ataylor Posted January 30, 2019 Posted January 30, 2019 I'm currently looking at upgrading our storage to SSDs, we store around 4TB of data which i'm sure I can reduce to around 3TB-3.5TB. Our two Hyper-V hosts are PowerEdge T620's with 2xE5-2650's, 64GB of RAM and PERC H710 Raid adapter. They hold 8 bays in total in capacity, the servers are currently running on 3x2TB Raid 5 & 3x4TB Raid 5 with two empty bays. I'm currently looking at dropping the 3x2TB drives, keeping the 3x4TB (for additional storage such as multimedia & for emergency storage in the event a host goes down) then install 4x1TB Enterprise SSD's in Raid10 - Doing this I can move the majority of our data and all of our services onto SSD storage. The only thing is that I'm not really sure what best practice is regarding SSD's and Raids - What have you guys have configured?
localzuk Posted January 30, 2019 Posted January 30, 2019 We run SSD only storage, and have done for a couple of years. We have it all in RAID 6 groups. 2 x 20TB RAID groups. 1
EddyAllen Posted January 31, 2019 Posted January 31, 2019 All SSD storage here as well. Might be a silly question but do you have Data Deduplication enabled on your VHDX libraries, the savings are absolutely crazy on REFS with Server 2019. I would personally definitely make the move, the performance benefits if you have a large SIMS database or even for file shares if you have lots of concurrent users are very much worth it. 1
HPlum78 Posted January 31, 2019 Posted January 31, 2019 (edited) Hmmm, so it depends on the workloads... for Virtualized work loads you should be looking at read intensive SSD's, for DB workloads you should be looking at write intensive SSD's. That said in your environment i would guess that you are not going to gain the benefits from the different types of SSD's v the additional cost of the type of drive, I add this for completeness as you should be looking at your environment and be armed with all the info. On a side note from memory i thought MS said using dedup on Hyper V where VHDX files are stored is only supported for a very certain type of VDI deployment/ environment. My initial thought about this for your general purpose live environment would be not to do it, even more so if tyou are just going to buy general SSD's off the shelf. Just reread what @EddyAllen has written he is saying on the libraries not on the actual Hypervisor drive used for the VMs (I hope). Edited January 31, 2019 by HPlum78
EddyAllen Posted January 31, 2019 Posted January 31, 2019 Hmmm, so it depends on the workloads... for Virtualized work loads you should be looking at read intensive SSD's, for DB workloads you should be looking at write intensive SSD's. That said in your environment i would guess that you are not going to gain the benefits from the different types of SSD's v the additional cost of the type of drive, I add this for completeness as you should be looking at your environment and be armed with all the info. On a side note from memory i thought MS said using dedup on Hyper V where VHDX files are stored is only supported for a very certain type of VDI deployment/ environment. My initial thought about this for your general purpose live environment would be not to do it, even more so if tyou are just going to buy general SSD's off the shelf. Just reread what @EddyAllen has written he is saying on the libraries not on the actual Hypervisor drive used for the VMs (I hope). Well I thought so too until a seminar recently, it appears as if Microsofts stance has changed on this and now they just suggest you DeDupe the libraries which include storage and the operating system for the VM's. The only case I would be not doing this would be for an SQL DB and logs drive. In which case they would just go in a subfolder that has an exclusion rule set on it. I believe this was also echoed in the 2016 MCSA exam. 2
PotNoodleTech Posted January 31, 2019 Posted January 31, 2019 (edited) All SSD storage here too running 6 x drives in each server in a Raid5 group. People will laugh at my raid5 but I wouldn't change them. I went with it as it still has a very good mixture of performance, storage capacity, redundancy and actually with SSDs what you really want to do to elevate their life span is reduce unnecessary writes on each disk which is the opposite of what Raid 10/01/1/6 does - overly write heavy thus your actually degrading your SSDs reliability/lifespan just to have extra parity, which doesn't make any sense to me! Edited January 31, 2019 by PotNoodleTech 1
Mako Posted January 31, 2019 Posted January 31, 2019 All SSD storage here too running 6 x drives in each server in a Raid5 group. People will laugh at my raid5 but I wouldn't change them. I went with it as it still has a very good mixture of performance, storage capacity, redundancy and actually with SSDs what you really want to do to elevate their life span is reduce unnecessary writes on each disk which is the opposite of what Raid 10/01/1/6 does - overly write heavy thus your actually degrading your SSDs reliability/lifespan just to have extra parity, which doesn't make any sense to me! I won't laugh at your RAID-5 as I have the same. I had to build 2 hypervisors on a budget, could only afford 3 storage SSD's, and 2 for the OS. Had to put the the storage SSDs in a RAID-5 and its performance is phenomenal. But I have plenty of bays for future expansion. 1
supportman Posted January 31, 2019 Posted January 31, 2019 All 2TB samsung SSD here, but no RAID. I don't see the point these days with instant recovery. I also read on here that the more disks you have, the more likely one is to fail.
EddyAllen Posted January 31, 2019 Posted January 31, 2019 All 2TB samsung SSD here, but no RAID. I don't see the point these days with instant recovery. I also read on here that the more disks you have, the more likely one is to fail. Kind of... if you have 6 drives in a RAID array you are decreasing the reliability(if you talk about reliability based on if a single drive fails) but gaining redundancy. A RAID0 array of 6 drives will have a MTBF 6 times lower than one drive on its own due to the fact that if one drives fails you've lost the lot. I'm a big fan of storage spaces direct in a Hyper-Converged environment and that effectively does away with RAID controllers in favour of HBA's where you present the drives directly to the OS and have the data distributed amongst several host nodes. Its scalability is amazing as well as there is none of this malarkey extending RAID arrays or needing to rebuild them as the data is redistributed as you add more drives / nodes to the system on the fly.
DrCheese Posted February 10, 2019 Posted February 10, 2019 Those that are doing all flash storage, What are the costs/what servers/disks are you using?
colacao82 Posted February 10, 2019 Posted February 10, 2019 Those that are doing all flash storage, What are the costs/what servers/disks are you using?I would also like to know. Can I buy cheap consumer grade SSDs. Does it matter if I'm a RAID?
3s-gtech Posted February 10, 2019 Posted February 10, 2019 My experience of consumer grade SSD in servers aren’t positive. I tried some for hosting a couple of VMs - the workload/reliability wasn’t as good as I’d hoped (Samsung 840 EVO). The drives still work, but it wasn’t reliable as an array in that server (in Hyper-V). However, I put our SIMS VM on its own (expensive) Intel DC S37xx SSD array in ESXi, that has been superb.
DrCheese Posted February 10, 2019 Posted February 10, 2019 Mixed use drives appear to be around £900/1100 each for 900gb if you add them to a R740 config on Dells website (I know getting a proper quote will be slightly cheaper than the web version) but that seems a bit crazy considering the amount you'd need to raid them properly and get a good amount of space You can pick up third party enterprise SSD's from here https://www.scan.co.uk/shop/computer-hardware/hard-drives-ssd/ssd-sata-enterprise (Or other resellers of course) for far far cheaper, but then naturally that won't count as part of the servers warranty (Any issues the OEM may blame on using non supported disks - I know Dells servers tend to complain if you do that)
DrCheese Posted February 10, 2019 Posted February 10, 2019 (edited) In fact yes, Dell are definately being a bit cheeky - They use the SSDs I've linked above https://i.dell.com/sites/csdocuments/Shared-Content_data-Sheets_Documents/en/dell-poweredge-sas-ssd-performance-specifications.pdf A mixed use 960gb in that list is an "S4600" series SSD. The S4600 has been replaced with this - https://www.scan.co.uk/products/960gb-intel-d3-s4610-series-25-ssd-sata-30-6gb-s-tlc-3d-nand-560mb-s-read-510mb-s-write-96k-51k-iops - Comes with a 60 month (5 year) Warranty & costs £239 ex vat A current Mixed use 960gb ssd on Dells website in the server configurator (Labelled as "960GB SSD SATA Mix Use 6Gbps 512 2.5in Hot-plug AG Drive, 3 DWPD, 5256 TBW) is £1,080.00 ex vat Edit: Apparently using third party disks will mean they don't report usage or failures to the raid card - but if you push Dell they can give you massive discounts. Edited February 10, 2019 by DrCheese
supportman Posted February 11, 2019 Posted February 11, 2019 As I mentioned above you don't need RAID with SSD technology assuming you buy the correct drives and not some Kingston or Crucial cheap thing. We use Samsung SSDs here for our main storage, both 2TB and 4TB variants.
localzuk Posted February 11, 2019 Posted February 11, 2019 Those that are doing all flash storage, What are the costs/what servers/disks are you using? We're running QNAP TVS-EC1580MU-SAS-RP NASs with Samsung PM863 2TB SSDs. Were a little pricey, but they're enterprise class so work well. As I mentioned above you don't need RAID with SSD technology assuming you buy the correct drives and not some Kingston or Crucial cheap thing. We use Samsung SSDs here for our main storage, both 2TB and 4TB variants. That *very* much depends on what you're doing with them. 2
HPlum78 Posted February 11, 2019 Posted February 11, 2019 @localzuk has beat me to it and I would have to agree SSD does not mean that you can disregard RAID and @supportman we need to be careful advocating a certain way as unless we understand the use case in its finite detail we risk misleading those who are asking for help. Sometimes it's better to give some general pointers than put your head and someone else's job on the block. Again as @localzuk says the options and decisions are very much dependant on what the work load looks like (as mentioned in my initial post) and on the appetite for risk and down time along with a load of other architectural decisions.
supportman Posted February 11, 2019 Posted February 11, 2019 @localzuk has beat me to it and I would have to agree SSD does not mean that you can disregard RAID and @supportman we need to be careful advocating a certain way as unless we understand the use case in its finite detail we risk misleading those who are asking for help. Sometimes it's better to give some general pointers than put your head and someone else's job on the block. Again as @localzuk says the options and decisions are very much dependant on what the work load looks like (as mentioned in my initial post) and on the appetite for risk and down time along with a load of other architectural decisions. I take your point (every setup and business need is different), but its worth looking at the technology of RAID specifically and why it was created originally. RAID became popular in servers for 2 reasons: both improved speed of mechanical disks and redundancy because mechanical disk moving parts tend to wear out and suddenly die. Speed is of course a non-issue these days with SSD so we can discount that one right away. Now if you take a look at SSD's they also don't wear out(OK technically they do but you are likely never to reach that point unless you write(not read) Terabytes of data for 5 years constantly), the only thing that tends to go wrong with them is the SSD controller on power events. Now Samsung and Intel both have internal protection against this and proven stable storage. I've abruptly power cycled some SSDs hundreds of times as a test and they have just booted straight back up without issue. In a test lab I challenge you to go pull the power cord from your RAID setups 100 times and see if it survives. Its an extreme test but should show that the biggest single point of failure is not the storage any more, its the controllers. It all comes down to a numbers game in the end, but I think personally RAID causes more issues than it fixes in the modern day. I have not tested a RAID restore recently but i'm pretty sure our virtualisation backup solution would be quicker than rebuilding an array from some 1990s bios interface. I know which one I would prefer anyway.
RobFuller Posted February 11, 2019 Posted February 11, 2019 Well I thought so too until a seminar recently, it appears as if Microsofts stance has changed on this and now they just suggest you DeDupe the libraries which include storage and the operating system for the VM's. The only case I would be not doing this would be for an SQL DB and logs drive. In which case they would just go in a subfolder that has an exclusion rule set on it. I believe this was also echoed in the 2016 MCSA exam. You couldn't post a link to this information about supporting vhdx workloads and dedupe, would like to read more? Thanks
PlantHead Posted February 11, 2019 Posted February 11, 2019 SSD's do fail though and they tend to fail instantly without any warning. If the data is important and you can't afford the downtime then it is still best to implement RAID. If it is the difference between rebuilding a server from backup or just changing a disk, I go with the disk change every time. 2
bossman Posted February 11, 2019 Posted February 11, 2019 (edited) @ataylor We've been using Synology units for the past 4 years with a hybrid mixture of SSDs and large capacity WD 8TB red drives in RAID6 config without any issues whatsoever... Fantastic IOPS read and write plus SSD caching with lots of fantastic apps for offline backups etc etc Currently running with 24TB across our 2 Hyper-V Cluster configured units which utilise Veeam to backup to another Synology unit with 35TB storage which then backs up to Google from the Hyperbackup Application to enable backup offline to virtually any other source... 5 years guarantee plus support... Worth a look if nowt else...we have found them to be excellent for running VMs from our DELL hosts.....and cheap too! Edited February 11, 2019 by bossman
DrCheese Posted February 11, 2019 Posted February 11, 2019 We're running QNAP TVS-EC1580MU-SAS-RP NASs with Samsung PM863 2TB SSDs. Were a little pricey, but they're enterprise class so work well. Brill, thanks for the reply - Do you have these setup in an active/passive setup or dedicate one per host (With Hyper-v/VMware replica between the two) & assume they're connected via iSCSI?/directly into the back of the NAS Apologies for so many questions - I'm toying with simplifying our existing setup (3 Vmware hosts + 1 SAN) as I think it's overly complicated & expensive to keep refreshing for what we do nowadays, given that large chunks of it are moving up to the cloud. We've been hit by quite a few VMware bugs recently that have caused me stress that we just shouldn't be hitting.
localzuk Posted February 11, 2019 Posted February 11, 2019 (edited) SSD's do fail though and they tend to fail instantly without any warning. If the data is important and you can't afford the downtime then it is still best to implement RAID. If it is the difference between rebuilding a server from backup or just changing a disk, I go with the disk change every time. And this is why we use RAID. Its an easy technology that allows for hardware failure whilst allowing your business to carry on running unimpeded. Swapping an SSD is a lot less disruptive to us than having to download everything that was stored on the failed SSD. 1 hour of downtime is a lesson lost that the children will never get back. Brill, thanks for the reply - Do you have these setup in an active/passive setup or dedicate one per host (With Hyper-v/VMware replica between the two) & assume they're connected via iSCSI?/directly into the back of the NAS We have them set up as 2 ISCSI targets, with half our VMs in 1, and half in the other. No NAS to NAS duplication, they're simply 2 NAS's. Edited February 11, 2019 by localzuk 1
EddyAllen Posted February 12, 2019 Posted February 12, 2019 You couldn't post a link to this information about supporting vhdx workloads and dedupe, would like to read more? Thanks It belongs on cringetopia but here is the Microsoft Mechanics video they were referencing. If you can get over the overall cheese-factor of the video its quite interesting. 1
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now