|
KS posted:Oh jesus christ. Now's the time to tell the story of how Unitrends decided an anime woman with fox ears was a good, professional corporate mascot. They pulled most of it, including a godawful youtube video, but some evidence remains. I think you mis understand why I go to con's, and why I go as dan vs. But wow this has nothing to do with the topic my bad. KS posted:
Three comes off as arrogent in many ways but he is a smart dude under it all. Seriously if he could perfect human relations, he'd be the best seller on the market. Dilbert As FUCK fucked around with this message at 04:55 on Jul 31, 2014 |
#
?
Jul 31, 2014 04:48
|
|
|
|
| # ? May 22, 2024 15:02 |
|
|
Dilbert As gently caress posted:EMC just has more money to throw at poo poo, doesn't mean they are the best at it. Hybrid arrays are the best bang for buck right now. It's not about favorites it's about choosing what is best fit for the customers' needs. Fit the customer needs not what your ego thinks storage should be. Remember 'it's about the customer'. Also, this is Something Awful. You probably shouldn't take it so seriously. It'd probably lead to less manic depressive episodes.
|
|
#
?
Jul 31, 2014 04:55
|
|
|
three posted:For the big vendors, I think EMC is best positioned for the future. I also like Dell's solutions for the more cost-conscious. EMC has placed a bunch of bets (XtremeIO, ScaleIO, VNX2, new VMAX, DSSD, VIPR, Data Domain, Avamar) and has a ton of money and also VMware. They will be fine because at least some of those bets will pay off, and even if they didn't they've got so much money and marketshare right now that they could bleed for 10 years and still be #1. I don't think that's got much to do with strategic vision at a company level though, it's just the benefit of being the 800 pound gorilla in the storage space. Dilbert As gently caress posted:I'd really like you to look at the new storage thread OP I wrote. 10011101 said you should be included, and I trust him a lot(mostly because he didn't make fun of me and wanted to talk the tech poo poo). I looked over it when you posted it a while ago and it seemed alright to me, other than something about NAS not being performance oriented relative to SAN, which I'd quibble with since NFS has been popular in the HPC space for a long time. It seemed good enough to cover the bases for new storage folks so that they could ask meaningful questions in the thread. So here's my thoughts on flash disrupting the storage market: there's this idea that Flash will change everything and that a new breed of storage solutions will take over that can leverage flash to provide you with billions of IOPs, and the old vendors will die off, but that's got it backwards as far as I can see. Flash makes performance easy. Everyone will have billions of IOPs and what's the differentiator then? Management. And the big storage vendors have a lot of time invested in building robust management tools and a lot of money to invest in building more to differentiate themselves from the smaller vendors that are trying to trade on being cheap and fast. Most of the small guys can't win that fight, and good luck guessing which one will come out of it alive and on top. Something fairly innovative, like what Tintri has done, is what I think would be required for a flash startup to stay alive. The problem for Tintri is that VVOLs (if they ever actually make it into a release) will make their offering a lot less unique. None of the other vendors really has a compelling "can't get that anywhere else" feature, so they have to rely on being faster or cheaper, which isn't sustainable. On another note, I do think it's sort of funny that some people here would NEVER recommend NetApp. Like...why? I work for NetApp and I can still see why EMC or Pure or Hitachi or whoever might have an appealing product, particularly for certain customers. I'm genuinely curious about what the perceived gaps are. The LHC at CERN uses NetApp. Lawrence Livermore uses NetApp. Apple uses NetApp. JP Morgan uses NetApp. Credit Suisse uses NetApp. Hell, I think Oracle might still use NetApp internally. It's just so bizarre to me to see someone say "Ugh, I would never recommend you deal with that company that some of the largest and most successful enterprises in the world have found value in partnering with." Anyway, that's my NetApp evangelism speech. It's not a perfect company and they don't have perfect products, but there are some ridiculously smart people working there and we have a lot of customers and many of them seem to like us.
|
|
#
?
Jul 31, 2014 04:58
|
|
|
three posted:It's not about favorites it's about choosing what is best fit for the customers' needs. I'm not going off of that, I am going off the way you have posted. I agree that is why I suggested it. I am not saying you are wrong dude, just realize how you are coming off. You're cool and all but realize how you sound to others, just as I should.
|
|
#
?
Jul 31, 2014 05:01
|
|
|
NippleFloss posted:On another note, I do think it's sort of funny that some people here would NEVER recommend NetApp. Like...why? I work for NetApp and I can still see why EMC or Pure or Hitachi or whoever might have an appealing product, particularly for certain customers. I'm genuinely curious about what the perceived gaps are. You just listed a bunch of customers with a bunch of money. Perhaps the disconnect is that while the systems are great in the petabyte range with professional services and large staffs, they're not cost effective in the 50-500 TB SME space where a bunch of us do business. I've been through the Netapp sales process twice in two years and they were not price competitive in either case. They spec out a pure 10k system with flash cache. When I tell them they're high, they try to add SATA disks instead of going pure 10k. However unlike VNX or Compellent or 3Par, Netapp can't autotier. I don't have the time or the inclination to deal with multiple aggregates and moving VMs between them when everyone else does it for me. Compellent showed me a better way four years ago. Why would I go back? CDOT is a bunch of features I don't need. I'm STILL considering buying Netapp so I can put it on my resume and go work at some of those big players you mentioned (Amazon and Google are on there too!). But it would not be the best choice for my company. KS fucked around with this message at 05:10 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 05:06
|
|
|
Dilbert As gently caress posted:You're cool and all but realize how you sound to others, just as I should.  NippleFloss posted:On another note, I do think it's sort of funny that some people here would NEVER recommend NetApp. Like...why? I work for NetApp and I can still see why EMC or Pure or Hitachi or whoever might have an appealing product, particularly for certain customers. I'm genuinely curious about what the perceived gaps are. The LHC at CERN uses NetApp. Lawrence Livermore uses NetApp. Apple uses NetApp. JP Morgan uses NetApp. Credit Suisse uses NetApp. Hell, I think Oracle might still use NetApp internally. It's just so bizarre to me to see someone say "Ugh, I would never recommend you deal with that company that some of the largest and most successful enterprises in the world have found value in partnering with." Anyway, that's my NetApp evangelism speech. It's not a perfect company and they don't have perfect products, but there are some ridiculously smart people working there and we have a lot of customers and many of them seem to like us. While EMC's products may be half-baked in many instances, they are at least trying to move with the current. NetApp has seemed to have lost the will to innovate (or at least stay with the innovators). What cool new things can NetApp do that they couldn't 3 years ago?
|
|
#
?
Jul 31, 2014 05:07
|
|
|
NippleFloss posted:So here's my thoughts on flash disrupting the storage market: there's this idea that Flash will change everything and that a new breed of storage solutions will take over that can leverage flash to provide you with billions of IOPs, and the old vendors will die off, but that's got it backwards as far as I can see. Flash makes performance easy. Everyone will have billions of IOPs and what's the differentiator then? Management. And the big storage vendors have a lot of time invested in building robust management tools and a lot of money to invest in building more to differentiate themselves from the smaller vendors that are trying to trade on being cheap and fast. Most of the small guys can't win that fight, and good luck guessing which one will come out of it alive and on top. Something fairly innovative, like what Tintri has done, is what I think would be required for a flash startup to stay alive. The problem for Tintri is that VVOLs (if they ever actually make it into a release) will make their offering a lot less unique. None of the other vendors really has a compelling "can't get that anywhere else" feature, so they have to rely on being faster or cheaper, which isn't sustainable. I agree, every time I have lunch with some VCDX or a storage engineer it comes down to "so how can you handle the data", or "IOPS are great but after IO count how do you handle and work the data best fit". A lot of flash people break down when they can't do inline dedupe/compression/or basic L2ARC. Half the battle when we are entering the flash age is IOPS, the other is how you control, secure, and make the most out of what you are given. Just throwing IOPS at someting is stupid, if it's all flash how can you use the data? quote:On another note, I do think it's sort of funny that some people here would NEVER recommend NetApp. Like...why? I work for NetApp and I can still see why EMC or Pure or Hitachi or whoever might have an appealing product, particularly for certain customers. I'm genuinely curious about what the perceived gaps are. The LHC at CERN uses NetApp. Lawrence Livermore uses NetApp. Apple uses NetApp. JP Morgan uses NetApp. Credit Suisse uses NetApp. Hell, I think Oracle might still use NetApp internally. It's just so bizarre to me to see someone say "Ugh, I would never recommend you deal with that company that some of the largest and most successful enterprises in the world have found value in partnering with." Anyway, that's my NetApp evangelism speech. It's not a perfect company and they don't have perfect products, but there are some ridiculously smart people working there and we have a lot of customers and many of them seem to like us. NetApp is good, and I love NFS 4.x; but 5.x does not support it. Hersey and I worked long and hard to do Nexenta acceleration on our NetApp. I look at all disk based controllers the same: How did I provision the storage. What I/O am I expecting? Where does my IO mean GB count? What data am I hosting? I like NFS for VDI and am actively pushing my near 600+ VDI cluster too it, I think it works well given the senario. NFS and iSCSI are neck and neck IMO; it depends completely on what you like if you are weighing NFSv4 vs iSCSI. Come 3 months from now I will love NFS, but right now iSCSI is great. NetApp + Nexenta = 28K+ IOPS, Hersey and I were really proud of the progress we made on that one. Dilbert As FUCK fucked around with this message at 05:13 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 05:09
|
|
|
KS posted:I've been through the Netapp sales process twice in two years and they were not price competitive in either case. They spec out a pure 10k system with flash cache. When I tell them they're high, they try to add SATA disks instead of going pure 10k. However unlike VNX or Compellent or 3Par, Netapp can't autotier. I don't have the time or the inclination to deal with multiple aggregates and moving VMs between them when everyone else does it for me. Compellent showed me a better way four years ago. Why would I go back? Every night, each one of my SQL servers runs a job that puts it into hot backup mode, snapshots the luns for the database, mounts them, runs a DBCC, and then replicates the verified snapshots to my other datacenter. My team gets emailed that it completed successfully (or failed) every night. It takes a windows admin about 3 minutes to set that up, one time when they are configuring the database. When I have to do a DR test, someone takes a small amount of time to create flexclones of these snapshots and mount them up. There are no tapes required, no driving to an alternate location, no extra disks needed. Just a few clicks of the mouse to clone a replicated volume. If someone contacts the helpdesk to restore an email, it takes someone with the proper rights about 5 minutes to restore it. No tape necessary, just run restore wizard and create a PST file. These are all things that add value to the array beyond the raw numbers that tell you usable space and IOPS. They save real dollars when I can put off hiring another person for a year, or avoid 10 hours of overtime while someone restores from tape for a DR test (which perform regularly throughout the year). They give my executive management peace of mind that we can post for someone with NetApp experience and find someone quickly if something happened and we lost our admins in quick succession. There is more to a technology purchase than just the spec sheet. Also, you guys are talking about NetApp being behind, but I think with what they have added with CDOT, they are ahead. I can live migrate my data to new hardware and retire the old without the clients noticing. I can push my reliability beyond 5 nines at this point, and like Nipplefloss said, in five years every schmuck will be able to fill his array with 1TB SSDs and call it good.
|
|
#
?
Jul 31, 2014 05:22
|
|
|
three posted:While EMC's products may be half-baked in many instances, they are at least trying to move with the current. NetApp has seemed to have lost the will to innovate (or at least stay with the innovators). What cool new things can NetApp do that they couldn't 3 years ago? I think the whole CDOT migration ended up really bogging 'em down for years, hopefully they can actually do some niftier poo poo now.
|
|
#
?
Jul 31, 2014 05:31
|
|
|
adorai posted:I have to ask, are you factoring in the added value that NetApp gives you with the tools, it's zero cost snapshots, and it's nifty cloning? NetApp is not the only vendor that has these things, but they add value to the package that not all storage vendors have. I have two NetApp SANs, and I don't need an extra backup product. I was going to break this down further but realized it doesn't matter. Doing all of the above with a 4-year-old Compellent array. I have powershell scripts that take SQL- and Exchange-consistent zero-cost snapshots and mount them to other servers for dev/test and backups. I do async replication with 15-minute consistent checkpoints, and I can spin up the DR site using SRM. Some of these are definitely features that Netapp had first, but a bunch of vendors have caught up and surpassed them, I think. Compellent's not perfect: they're missing a fairly basic feature like compression just like Netapp is missing auto tiering. But they were 40% cheaper this time around, with a much bigger SSD tier. This is definitely where the new players fall down, though. Tintri, for instance, is completely missing SRM support. Many don't have scriptable interfaces or the VSS integration with Exchange and SQL. adorai posted:I can live migrate my data to new hardware and retire the old without the clients noticing. KS fucked around with this message at 05:39 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 05:33
|
|
|
We just did a storage refresh and bought NetApp, coming from IBM it is a breath of fresh air. Looking at storage arrays NetApp has been pretty much ahead of all the others with WAFL and flexibility it gives you. HP 3Par: Software was clunky when compared to Netapp and the integration with something like VMware was still lacking. Monitoring and the like was very rudimentary. The tech behind it is very cool but the whole package felt a bit lacking. IBM: They finally made a good user interface which they ported form the XIV. Storage was alright, but for snapshot backups you need another box and for real HA you would need an SVC on top the V7000. TSM integration with the array was also a weak spot. Did not look at EMC, the guys who did the offer pulled out. It isn't all good though: Snapprotect (commvault) is a beast. And you get some weird stuff like exchange needing ISCSI whereas the rest will run on NFS. Thanks MS. Not too many VMs in a volume because of VM stun issues if you want them consistently backed up. But this is more of a VMware issue than a NetApp one. These are little issues that crop up when doing an implementation but are still important when designing the whole solution.
|
|
#
?
Jul 31, 2014 05:47
|
|
|
KS posted:You just listed a bunch of customers with a bunch of money. Perhaps the disconnect is that while the systems are great in the petabyte range with professional services and large staffs, they're not cost effective in the 50-500 TB SME space where a bunch of us do business. Those are generally problems with channel. How much you like NetApp is often extremely dependent on whether or not you end up with a good partner or a bad partner. You ended up with a bad one. Flashcache isn't really meant to front end 10k disks. It's one or the other. The architecture is very similar to Nimble (in fact, it predates them!) where the flash, either in the form of SSD or PCI, absorbs your random read IO while the write accelerated SATA back end absorbs writes. Whether you like auto-tiering or not seems to be a religious discussion, but I tend to think that cache is a better way to accelerate data access than physically moving the data around. It's just an extension of the same model that has worked for processing for years. CPU registers to L1 to L2 to system memory to disk. Auto-tiering is sort of dying off as hybrid and all flash takes over anyway. We do have problems in the SMB space due largely to pricing, and it's something that's discussed loudly and often by SEs and sales folks internally. I don't know if it's a conscious management decision or just a misreading of the market, but there's a price band or two at the low end that we've sort of abandoned, or are only going to be strong in if we are already the incumbent and they've bought in to the ecosystem. Nimble, or Tintri or Nutanix will beat us there most the time, but then they'll also likely beat EMC because it's just a really super competitive space with a newcomers selling at ridiculously low margins (or at actual losses, based on Nimble's most recent yearly financials). But yes, if you're in that space then there may very well be better options, especially if your reseller is staffed by morons, which isn't unlikely. three posted:While EMC's products may be half-baked in many instances, they are at least trying to move with the current. NetApp has seemed to have lost the will to innovate (or at least stay with the innovators). What cool new things can NetApp do that they couldn't 3 years ago? There's a lot of little things and some big things centered around CDOT reaching feature parity with 7-mode. Being able to do secure multi-tenancy is awesome, particularly for the ability to containerize a subset storage resources within a cluster and delegate management to another group. It will also make chargeback much easier to implement. Some basic QOS allows us to easily fence off sets of data to keep them from crowding out other workloads. Moving volumes non-disruptively between nodes and being able to non-disruptively evacuate and remove both nodes and shelves from a cluster, and add new ones in to the cluster makes storage lifecycle management a lot easier. Huge improvements in multi-threaded performance, meaning big performance improvements on multi-core systems that now balance CPU load evenly; a new space efficient replication engine that maintains compression and deduplication space savings when transferring data, but also allows for different snapshot retention periods at the source and destination; support for SMB2 and SMB3; support for pNFS; Infinivols. There's a lot of things that will make my life easier and help my customers. Some of them are things that other vendors already had and some are things that we had in 7-mode that are just much more useful in CDOT. Some are simple things like huge increases in aggregate sizes and SSD caching limits, or just recently refreshed hardware with large increases in CPU count and system memory. Some are low level things like better predictive readahead algorithms or improved background scanner processing. Whether a random customer would find them exciting I really don't know. But I'm sure there are plenty that would find them useful, which is better than exciting. But I also think that's sort of a loaded question because it implies that what existed everyone started at the same place and if NetApp hasn't innovated as much they've fallen behind, when realistically NetApp was an incredibly innovative company for a long time and they have a lot of IP that is still very useful and worthwhile even if it isn't brand new. The useful question, to me, is what is Netapp missing? What segments of the market are the lacking in? And who *IS* innovating? What's innovative about it?
|
|
#
?
Jul 31, 2014 05:59
|
|
|
KS posted:It can also read "not many of these new trendy vendors support {SRM, SQL snapshotting, Exchange snapshotting, API or scriptability}" which is a pretty tough feature set to give up once you build around it. I'm replacing a Compellent and am limited because of it too. Netapp, 3PAR, and Compellent are waaaaay ahead of most here. See, I feel the exact opposite way. Compellent is a company that was hot poo poo for 12-24 months, then got bought by Dell and now they're dead. Everyone that will ever own a Compellent already bought it. 3PAR is in a similar place with HP. I would much rather stick with an independent company that has the financials to sustain itself over picking a new storage startup that is just looking to get acquired. From what I heard through the grapevine about Nimble, their R&D stalled hard after the IPO happened and the early engineers cashed out. Many engineers have been using the advanced NetApp workflows for years (decades?) and I would rather not guess whether Nimble will ever actually get around to finishing that API they've been promising. NetApp is a lot like Windows, where it's got so many little things baked in now after years of R&D and customer requests. You have to work very hard to find something a NetApp can't do (encryption at rest, advanced replication/snapshot options, etc). e: for the record, I hate auto-tiering and I think it's a bad technology. You under-buy on fast disk in order to save money but get burned by any anomalous usage patterns (I guarantee you have some). It was all the rage, and now nobody even talks about it any more. It's all SSD/Flash-based caching now instead of actual tiering. madsushi fucked around with this message at 06:06 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 06:04
|
|
|
Maneki Neko posted:I think the whole CDOT migration ended up really bogging 'em down for years, hopefully they can actually do some niftier poo poo now. Getting CDOT to feature parity has been a huge burden for the engineering teams which is part of the reason why there has been less nifty stuff. 7-mode ends at 8.2 and from 8.3 and up will be CDOT only with 7-mode going into maintenance mode only, so that should free up some resources to do some of the cooler stuff that I know we've had on on the table for a while. But CDOT is a massive change from the old 7-mode architecture so it's really a huge shift, it's just been such a prolonged rollout that it hasn't really registered as a big new thing. But there's still been some cool work done. 8.2, for instance, is VVOL ready. We're literally just waiting on VMware to turn it on. Mr Shiny Pants posted:Not too many VMs in a volume because of VM stun issues if you want them consistently backed up. But this is more of a VMware issue than a NetApp one. This should not be an issue. Are you using NFS datastores? Do you have VMWare snapshots disabled? KS posted:I was going to break this down further but realized it doesn't matter. Doing all of the above with a 4-year-old Compellent array. I have powershell scripts that take SQL- and Exchange-consistent zero-cost snapshots and mount them to other servers for dev/test and backups. I do async replication with 15-minute consistent checkpoints, and I can spin up the DR site using SRM. Some of these are definitely features that Netapp had first, but a bunch of vendors have caught up and surpassed them, I think. You can do almost anything with any storage if you're willing to put the time into developing and maintaining custom tools. A lot of places don't have the resources to do that, or don't want to trust their admins to develop and maintain them. As you scale up in size these problems become worse and vendor supported tools become more important. The baseline capabilities of most storage arrays these days really aren't that different, the crux is how easy they make it to use them. A good admin can easily close the gap, but a most places aren't guaranteed to always have a good admin around. As far as using VMware for data movement, there are still benefits to doing it at the array level which is why VAAI uses back end copies or cloning whenever possible instead of taking up a bunch of bandwidth on the storage network copying data. It's also another thing that gets worse at scale, as storage vmotioning thousands of VMs can take a very serious amount of time and planning, whereas moving the volume backs the datastore does not. From a NetApp specific perspective, because snapshots live in the volume and do not follow the VM when it is vmotioned, there is a really stark difference between moving the data at the array layer versus the vmware layer. If I just move the VM off of the volume and then blow the volume away and retire those disks/that array then I've lost all of my snapshot backups. This would also affect ZFS based appliances. Not sure which other vendors would also have that issue.
|
|
#
?
Jul 31, 2014 06:22
|
|
|
NippleFloss posted:
Yes. Well that is a bit of a problem, they are not totally consistent when they are not in VMware snapshot mode. It is like yanking the powercord of the VM. It is not a slight at NetApp, all other backup tools have the same problem. The downside is that snapshots are made on the volume level requiring you to make more volumes than you may want.
|
|
#
?
Jul 31, 2014 06:54
|
|
|
Mr Shiny Pants posted:Yes.
|
|
#
?
Jul 31, 2014 12:25
|
|
|
NippleFloss posted:Those are generally problems with channel. How much you like NetApp is often extremely dependent on whether or not you end up with a good partner or a bad partner. You ended up with a bad one. I actually went through the sizing/config process with a Netapp sales engineer and only involved a "VA"R for pricing. I trust my reseller pretty well, but granted maybe this is the time they jacked up the margins on the Netapp to encourage me towards something they want to sell. madsushi posted:See, I feel the exact opposite way. Compellent is a company that was hot poo poo for 12-24 months, then got bought by Dell and now they're dead. Everyone that will ever own a Compellent already bought it. 3PAR is in a similar place with HP. What Tintri does is still tiering in my mind. I mean, they write to flash, then move less-used data in chunks down to 7k. How's that not tiering? Caching, to me, means the data lives on disk but a copy is maintained in flash for read acceleration of repeatedly accessed data. By that measure, I think it's a pretty even split. Hell, even Netapp can do both caching (flash cache) and tiering (flash pools) with SSD -- just not with disk. In terms of disk tiering, yes, you can get a lovely partner that undersizes to win a bid -- but you can also do it right, and when it's done right, there are no drawbacks. But saying 3PAR and Compellent are dead since acquisition is laughable. Compellent sold 8500ish arrays prior to acquisition and another 20000+ since. 3PAR gained more market share than anyone last year.
|
|
#
?
Jul 31, 2014 14:02
|
|
|
KS posted:Oh jesus christ. Now's the time to tell the story of how Unitrends decided an anime woman with fox ears was a good, professional corporate mascot. They pulled most of it, including a godawful youtube video, but some evidence remains. Wait, that's the PHD Virtual people? Guess they're off my list of products I've heard good things about.
|
|
#
?
Jul 31, 2014 14:50
|
|
|
Dilbert As gently caress posted:Ah okay I can get that, iSCSI is good, but doing CIF's and NFS straight off the box does have nice advantages. The last place I worked, anytime I brought up iSCSI it was shot down because "iSCSI is a broken protocol" Even though I still to this day don't understand how, maybe I am missing something. iSCSI raw block storage over the network accessed with raw SCSI commands as fast as fiber channel and supports multipath vs SAS raw block storage over the PCIe bus accessed with raw SCSI commands as fast as fiber channel and supports multipath I dont have any idea what your coworkers were smoking but iSCSI is fine. iSCSI is less complicated and MUCH MUCH easier to tune for high IOPs and high throughput. Also, VAAI exists which probably covers their concerns? Now AoE -- that's something I stare at (  ) and wonder what people are thinking because it is so hard to fit into most infrastructuresedit: cleanup feld fucked around with this message at 15:42 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 15:39
|
|
|
Erwin posted:Wait, that's the PHD Virtual people? Guess they're off my list of products I've heard good things about. Unitrends bought PHD Virtual like 8 months ago or something, just finished releasing their re-branded version of it. I am using it (and have been using it through their beta) and it works great. Our rep at Unitrends couldn't answer my question about why they have a catgirl mascot. feld posted:iSCSI I have been strictly running straight iSCSI for a years now and have no issues. I honestly prefer to throw a VM infront of file shares vs having a storage unit that can share out CIFS.
|
|
#
?
Jul 31, 2014 15:49
|
|
|
KS posted:What Tintri does is still tiering in my mind. I mean, they write to flash, then move less-used data in chunks down to 7k. How's that not tiering? Caching, to me, means the data lives on disk but a copy is maintained in flash for read acceleration of repeatedly accessed data. By that measure, I think it's a pretty even split. Hell, even Netapp can do both caching (flash cache) and tiering (flash pools) with SSD -- just not with disk. What Tintri does is write-back caching. The difference between tiering and caching is a big fuzzy anyway, and some NetApp marketing materials refer to flashcache and flashpool as Virtual Storage Tiering to play on the auto-tiering craze from a few years ago, but it's really just cache. Data writes to cache initially, then destages to SATA, then when it is read from SATA it is inserted back into cache, but the blocks still also live in SATA. This is actually functionally identical to what WAFL does with writes, we just do it with NVRAM, rather than SSD. Writes in WAFL are never written directly to HDD, they are always written to stable flash storage (NVRAM) and acknowledged, then destaged at the next consistency point to whatever aggregate holds them. Flashpool can also be configured as a pure read cache, or as a read and write cache where certain types of volatile write data are written straight to the SSD layer, rather than the HDD backing it. The issue is that tiering only really makes sense for certain architectures. WAFL is write accelerated by design, so moving a bunch of data between tiers just to accelerate writes for some process doesn't make any sense. It won't get any faster because unless the system is falling on it's face write latency is completely divorced from the disk latency on the back end. Which leaves tiering for read data. But sequential reads don't get significantly faster moving from 7.2k to 10k to 15k to SSD. All of those technologies are more than adequate to satisfy sequential reads. So random reads are the only use case where tiering might make sense, but that's exactly the problem that a large flash caching store solves. So the merits of needing to migrate between storage tiers on NetApp are a little suspect due to the design. In a more traditional storage design it might make more sense. Now once you get into being able to set QOS on storage objects (LUNS, volumes, files, whatever) and have the storage make intelligent decisions about where to place that data, alone with the ability to non-disruptively move it, then there's some real value there. I think 3PAR does some of this, and VMware is trying to build that intelligence into the hypervisor layer as well. That's a lot less clunky than some of the older tiering designs and makes sense for any storage architecture. feld posted:iSCSI I'm not sure why you're talking about SAS, as it's not a network storage protocol. Anyway, iSCSI is fine, we use it here, but it's definitely inferior to FC as a storage protocol due to the reliance on ethernet. It's not a massive difference, but all things being equal FC is going to provide more consistent and reliable latency and throughput as well as a more stable transit network. TCP's flow control and reliable transit mechanisms were not designed with ordered delivery and consistent performance in mind. Ordered delivery is a requirement for SCSI, at least within the context of a single path, so TCP retransmits can add significant latency as IOs are queued up waiting for the out of order PDUs to be delivered. TCP window changes can cause unpredictable changes in available throughput and because they are reactive and only endpoint to endpoint you can flood intermediate switches or ports and won't know about it until your traffic slows down, or even comes to a complete halt due to pause frames. FCP is designed to avoid those issues, and FCoE with DCB enabled switches does the same. That's why FCoE exists. Because it fixes the problems inherent in relying on TCP/IP for storage delivery and it utilizes technology, 10G ethernet, that almost everyone already has or will be putting into their data center. FCP is also really really easy, so I don't get where the idea that it's complicated comes from. You can make some really complex topologies and some of the advanced features of FC switches can be a little daunting, but zoning is really easy and straightforward and the protocol is loop free by design, determines best path automatically, and as long as you don't screw up your bridge IDs you just plug a new switch in and it magically has all of the requisite zone information. It's pretty cool. YOLOsubmarine fucked around with this message at 17:42 on Jul 31, 2014 |
|
#
?
Jul 31, 2014 17:40
|
|
|
Mr Shiny Pants posted:Yes. Like Adorai said, don't bother with the VMware consistent snapshot. NTFS is journaled and WAFL is also journaled so your risk of getting a corrupted VMDK filesystem using array level snapshots is exceedingly low. We use NetApp snapshots without VMWare consistency in an environment with about 4000 VMs and have never had and issue with a corrupt VMDK. We had to disable the VMware snapshots to make completing backups feasible, and it was done at the guidance of NetApp support. Anything that needs application consistency should be backed up with something like SnapManager, NetBackup, or some other tool that can ensure that IO is fenced properly at the application level.
|
|
#
?
Aug 1, 2014 19:40
|
|
|
So towards the end of the parallel SCSI days with 300G 10k/15k Ultra320 drives, were any of these drives acceptably quiet? I've worked with some older 10k SCSI drives that were really loud and newer ones that were relatively quiet but nothing quite as new as a 300G Ultra320.
|
|
#
?
Aug 5, 2014 15:43
|
|
|
Shaocaholica posted:So towards the end of the parallel SCSI days with 300G 10k/15k Ultra320 drives, were any of these drives acceptably quiet? I've worked with some older 10k SCSI drives that were really loud and newer ones that were relatively quiet but nothing quite as new as a 300G Ultra320. Every one I worked with sounded like a banshee.
|
|
#
?
Aug 5, 2014 16:42
|
|
|
thebigcow posted:Every one I worked with sounded like a banshee. Thanks. I'll stay away from them, at least not the crazy expensive ones. Anyone care to comment on 10k vs 15k noise?
|
|
#
?
Aug 5, 2014 18:26
|
|
|
Shaocaholica posted:Thanks. I'll stay away from them, at least not the crazy expensive ones. 15k is obviously louder and hotter. But it's 2014. You can get a 250gb SSD which is silent, low power, and will blow the socks off any parallel SCSI drive for $100. If it's not going into a rack where you don't care about noise, don't use SCSI. This is the enterprise storage thread, so I hope it's going into a rack. Why are you concerned about noise?
|
|
#
?
Aug 5, 2014 18:29
|
|
|
evol262 posted:15k is obviously louder and hotter. These are going into 'legacy' workstations I'm playing around with in my home. Not for actual work/production. I just thought I'd ask here since you guys would have the most experience.
|
|
#
?
Aug 5, 2014 18:45
|
|
|
How legacy are these things? You may be better served by the cheapest PCI SATA HBA and modern 7.2k desktop drive. And by may I pretty much mean you will be. edit: If they're *that* legacy you can get better performance out of a couple virtual machines running on the lowest garbage tier modern Intel processor and an SSD. The power savings may pay for most of it depending on how long you leave this stuff running.
|
|
#
?
Aug 5, 2014 18:54
|
|
|
thebigcow posted:How legacy are these things? You may be better served by the cheapest PCI SATA HBA and modern 7.2k desktop drive. And by may I pretty much mean you will be. Well absolute speed isn't the point. I just wanted to gently caress around with parallel SCSI because I never really did when I was in high school.
|
|
#
?
Aug 5, 2014 19:05
|
|
|
Shaocaholica posted:Well absolute speed isn't the point. I just wanted to gently caress around with parallel SCSI because I never really did when I was in high school. You probably also never played with FDDI, UNIX workstations with weird expansion buses, PCI-X, and a lot of other technologies which are dead. Let them stay that way. evol262 fucked around with this message at 19:25 on Aug 5, 2014 |
|
#
?
Aug 5, 2014 19:23
|
|
|
Shaocaholica posted:Well absolute speed isn't the point. I just wanted to gently caress around with parallel SCSI because I never really did when I was in high school. If a job has this as a requirement and it's not for some very specialized setup and operations; run the other way. Fast
|
|
#
?
Aug 5, 2014 19:39
|
|
|
There really isn't anything to gently caress around with though. Its a giant rear end ribbon cable but otherwise....? If you already have everything go nuts I guess, but do not spend any money on SCSI discs.
|
|
#
?
Aug 6, 2014 00:14
|
|
|
thebigcow posted:If you already have everything go nuts I guess, but do not spend any money on SCSI discs.
|
|
#
?
Aug 6, 2014 00:33
|
|
|
thebigcow posted:There really isn't anything to gently caress around with though. Its a giant rear end ribbon cable but otherwise....? Termination, chain length, did you get the correct type (SCA, LVDS...) et c.
|
|
#
?
Aug 6, 2014 10:32
|
|
|
Anyone here use "storage accelerators"? I know nexenta will front end poo poo with ZFS, wondering what else there is out there besides nexenta.
|
|
#
?
Aug 6, 2014 14:49
|
|
|
luminalflux posted:Termination, chain length, did you get the correct type (SCA, LVDS...) et c. I do not miss the days of changing jumpers on SCSI drives to set their ID. God help you if you dropped one of those tiny bastards on some like-colored carpet. You'd never find it and of course you'd have no spares.
|
|
#
?
Aug 6, 2014 15:06
|
|
|
Re: legacy SCSI I'm sure I'll get tired of it soon. As a bonus u just found a box of ultra320 drives I 'borrowed' from my former and now defunct employer. The 147G and 300G 10k drives aren't that noisy especially inside a case. Cables are a mess though but the round sleeved ones available seem to make it less messy.
|
|
#
?
Aug 6, 2014 15:41
|
|
|
While you guys are talking retired Enterprise storage I'd like to mention my personal VM labs run a pair of NetApp MK14's. I picked up three from my friendly recyclable about 3 years ago for about 300$. They are loaded with 300GB 15k's. Kept the third shelf for spare drives but not one has gone out on me. I run about 10~15 VM's off the setup for video server testing at any given time and it's no SSD shelf but for the price I cannot complain. You have to update the drive parity to get them to play nice with non-NetApp appliances, but totally doable. If anyone has any of these shelves lying about and wishes to re-purpose them, I'd be more than happy to ship them my notes. Otherwise feel free to call me a moron for using legacy systems for anything other than a door stop. 
|
|
#
?
Aug 6, 2014 16:48
|
|
|
BlueBlazer posted:While you guys are talking retired Enterprise storage I'd like to mention my personal VM labs run a pair of NetApp MK14's. I picked up three from my friendly recyclable about 3 years ago for about 300$. They are loaded with 300GB 15k's. Kept the third shelf for spare drives but not one has gone out on me. I run about 10~15 VM's off the setup for video server testing at any given time and it's no SSD shelf but for the price I cannot complain. I think there's a difference between getting experience with SAN kit and using it for shared storage versus intentionally spending time on a dead technology. Not that I didn't spend time when I was first getting into the industry playing with all the old, weird UNIX variants I could get my hands on, but "learning" parallel SCSI is about as useful as learning parallel ATA or MFM. 520->512 byte sectors takes forever to do.
|
|
#
?
Aug 6, 2014 17:05
|
|
|
|
| # ? May 22, 2024 15:02 |
|
|
I wouldn't really call it 'learning' so much as loving around with the best storage available for a given legacy platform. People still collect and fix up compact audio cassettes. It's just a vintage thing.
|
|
#
?
Aug 6, 2014 17:15
|
|