4

I'm looking into purchasing a new storage backend for our VMware vSphere Essentials package. We have 3 ESXi hosts, currently mainly using DAS. The goal is to upgrade to Essentials Plus this year, to leverage its extra power in High Availability and other integrated availability options.

As a long-time Synology user, I was thrilled at today's announcement of the new RS815+ models, with 400MB/sec throughput and support for serverside HA and SSD caching. I'm intent on deploying this as the iSCSI backend for our vSphere data store. I'm not sure however on what is required to achieve the highest possible level of redundancy.

Assuming a single server setup, the obvious options are:

  1. 3x HDD in RAID-5 with 1x SSD as cache
  2. 3x SSD in RAID-5 with 1x SSD as cache
  3. 4x SSD in RAID-10

The data throughput is limited by definition at 400MB/sec because of the 4 trunked Gbit LAN ports. This brings the main questions, all related directly:

  • Am I right in assuming RAID-5 is still not advisable for present-day SSDs, thus making option 3 the best for durability? Current Samsung 850 Pro series carry an unlimited 10-year guarantee, so it would appear this is no longer relevant.
  • Am I right in assuming that option 1 will perform worst of these setups, as being the storage backend of 10+ servers means continous unpredictable random access, causing too many cache misses, thus making the HDD RAID array the bottleneck instead of the network?
  • Is there any advantage of option 2 over 3 or vice versa, as I think both will perform identically in real world scenarios?

To complicate, I'm considering shelling out for 2 rackstations, to utilize the HA features of DSM. This means there will be extra network and disk load for synchronization. Would I need to set these servers up simultaneously, or could I add the second one months later (budget constraints...)? And would it need to have the exact same storage config, or could I for example fill it with much cheaper HDDs, as it will only run in passive mirror 99.99% of the time anyway?

Or is redundancy at the Synology level not even required, considering that vSphere Essentials Plus provides vSphere Replication, which should allow me to mirror to an NFS-datastore on one of our much slower other storage servers, as it's purely for fallback anyway?

  • What are these ` integrated high performance options` that you speak of in Essentials Plus? – joeqwerty Jan 21 '15 at 13:11
  • @joeqwerty was poorly worded - I meant the other high availability and disaster recovery options like Replication and the Data Protection APIs. – Niels Keurentjes Jan 21 '15 at 13:20
  • This should be interesting. One of my clients recently moved from local storage on ESXi (using HP ProLiant servers) to a Synology NAS... I'm opposed to the idea because it seems like the solution is now less robust. – ewwhite Jan 21 '15 at 13:59
  • @ewwhite with DAS however most HA scenarios will not work, as for example vMotion is impossible without shared storage. Why do you consider it less robust? – Niels Keurentjes Jan 21 '15 at 14:42
  • 1
    I say "less robust" versus an enterprise SAN with support... I think Synology are good NAS devices, but not for primary virtualization storage. Here's a good read from Spiceworks - Synology NAS for VMware Environment? Anyone do this and their thoughts? – ewwhite Jan 21 '15 at 14:45
4

To answer your questions:

  • RAID5 is not advisable, period. A single bad read during a rebuild kills your entire array. This is more of an issue with HDDs than SSDs, but it's still a consideration. RAID10 is generally the best option, but RAID6 (dual-parity) can also do quite well.
  • You're comparing Jaguars to Hyundais, there. SSDs have better random access, but are vastly more expensive per GiB. You need to have a good idea of what your current IO usage is to get any useful indication of what you should go with. Enterprise-class HDDs aren't exactly slouches, and reading data from one isn't going to bring your server to its knees.
  • The performance difference between option 2 and 3 would be dependant on your current IO usage. If you have completely random reads, option 3 would be better.

If I were to suggest something to start with, not knowing your IO usage, it would be to get the bay expansion and populate it with 5-6 reasonably sized HDDs in RAID6 and a SSD for caching.

  • As for the Jaguars vs Hyundais: price is not really an issue there. I don't need terabytes of storage, so the price difference between 3 SSDs and 3 HDDs is acceptable - if, and that is the main question, the performance gain is measurable. Thanks for your answer though. I like the point about RAID-5, so it seems 4 SSD RAID-10 is the way to go. Do you have anything to add about the followups on server-level redundancy? – Niels Keurentjes Jan 21 '15 at 23:38

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.