancy.
|
|
#
¿
Jun 20, 2012 02:36
|
|
|
somecallmetim posted:I just got off the phone with the Rep from Dot Hill along with a rep from EMC (Zones conferenced) and here is what I took away: It sounds like the Dot Hill system you are describing is what HP is reselling as the P2000 G3, so you may want to cross shop there as well. http://h10010.www1.hp.com/wwpc/us/en/sm/WF25a/12169-304616-241493-241493-241493-4118559.html?dnr=1
|
|
#
¿
Jun 21, 2012 03:07
|
|
|
Internet Explorer posted:NTFS will break if more than one host accesses it. NTFS does have some limited cluster support since 2008 R2 via Cluster Shared Volumes. http://en.wikipedia.org/wiki/Cluster_Shared_Volumes
|
|
#
¿
Jul 3, 2012 20:57
|
|
|
I would say firmware upgrades are another place where you can get nailed. I worry less about the hardware and more about the software. The idea that a glitch could wipeout the configuration or that a firmware update could blowup somehow but still proceed to the 2nd controller scares the crap out of me.
|
|
#
¿
Aug 16, 2012 21:48
|
|
|
Just out of curiosity, why is RAID 5 instead of RAID 6? I'm not sure I would be comfortable with single parity in a production setting, even if you did have a hot spare.
|
|
#
¿
Mar 11, 2013 17:23
|
|
|
A lot of people really just don't recognize storage as having more than 4 features; big/small and fast/slow. There's a whole log more engineering that goes into a SAN that people don't recognize because to them storage is connecting up a single drive to a computer's motherboard. They don't see the redundancy baked in, the advanced caching algorithms, the fault prediction, the dedup, and replication. That said, storage manufacturers can get hosed too because their margins are outrageous and you have to do this lovely dance with vendors that would make a car salesman weep to get anywhere near the real price for one of these things. I found this article on Anandtech interesting the other day. http://www.anandtech.com/show/7170/impact-disruptive-technologies-professional-storage-market Things are changing rapidly in the storage market. Insanely cheap compute power and relatively cheap reliable speedy storage in the form of PCIe SSDs threaten to turn the industry on its head in a few years. Most of the energy SAN manufactures have been expending has been to compensate for all the shortcomings of rotational magnetic disks. Once you start introducing an extensive amount of flash and high powered compute resources, you can start making fast and reliable storage systems out of commodity parts for far less.
|
|
#
¿
Aug 20, 2013 20:42
|
|
|
NippleFloss posted:
I think the takeaway from that article is they can get away with it now due to people inherently trusting storage vendors to keep their data safe because they have overcome the shortcomings of the mechanical disk in the past. In 3-7 years, this may not be the case anymore. Basically, we won't need the big engineering input because the increased reliability of the storage will let any vendor churn out high (enough) performance solutions to meet the demands of most customers. At the same time, writing is on the wall for traditional shared storage in quite a few cases. The application technologies are starting to come around to the "shared nothing" way of doing things. Exchange is doing it now with DAGs. Hyper-V 2012 is doing it with shared nothing live migration. Even VMware is dipping their toe in the water in the form of the vSphere Storage Appliance. Yeah, none of these implementations are perfect yet, but the products keep iterating. The only issue in the past was being able to get enough performance out of local storage for this to be feasible (while at the same time being reliable enough), but PCIe SSDs look like they are going to bring us a long way there. There will always be an application for SANs, but things are very much in a state of transition right now. It will be interesting to see how things play out. If nothing else, I will enjoy seeing SAN vendors get knocked down a few pegs and be forced to price their hardware like anyone else. gently caress "call for pricing." bull3964 fucked around with this message at 22:28 on Aug 20, 2013 |
|
#
¿
Aug 20, 2013 22:25
|
|
|
skipdogg posted:This is pretty much IT.txt all the time. There's always new poo poo in the pipeline and it's loving cool. Oh yeah, absolutely. Every once in awhile I sit back and think that the industry I'm in didn't even exist at all when I was in college 12 years ago.
|
|
#
¿
Aug 20, 2013 22:30
|
|
|
Density isn't really going to be a huge issue with flash though long term. Today, you can buy 10TB that fits into a double height PCIe bay. Sure, it's expensive as hell, but this is young technology. Reliability is a concern, yes. However, we can't frame this around traditional reliability concerns that we had for mechanical disks either. It doesn't make sense to use traditional RAID levels or interfaces for devices that do not share anything in common to their predecessors. Yes applications will eventually be written to take advantage of the speed of PCIe SSDs and we will have to find ways around new bottlenecks. This is absolutely the case and even more of a reason why the traditional SAN isn't necessarily the proper fit for these technologies. Why would we want to hamper a PCIe SSD to the latency of iSCSI? I'm not saying any old person can slap together a white box server and have enterprise class storage. I AM saying that new vendors with commodity priced products combined with new approaches to data storage by application developers is going to put a squeeze on the traditional SAN for many implementations. I mean, if you merged something like a FusionIO drive into a Dell VRTX chassis, you've basically killed most of the low end SAN market overnight. At that point you could afford to have two of the things and replicate between them for added redundancy. bull3964 fucked around with this message at 23:51 on Aug 20, 2013 |
|
#
¿
Aug 20, 2013 23:41
|
|
|
The presentation of VRTX I sat through said they were planning on coming out with a redundant controller version sometime in the future. May take on it is do a 4 blade configuration, 3 vmware essentials plus and the 4th is a backup server node running off of internal drives connected into cheap iSCSI or DA storage for the backups. Run the remainder off of internal SD cards and make sure you have the 4hr mission critical support. Worse case, you lose the entire storage of the cluster. You get dell to fix the hardware problem within 4 hours and then blow the VMs back on to the rebuilt storage from the backup server. You won't even have to redo vmware configuration. Obviously, that's not ideal if you can't have any downtime at all, but that's not the target market for these devices anyways.
|
|
#
¿
Sep 25, 2013 23:02
|
|
|
For something like an Equallogic PSx100 series SAN, how exactly would you do different VLANs per path? You have two controllers with 4 ports each. Only 4 are active at once and they do vertical port failover so you would need 4 VLANS. If each vertical port group was on a different VLAN, wouldn't you nee 8 ports per host to connect to all the paths?
|
|
#
¿
Oct 26, 2013 04:24
|
|
|
Misogynist posted:You don't need active/passive network connections to be on separate VLANs -- I don't even see how that would work, since you couldn't have the same IP address between interface pairs. I know, that's why I said 4 VLANs. Each pair of vertical failover ports would be their own VLAN. For example. 0 A/B 10.100.1.x 1 A/B 10.100.2.x 2 A/B 10.100.3.x 3 A/B 10.100.4.x But, if you want to have two switches for redundancy, each pair of a vertical port group would be connected to a different switch (A plugged into one, B plugged into the other). If you wanted all 4 ports to be accessible from each host even during a switch failure, then you would need 8 ports on each host, 4 connected to switch A (one on each VLAN) and 4 on switch B (one on each VLAN). If you had fewer ports on a host, I don't see how the host would be able to see all ports on the storage in the event of a switch failure and you would lose half your storage bandwidth. Dell's best practices for that array has all 4 ports being on a single VLAN and using a single VLAN for all your storage traffic. bull3964 fucked around with this message at 05:35 on Oct 26, 2013 |
|
#
¿
Oct 26, 2013 05:32
|
|
|
No mention at all yet of Seagate's Kinetic Open Storage that was announced over the weekend? http://www.seagate.com/tech-insights/kinetic-vision-how-seagate-new-developer-tools-meets-the-needs-of-cloud-storage-platforms-master-ti/ It seems like an interesting concept, though I'm still trying to wrap my brain fully around it. Ethernet enabled low level storage devices (drives) manipulated directly via an API (PUT, DELETE, GET commands). No controller, no RAID subsystem, no operating system, no SAN.
|
|
#
¿
Oct 28, 2013 15:24
|
|
|
Speaking of backup storage, looks like facebook wants to jumpstart using blu-rays for long term cold storage rather than tape or slow disks. http://arstechnica.com/information-technology/2014/01/why-facebook-thinks-blu-ray-discs-are-perfect-for-the-data-center/
|
|
#
¿
Jan 31, 2014 19:33
|
|
|
Docjowles posted:The guy from Facebook addresses that in the article. He feels that if a bunch of large enterprises start ordering hundreds of thousands of Blu-Rays, the economies of scale would reduce the per-disc cost to basically nothing. Yeah, the cost is high right now because the volume is really low. BD-R production never reached what DVD-R or CD-R did because everyone just bought a flash drive for portable storage and everyone had large enough hard drives that they didn't need to offload files to disk. The big drivers for this are power and the fact that the moving parts are confined to a single reader rather than the things storing the data. As far as shelf life, there's no reason why a quality manufactured disc couldn't have a high shelf life in a controlled environment, most of the crap that's in use for home is just that, crap. BD-R, in general, is supposed to be better than CDR or DVDR right off the bat anyways. That, combined with some sort of cross disc error correction should be good to ensure integrity of the data.
|
|
#
¿
Jan 31, 2014 20:06
|
|
|
Well, that's the beauty of some of the stuff Facebook is doing. They are assuming the risks and hopefully it paves the way for innovation. A big chunk that's driving this is power. You really don't want to be spinning disks up and down or storing them powered off as it affects their lifespan. However, keeping them spun up and ready all that time is a huge power draw. So, there really is a need here for SOME sort of near zero power cold storage with a high reliability and long shelf life. It makes sense to start with off the shelf tech first and refine it.
|
|
#
¿
Jan 31, 2014 20:41
|
|
|
On the subject of VSAN, I wonder if they are going to let people start using SSDs for actual storage rather than just cache. The new intel PCIe SSDs were announced the other day and those seem perfect. Two 400gb DC P3600s in a 3 host cluster would be screaming. That would give you 1.2tb of insane fast storage for around $4500. Not a ton of raw storage to be sure, but if you had a handful of VMs, you could get 3 lower end dell servers like the R420 an put together a fast as poo poo 3U SMB cluster for very little money or complexity. You could probably even get away with DC P3500s if your write load wasn't too high and save another $1200. It actually wouldn't be a bad compliment to a Dell VRTX since it has 8 PCIe bays internally. Setup a VSAN with PCIe SSDs across the blades you have installed (assign two PCIe slots to each) for OS volumes and then use the shared PERC8 mechanical drive backplane for bulk storage.
|
|
#
¿
Jun 5, 2014 04:20
|
|
|
Dilbert As gently caress posted:
What kind of Pure meltdowns have you heard about? We're actually looking at a 420 right now.
|
|
#
¿
Jul 23, 2014 19:49
|
|
|
Dilbert As gently caress posted:Basically a large school system in my area got some Pure storage in for a try and buy, after punching it in and reving it up. All the Pure storage shat itself and fun times were not had by any. Do you have any more details? How long ago, did they identify the bug, total data loss, models, scale of implementation? One of the things that attracting us to them right now is one of our partners (which happens to be driving our data needs right now) currently use them. I mean, I'm sure someone has a bad story about every product that's out there (I remember DSLReports badmouthing equallogic when an array crashed). This is just a pretty sizable purchase for us and if this implementation is anything but a hail mary, there will be blood.
|
|
#
¿
Jul 24, 2014 22:36
|
|
|
Thanks for the feedback. Our primary use case of this initially is for 3tb of MongoDB data. It's super compressible data, but there's no native solution for it right now so inline compression on the storage seems like the way to go. It's a replica set of 3 nodes, so the dedupe is very attractive as well. At least on paper, it seems like we could not only greatly shrink the size of a single datastore but also have a great reduction in total data size due to the nodes deduping. Encryption at rest is also a necessity at this point, and that's handled natively by the array. Assuming everything works out well with the storage, the long term plans are to move our IO intensive virtual machines to the Pure storage and our main SQL 2012 R2 database to the Pure storage as well (likely after adding more shelves) since our MSA2000 G2 SAN is getting up there in years. Adding to the cost for us right now is we have no 10G infrastructure. To this point, our switching has simply been a bunch of HP Procurve 2810 switches with our current Equallogic SANs on their own segregated network on two Dell PowerConnect switches. The goal is to unify storage and network switching for simplification and flexibility while adding 10G capability. So, that's just an added cost on top of everything else. It will be the single largest individual infrastructure purchase we have ever done, so I'm listening to as much as I can.
|
|
#
¿
Jul 25, 2014 03:02
|
|
|
pixaal posted:The Equallogic UI displays nothing that I can find about IOPS If you are using Equallogic, you should REALLY be running SAN headquarters. It gives you all of this sort of information. You can download it from your support portal and it's pretty much a requirement for later firmwares (you'll get a persistent alert in the EQL management console if you don't have the group hooked up to SAN headquarters.)
|
|
#
¿
Jul 31, 2014 00:11
|
|
|
Not a SAN, but Dell released a thing today that has some impact on storage density. R630 server. They added a 1.8" backplane option now. So, their 1U servers can have 24 1.8" hot swap SSD drives now. That's some crazy density. I put together some spitball configs. You could buy a server with 24 cores, 256gb of ram, and 4.8tb (raw) of flash storage for right about $20k. All in 1U.
|
|
#
¿
Sep 11, 2014 16:29
|
|
|
We're demoing a Pure FA-420 right now. We just got our 10g networking up and running last week so we don't have a ton of stuff on it yet, but there's some mix of VMWare and just LUNs carved out to individual machines for various things. On the two ESX volumes I have allocated, I'm seeing a 5.1:1 and 5.2:1 compression ratio. I have a physical SQL server attached and I'm seeing a 4.9:1 compression ratio for that data. Most recent thing we added was a Mongodb node, and that's coming out to be 3.0:1 compression. The compression ratio given does not take into account any deduplication you may be seeing (I think) and also ignores empty space. For example, that MSSQL volume I mentioned has a lot of free space in a few of the DB files. So, the 4.9:1 is only taking into account used space within a file. The empty space in the MDF files doesn't even register. If you compared the size taken on disk in the OS to the amount of storage taken on the array, it would be closer to 12:1. There's about 200gb of data shared via deduplication right now. We plan on putting more stuff on it over the next few weeks to run it through the paces. We don't have any 10g host adapters yet, so we've been limited by 2x1g iSCSI right now as far as performance goes. If you can just drop it right into your environment, it's an easy thing to demo. They'll crate up and take the whole thing back after a 45 day trial if you don't want it.
|
|
#
¿
Oct 28, 2014 20:05
|
|
|
NippleFloss posted:
It's funny because I just had the same thing told to me by a re-seller about Nimble. Not saying it's true one way or another, but it's interesting what you hear from various sources. They deal with Nimble, NetApp, EMC, Dell, and Pure and they are who we bought our Pure array from. We've been really happy with our Pure and are adding another 12tb raw shelf, but we also have data needs beyond what Pure will scale to. Currently, they are being served by Equallogic, but scaling an Equallogic SAN group to 300-400tb sounds like a bad idea even if they can technically scale way beyond that (not to mention that they don't have things like compression or dedup and some of the stuff we'll be storing will compress really well.) Right now, it's looking like NetApp is going to be the front-runner for what we need. We're going to likely need around 1/3rd of a petabyte by the end of 2015 with the ability to scale beyond that depending on demand and it's going to be an extremely mixed workload with a lot of sequential data loading going on as well as a decent chunk of random workload that we won't necessarily want to put on the Pure due to space concerns. Ability to serve up CIFS is a real plus to us too because there's parts of our app that extensively use file stores and right now the dual VMs under a DFS namespace and DFSR cause occasional issues that would be simplified by punting that stuff directly to the storage.
|
|
#
¿
Dec 31, 2014 00:06
|
|
|
I don't know if I would really call it low density. From a raw perspective, perhaps, but I'm a pretty firm believer in their claims about compression and dedup seeing it first hand. We're getting a 5:1 ratio for our databases (not counting empty space within the DB files which would push the ratio even higher.) Our general VMs are around 4.7:1 and we're seeing about 3.6:1 on our Mongodb data. In all, we're getting above the 4:1 ratio they use for their usable stats. The big thing though is density of IOPs. That was one of the perks for us since we currently have limited rack space in our colo before we would have to consider moving to another one of their locations. Another thing is to not discount the niche of hardware covering for inefficiency. A good deal of our software systems could use more engineering to make them less resource intensive. We could probably achieve the performance we desire through software optimization and careful planning of a hybrid array (flash sizing for the working set, DB partitioning between tiered volumes, etc.) That takes time and people and we are in short supply of both right now. So, dropping $200k hardware becomes appealing in the short term as opposed to those other concerns if it basically eliminates one bottleneck completely. I know our company isn't unique in this situation and those are the areas where Pure is going to find a good audience. They are working with partners more though. I don't know if you've seen the FlashStack CI solutions they launched earlier this month. These are forklift installs of Pure, VMWare and Cisco UCS and Nexus for either drop in vSphere or or Horizon View environments along with a single point of contact for support for the whole stack. VDI should be a good fit for their dedup and compression algorithms since so much of each desktop is going to be the same system to system. In the end, it's not a fit for everyone. However, holy hell is it a nice solution where don't have exact models in how you are scaling performance and need something that can absorb all the IO you can throw at it without dedicating 3 racks to spindles. But no, it's really not going to be general storage unless you have data that really compresses and dedups well or you have a small working set.
|
|
#
¿
Dec 31, 2014 02:22
|
|
|
chutwig, thanks for the info, that stuff is good to know.
|
|
#
¿
Dec 31, 2014 16:20
|
|
|
Speaking of OnTap, we just finished our Netapp 2554 install today. 20x 4tb SATA, 4x 400gb SSD, 48x 900gb SAS. There's definitely going to be a bit of a learning curve to this as it's not quite as point and shoot as the Equallogic or Pure I've used so far. We haven't configured the Flash Pool yet on recommendation of the tech we were working with from our vendor. OnTap 8.3 allows for partitioning of the flash pool, so we would rather wait until we upgrade to it and allow both the SAS and SATA aggregates use the flash pool than choose one now.
|
|
#
¿
Mar 6, 2015 01:34
|
|
|
Any tips or tricks on Volume/LUN provisioning with NetApp when the primary consumer of storage is going to be VMWare via iSCSI? With the previous SANs that we've had (Equallogic and Pure) Volume to LUN mapping was 1:1. Obviously this is different on NetApp where you can have multiple LUNs per volume and multiple volumes per SVM. I understand that things like dedup are siloed within a particular volume and if the LUNs in the volume grow beyond the max volume size, the volume will go offline. At the same time, if you don't overprovision somewhere you really aren't taking advantage of the extra space that dedup gives you. It seems like there's two ways to go about doing this. Thick provision the volume and overprovision LUNs on that volume based on the dedup ratio you are usually seeing. So, if you thick provision a 10tb volume and are getting a >50% dedup rate, you are pretty much safe to provision 20tb worth of LUNs on that volume (minus any room you would need for snapshots. If the dedup rate gets suddenly worse and you fill up your LUNs, you have a risk of the volume going offline. Alternatively, you could thin provision both the Volume and the LUNs and size the volumes for the total size of all the LUNs in a volume (plus overhead for snapshots). That way, even if you maxed out all the LUNs, the volume wouldn't go offline. You would then size the Volume based on available disk space in the aggregate with your average dudup ratio taken into account. So, if you wanted 20tb worth of LUNs, you would make the volume ~20tb thin provisioned and assuming a 50% dedup you just need to make sure you had 10tb of space in the aggregate to be safe. That seems like the more dangerous route though since if the LUNs + snapshots managed grow beyond what you assumed for dedup, you could fill the whole aggregate which could cause multiple volumes to go offline.
|
|
#
¿
Mar 6, 2015 16:05
|
|
|
Yeah, right now I have 2.9tb of SQL databases consuming 649gb on disk on an FA420.
|
|
#
¿
Mar 6, 2015 16:23
|
|
|
Yeah, it seems like Pure is expanding their team pretty rapidly. Just months after we got our array, they added personnel local to our area. It is some nice piece of mind knowing they have a tech that lives within spitting distance of our datacenter.
|
|
#
¿
Mar 6, 2015 19:02
|
|
|
NippleFloss posted:
It's certainly a consideration and something I was actually researching today. This is the first NFS capable storage device we've had so far, so it hasn't been an option before now. I understand that NFS handles free space reclamation a bit more gracefully and we would have some more granular options for snapshot recovery if we went the NFS route. We will also have a few select very high IOP VMs on Pure which is iSCSI only, but that shouldn't be much of an issue. While we're on the subject of NFS (and CIFS for that matter) what is generally the best way to handle offline backup of native CIFS and NFS (non-VMware) on something like netapp? Snapshots are great for recovery assuming the filer isn't compromised in some way. However, we only have one NetApp install so far and aren't likely to get another for some time. Not only that, many of the most recent hacks have been about data destruction as much as they've been about theft for profit. So, backup of the data to wholly different storage and potentially to tape still seems like the only option to completely cover your rear end from an absolute loss perspective. I've been thinking of a few different solutions such as copying data to a VM and then backing up the VM itself or copying the data to other storage and shoving it off to tape from there. I'm planning on using our soon to be vacant 80tb Equallogic SATA array as a local data store for the backups while I come up with a new strategy for offsite. We're currently using backup exec which I don't particularly like and aren't really married to. So far, my options range from going with a purely VM based backup solution like Veeam and doing a few tricks to get that to backup some of the physical machine data we have or go with another solution that covers both physical and virtual options. Sadly, nothing cloud is a option as our client contracts forbid it. This has all been a massive reorg and expansion. We started with a Dell 3-2-1 solution with 3 R620s, 2 1gb power connect switches, and 1 Equallogic array (which we later added a 2nd unit on to). Since then, we have upgraded our production switching to Nexus 5672UP, added a Pure FA-420 for very high IOP processing, and are now in the process of replacing our Equallogic storage with Netapp for our main VMWare workhorse due to better availability and needing things like dedup. On top of all that, we're expanding from 3 to 6 VMWare hosts and redoing the NICs on all the old hosts with 10gbe and will be updating VMWare from 5.1 to 6 across our whole environment (so I'm also trying to figure out how vVols figure into all of this.) When all is said and done, we've have a much faster, less complex, and easier to scale infrastructure. But we're also at a crossroads to a lot of decision points for setup and backup/recovery that I'm researching. bull3964 fucked around with this message at 20:19 on Mar 6, 2015 |
|
#
¿
Mar 6, 2015 20:16
|
|
|
Kaddish posted:I'm not a big Netapp guy but isn't BackupExec NDMP capable? Oh, it is. I'm just not too keen on keeping BackupExec around if we move to something else for our VMWare level backups in our prod environment. We're pulling back our SAS Equallogic storage to build a small capacity 3 node VMWare Essentials Plus cluster in our office and I had planned on moving the current production backup exec licensing into our much smaller office environment where I feel it will work a bit better. So, I'm open to other (preferribly better but not horrifically expensive and complex) NDMP capable backup options.
|
|
#
¿
Mar 6, 2015 21:03
|
|
|
toplitzin posted:I didn't see this addressed, but is pretty important. When you go to provision the FlashPool you need to keep in my what the workload will be. Certain workload profiles won't even leverage the FlashPool so allocating it would be a waste. Also make sure your SSD Aggr is raid 4 instead of the default DP, otherwise you'll lose two disks to parity and only have two for data use. Two nodes. One node owns the SAS aggregate and one node owns the SATA aggregate. It will, of course, failover to the other node if necessary, but we split them so as to not cause any performance issues.
|
|
#
¿
Mar 30, 2015 18:18
|
|
|
Bilirubin posted:10 TB RAID 5 file server Run. Run far. That's terrifying unless it's all really tiny disks.
|
|
#
¿
Jun 11, 2015 20:10
|
|
|
So, thin provisioning (especially with virtualization). You have thin provisioning at the VMWare level You have thin provisioning at the storage level, sometimes even twice at the storage level (NetApp thin provisioned LUN inside of a thin provisioned volume.) The only reason to thin provision is to oversubscribe resources. The question is WHERE to you oversubscribe. What's best practice? If you have dedup/compression, does it make sense to even thin provision VMs at the VMWare level anymore? At that point is it better to thick provision them and size the drives accordingly so you don't have to worry about oversubscribing each VMWare volume, instead focusing all your attention on the storage level?
|
|
#
¿
Aug 14, 2015 16:19
|
|
|
If you don't thin provision at the storage level though, how is dedup providing you any benefits at all? If I thick provision a 10tb volume on netapp and put 2 thick provisioned 5tb LUNs on it and then put 50 100% full but identical 100gb VMs on each LUN I'll be wasting a ton of space. I will have reserved 10tb of storage on the SAN, both of my 5tb VMware volumes would be full, each guest would also be full, but I would only be consuming around 100gb (give or take) of real storage.
|
|
#
¿
Aug 14, 2015 22:21
|
|
|
|
| # ¿ Apr 19, 2024 13:46 |
|
|
Internet Explorer posted:NetApp is actually one of the big names I haven't worked with. I don't know how their dedupe works but I'd assume it doesn't need thin provisioning. It would just be deduping empty space? Maybe I'm not explaining myself well enough. If you don't thin provision, you are making a reserved storage commitment on the SAN (doesn't have to be netapp, the principles should be the same regardless.) If you have some dedup or compression ratio, even a small one, the actual raw storage you use is going to be less than what the host sees for the LUN. The size of the LUN is irrelevant. If you get a 2:1 dedup ratio on data you store on the LUN, then you are always going to be using half the raw storage compared the consumption the host is seeing on the LUN. You write 100gb to the LUN, the host sees that 100gb more of the LUN is used, but in reality you only consume 50gb of raw storage. At some point, you'll fill the LUN, but only half of the actual raw storage will be used for that data. But if your LUN isn't thin provisioned, you have that other half of raw storage sitting there reserved for the LUN and no other LUN can use it, but you also can't possibly use it since the host is going to see a larger percentage of consumption than you have on raw disk. So, you essentially have raw capacity that you cannot access anymore. Dedup doesn't need thin provisioning, but it doesn't make sense to dedup a thick provisioned LUN (or volume, or whatever) because you are reserving capacity that will never get used as the amount of raw storage being used will always be less than the consumption that the host sees. bull3964 fucked around with this message at 00:01 on Aug 15, 2015 |
|
#
¿
Aug 14, 2015 23:54
|
|