Calling for the Best Storage Recommendations

The funny thing about technology is that there is rarely one answer that fits all situations. This is the root of the reason behind the all-too-familiar IT answer, “It depends.” Rarely is this challenge more evident than when trying to write a book that is applicable to as broad an audience as possible. The challenge is compounded when the topic of your book is a large, complex suite of software like VMware vSphere.

As an author, you want to provide information that is helpful, yet at the same time you know that you simply cannot provide the “right” answer for all the possible readers out there. And you also know that you simply don’t have enough pages or enough time to discuss every single potential factor that might play into how a particular technology should be deployed or configured. As an author, you rarely have the time and page count to go as deep as you’d like. (There are exceptions—the HA/DRS clustering deep dives are good examples.) So, you try to take a balanced approach, making fairly broad recommendations that will apply to the largest group of individuals in the most common set of circumstances or situations.

Because you know you can’t provide the “right” answer for all possible situations and environments, you also know that there will be individuals for whom the book isn’t as helpful as it is for others. Perhaps their particular environments have circumstances that aren’t common, or perhaps their functional requirements drive their vSphere environment in directions most organizations don’t go. That’s fine—that’s the whole reason “It depends” is such a popular answer.

As a result, I can’t say that I was terribly surprised when someone on Twitter pointed me to this Spiceworks thread criticizing some broad RAID recommendations that I made in Mastering VMware vSphere 5. As I’ve already stated, it’s impossible (in my opinion) to provide the “right” recommendations for every possible user in a relatively-mainstream IT book such as Mastering VMware vSphere 5. What did surprise me, though, was that the thread was surprisingly vitriolic, focusing more on me personally than on the recommendations themselves (“Mr Lowe…should not be spreading his ignorance” and “I’ve seen some stuff he’s written and I’m none too impressed”). That’s a shame—it would have been so much more helpful for other Spiceworks community members and the virtualization community as a whole if the thread had focused more on providing “better” RAID recommendations than what I included in the book.

So, I’d like to take this opportunity to issue a call to the storage experts out there (which I am not, and have said that on numerous occasions) to chime in and provide what they think are the “best” RAID recommendations or storage configurations to support a VMware vSphere environment. Keep in mind the challenges I described earlier—you need recommendations that are broadly applicable to many users, many environments, many situations. Please add your thoughts in the comments below, and—where applicable—please provide any relevant affiliations or disclosures. All courteous comments are welcome!

(Full disclosure: I do work for EMC, although this is not an EMC blog and “Mastering VMware vSphere 5″ was not sponsored or supported by EMC in any way. I speak only for myself.)

Tags: , , ,

  1. Snehal Dasari’s avatar

    Not really helpful to the core point of your post, but its extremely poor on behalf of the SpiceWorks community that they didnt even try to provide context around your comments. Part of that section indicates that RAID selection is inherently flexible, and my take away from it was that there is no single silver bullet RAID type.

    Fundamentally, it comes down to performance, availability and efficiency of the storage and how you best match them to the requirements of the solution. For the SpiceWorks users to miss that, is a failing on their behalf.

  2. Paul’s avatar

    Ignore the haters Scott; you are doing the community a service. If someone disagrees with a design you put for then that is one thing, being derogatory and a troll is unacceptable.

  3. Michael Webster’s avatar

    Hi Scott, I must be really dumb here or just simply missing the plot as I can’t see the problem in the generalised statement that they’ve quoted, given the context of the book. Just like you’ve said – it depends. If someone can explain it then I’m all ears and I’d love to know. You could write a whole book on storage for vSphere. Oh wait, Mostafa Khalil did that already and I recommend that everyone reads it.

  4. Tom’s avatar

    As a P.S. to this blog entry you could tell us what you said in the book (not all of us have the book) and then we’d know what we are comparing & contrasting to.

    Also, isn’t it said that RAID5 doesn’t like more than 7 spindles/disks??
    I ask because I have 11 disks in RAID5 in a SAN but the SAN is not over-taxed etc. I/O-wise or IOPS-wise.

    Any recommendations for RAID must consider the load — light, medium, heavy — and of course — “It depends.”

    Thank you, Tom

  5. JC’s avatar

    I think the answer is still “It depends”. IOPS requirement, data protection, capacity, business needs, and budgets (the most important one) are all need to be considered. And it gets much more complicated when SSD offloading is in the picture.

    Cheap, Fast, Good… you can only get two.

  6. Fabio Rapposelli’s avatar

    As you said it depends but your statements are absolutely correct, I would love to see their storage recommendations with a good dissertation on pros and cons.

  7. Andre Carpenter’s avatar

    Firstly, I absolutely agree with your point there is no one configuration to rule them all, if there was then we as architects would be out of a job!

    IMO – Multiple RAID types certainly hold a lot of value from a pure performance standpoint, sometimes selecting multiple RAID types is not always possible as storage vendors tend to lock down the number of RAID options available to us, but High IO workloads can be offset by the use of technology such as SSD and tiering etc

    These days, your leading storage venders are aggregating storage into pools (in the effort to support cloud computing) so that IO profiles are more or less spread across as many spindles (same tier) as possible.

    There are different arguments around RAID type selection to be had regarding availability, capacity efficiency, raid penalties etc. You would probably have to write a separate book altogether taking into account other workloads other than virtual environments giving the shared pool architecture to really cover the subject.

    Also consider technologies such as storage tiering, array based cache as well as server based cache and technologies like storage DRS (which probably shouldn’t be used in conjunction with array based tiering) and RAID selection becomes just one ingredient of many that needs to be thought thoroughly through.

    Dont take those comments to heart Scott, you made a good point they their approach hasn’t been thought through – they should attack ideas with respect rather than the person.

    My 0.02c

    @andrecarpenter

  8. iwan rahabok’s avatar

    Sad to say Spiceworks members focus on emotion instead of logic.
    Here is a good response on such immature behaviour:
    http://sg.news.yahoo.com/silencethehate/

    Have a good day Scott!
    e1

  9. Max’s avatar

    Ignore it. You simply visited a prayer house thinking it was a science conference. It’s really just myths and legends. RAID10 is not an instrument of gods magically protected against any and all failures, it simply has somewhat lower probability of a double disk failure. This crowd can’t acknowledge it, because doing so would mean they’d have to look into setting up reliable backups, and that, in turn, would reduce the potential of a RAID5 data loss to a non-issue.

    As for the right RAID level, I believe the correct position for a sysadmin in the modern world is “I couldn’t care less”. I need X IOPS, Y GB/sec, and I trust my storage vendors to know their products’ reliability and adequately reflect it in their price list. And I know that no matter what I do, it won’t be 100% reliable, therefore backups.

    The reason for this position is the complexity of any decent modern solution. How do I calculate the probability of data loss on a VNX storage pool with 5 SSDs, 15 SAS, and 30 NL-SAS disks, or on a 3PAR system where RAID is not applied to physical disks at all, operating instead on logical disks built of chunklets that come from wherever? Or how do I define RAID level on a Tintri system that would only show me its logical volumes on its console when I boot it for the first time? If I really worried about stuff like this, I would be forever stuck with Dell Powervaults or HP MSAs or other things like these, happy with my little free ESXi hypervisor on my little standalone R610.

    (Actually, with the new VMware storage solutions, those Powervaults and MSAs may experience a bit of a comeback, but that’s a conversation for another day.)

  10. Hans De Leenheer’s avatar

    Couldn’t silence the hate :-) Got in there and got on topic …
    Some people in the industry do deserve some backup, Scott is one of them.

  11. Jason Boche’s avatar

    If I may, the Raid Triangle found in a January 2010 blost post of mine:
    http://boche.net/dropbox/raid-triangle.gif

    The image above is not my original work but serves as a talking point for this discussion.

    Your question specifically is: What do I think are the “best” RAID recommendations or storage configurations to support a VMware vSphere environment.

    I haven’t met a single RAID type yet that satisfies all business use cases but I do think the best storage technology and integration is yet to come. There are common storage challenges that vary quantity and complexity from datacenter to datacenter today. The best overall solution is of course the one that is going to be flexible enough to meet availability, security, performance, capacity, and budgetary requirements or constraints. RAID5 is and has been the sweet spot for a blend of availability, performance, and cost efficiency. I belive that is one of the points you were attempting to make and it does resonate with me. However, without complimentary offload components such as cache, SSD, FusionIO, etc., it’s not going to stand up as well to abusive write I/O workloads as well as RAID10. Unfortunately RAID10 by itself isn’t very efficient from a raw storage utilization perspective. Beyond RAID5, with what’s available today, the best answer isn’t a single RAID type. It’s a virtualized storage solution which leverages multiple RAID types and multiple tiers and types of storage, be it spinning disk, or high I/O flash/memory/silicon based componentry mentioned earlier.

    As for the Spiceworks discussion thread, I wouldn’t expend too much energy on that. The behavior exhibited speaks volumes for those individuals as well as the community if the post is allowed to remain visible. Nothing to see there.

  12. Brendan Clover’s avatar

    I took some time to understand why those comments might of been made and have come to a couple of conclusions:

    1) The Spiceworks forum contributers base all their comments around every word that Scott Allan Miller has ever uttered and treat it as gospel.

    2) Scott Allan Miller (SAM) seems to only work with small business (not that this a bad thing but small business ideas don’t always translate to big business).

    The beef they seem to have with your work is that you recommend RAID5 as against RAID6 or better still RAID10. Now SAM makes some good points about RAID5 and it’s vulnerabilities. SAM also believes every LUN or array should be RAID10 and that’s not a bad option or that costly in an environment where you only have 10TB of disk but move to 50 TB or 100TB and suddenly the cost of RAID10 becomes incredibly prohibitive.

    I think that the text needs updating to reference RAID6 instead of RAID5, the rebuild times on today’s disks make RAID5 very risky. And possibly, to appease those Spiceworks types, include some information about using RAID10 in smaller environments.

  13. matt roeske’s avatar

    I’m currently using two different netapps, an emc vnx, equal logic and nimble and as you said it depends hut I’ve got my money on VNX for primary storage nimble for test/dev and we’re building a supermicro as our cheap low end log store. Also used data domain for dedupe but feel they’re to expensive hopefully the super micro array can best it

  14. Bas Raayman’s avatar

    Disclaimer: EMC employee

    Now that that is out of the way, let me start with something different: context.

    Many of the folks who I see responding here, work with large enterprise level arrays on a frequent basis. Things are different there. :)

    There is no way in which I can compare a Drobo B1200/B800, a Synology DS3612 and other similar arrays to the high-end market that is being served by companies like HDS, EMC, NetApp, IBM, HP, etc. The limitation in the number of drives, the drive types I can use, redundancy and the intrinsic design are different by nature.

    Does that mean that one is better or worse than the other? Well, yes. Yes in a sense that I probably wouldn’t put an HDS HUS100 with 960 disks in place to serve my company with 20 employees. I also wouldn’t put multiple QNAP 16-bay arrays in my environment as the core storage backbone for my company with over ten thousand employees.

    While the Spiceworks community seems to have lost it’s general appreciation for RAID5, I see a lot of customers using this for highly productive workloads in large environments. On the other hand, I know that RAID5 isn’t the best choice for my smaller array at home that I use to save my personal data (I’m using RAID1 there), but I’m happy to use RAID5 for my home lab which can be easily rebuilt.

    The one key thing missing in the entire discussion on the Spiceworks community discussion is context. The person starting the discussion, for example, left out that there are roughly 6 pages prior to the quoted part of the text, explaining the various RAID levels and setting context on common storage architectures and how to define performance requirements. If I don’t have this context, I can’t give out a specific guideline/recommendation.

    All in all, I would say that Scott made a valid statement, and that this statement applies to arrays that are designed for the mid- to large SMB and upwards, and that people might want to put more focus on their array protection and performance level choice for smaller environments, since choices tend to be more limited in smaller arrays, and your design choices tend to have a larger impact on the overall environment.

  15. Scott Alan Miller’s avatar

    In the SpiceWorks community, to which you link above about the discussion, has addressed RAID recommendations at extensive length over the past several years and EMC is a participant in those discussions and in the community as is VMWare. The recommendations, by and large, are RAID 10 as the most common option and RAID 6 as a less common, but perfectly viable option. RAID 5 has been tested and the reliable answer is that the niche for it appears to not exist with the point where RAID 5 starts to become viable being so far out of norm that RAID 0 actually starts to pass it in reliability. I have distilled much of the community information in these posts and talks:

    http://www.smbitjournal.com/2012/07/hot-spare-or-a-hot-mess/
    http://www.smbitjournal.com/2012/05/when-no-redundancy-is-more-reliable/

    http://www.viddler.com/v/ade548be

  16. Scott Alan Miller’s avatar

    I see that there are a lot of people talking about “double disk failures”. I cannot stress enough that the discussions of RAID need to be around relibility, not around disk failure counts. Multiple drive failure is a trivial failure node and used as a means to mislead and misdirect storage reliability discussions. It is easy to talk about redundancy (which we don’t care about in reality) and to ignore reliability (which we do care about) because one is concrete and one must be looked at from a holistic system perspective. RAID 5 is parity RAID and suffers from many failure modes, one of the lesser of which is double drive failure.

    Any discussion using multiple drive failure as a primary topic and not a footnote is not addressing the concerns raised by the SpiceWorks community in question.

  17. Scott Alan Miller’s avatar

    Scott, in this post you say “So, I’d like to take this opportunity to issue a call to the storage experts out there (which I am not, and have said that on numerous occasions)…” but the subtitle of your site is… “an IT pro specializing in … storage” so I think that we are rather harsh in our criticism as you seem to claim to specialize in storage and you work at one of the primary storage vendors.

  18. Tom’s avatar

    Max’ reply seems elitist. PowerVaults and MSAs are perfectly fine and appropriate in many more situations than his beloved high-end storage servers. :) :)

  19. Jamie’s avatar

    This is my favorite topic because everyone argues what is optimal while their applications are pushing 100 IOPS with nothing in the queues. I think if people did a little more reading on data sheets and less on wikipedia there would be more understanding of what is, and is not, impacting their disk performance. Anyway, from another non-expert I would argue there are very few applications out there running in mid-size companies that receive performance benefits of RAID 1/0 in today’s storage architecture with tiering and flash components. Keep it simple, RAID 5 for SAS and RAID 6 for NL-SAS to minimize rebuild issues. Supplement that with some caching technologies and you are probably meeting the demand of your application 10 fold. Worry more about spindles and disk types and less about RAID types and you will probably address your storage performance requirements better.

  20. Brian Boyd’s avatar

    Best thing about VMware is that you can mix differing RAID types and storage vendors…and if things don’t perform you can move them. We tend to create several tiers to try to have customers think about where things should be placed. Is it the answer to everything? No. Does it solve most guest OS requirements? Sometimes. Best thing to do is have VM and Storage admins talk to each other (sometimes this is the same person so it’s easy), understand IOPS, latency, etc and work from there.

    In the very near future, RAID is being replaced with very smart algorithmic parity calculations and less mechanical disk. As things get smarter and faster we won’t need to think about RAID anymore. You’re already seeing a slew of vendors out there that are doing this right now. As it becomes more popular I think we’re going to see more ‘node based’ storage or even the re-emergence of local/server based storage. I.E. VM servers providing their own storage to their guests and some magic behind the scenes that will keep track of all of the ESX clustered hosts storage.

    Fun times ahead!

  21. Mike’s avatar

    Love this – thanks for the opportunity Scott. We’re big fans of your sage advice.

    “It depends” Truer words have rarely been spoken. As an avid problem-solver for clients of all shapes, sizes, and performance profiles, I can heartily say we’ve used RAID1, RAID5, RAID10, and occasionally RAID6. We pick the ‘right tool for the job’ based on defined criteria. Said another way, ‘form follows function’

    However to the point of Max above, the correct position is “I don’t care” – and that we should simply define the performance profile required, then expect the system to deliver it – with a data protection level that is appropriate to the risk profile of the spindles in the set.

  22. slowe’s avatar

    All, thanks for all the responses—it’s great to see everyone sharing information openly. I’m also glad to see that the aforementioned Spiceworks thread seems to have gotten back on track with regards to discussing RAID protection (or other forms of protection) for customers instead of bashing people.

    SAM, one response for you—a person indicating they “specialize in…storage” does not make them an expert. I’ve never claimed to be an expert in anything, as I know there is always more to learn. In fact, even though I am one of less than a 100 people worldwide who hold an expert-level VMware certification, I still don’t consider myself a VMware expert. There is still so much more to learn. Further, although I work for a storage vendor, I don’t speak for that storage vendor (never have and never claimed to). Thanks for taking the time to add your comments here!

  23. VMTyler’s avatar

    Scott (Miller)-
    Your points about RAID5 apply to some of the scenarios that you laid out in your blog posts (large arrays of SATA drives), and I’m glad to see more people talking about the effects of URE on RAID rebuilds. That said, there are still plenty of use cases that RAID5 protection is sufficient (Enterprise SAS/FC drives, smaller raid sets) and is common in virtualization environments. That’s not even getting into some of the technologies used by storage vendors to minimize the exposure to these failure modes and well as other failure modes that have nothing to do with the drives themselves.

  24. Nic Tolstoshev’s avatar

    Spiceworks community manager here. Scott, I’d like to apologize for the comments on our site that got personal. They aren’t acceptable on our community and I’ve removed them. Thanks for engaging in the debate anyway, and hopefully we can keep this about the technical issues and not about personal insults from here on out. Unfortunately RAID 5 is a hot button topic among our community, that is no excuse for taking the debate into the gutter.

  25. Scott Alan Miller’s avatar

    Scott, understood. I think that being a CTO and specializing in storage makes us assume that that implies specialist. As an EMC customer, I expect that the EMC CTOs are the world’s absolute authorities on storage. I’m a little confused that you would not be considered an expert?

    I appreciate that you do not speak for EMC directly nor consider yourself a storage expert. But as a member of EMC’s C-suite, I think that there is a certain implication that you are a major component of EMC’s interface to the world.

  26. John Nicholson’s avatar

    Scott,

    Sorry for things getting out of hand in the forums there briefly. I nuked what I could last night and the rest of the admin’s cleaned it up this morning when they woke up.

    I do think that RAID and disk selection is one of the least understood things in virtualization today. (I’ve had customers request I configure or configure on their own 92 disk RAID 5/50 configurations with SATA). I’ve had fortune 1000 companies try to run VDI on RAID 5 NL-SAS and be in shock and horror when I question their SAN configuration. Considering storage is often as much as 40% of the cost of virtualization projects, some good deep dives into profiling, sizing, and planning are badly needed given the amount of million dollar mistakes I see on a daily basis in storage sizing and RAID selection. I know EMC SE’s (as with most major vendors) try to do this work before sales, but I feel like for every SAN sale where things are properly designed, I run into 10 that are based on RAW TB of space per $ blindly, with zero pre-sales consulting done by vendor’s or requested/allowed per customers.

    Storage is strangely the thing where I see the most egregious wastes of money today that should have gotten someone fired, but strangely its the last one that ever gets discovered or looked at. Its a strange world where customers often don’t actually know what your performance or usable space is REALLY going to look like until long after the sale.

    Its something that you might want to consider tapping another resource for the next update on the book because while I know its not the main focus its honestly one of the hardest and most expensive things to undo when done wrong.

  27. Christopher Waltham’s avatar

    Disclaimer: I am a NetApp employee.

    Umm, RAID-DP is pretty freakin’ good..!

  28. Eric Price’s avatar

    After 20 years of IT work (I learned RAID 1 / RAID 5 with everyone one else in the 90s) Im ready to see us as a community move beyond “it depends” to something more substantive and specific. Ok, it depends – On What? When does RAID 5 make sense? Where does it stop making sense? What are the risks? How common are those risks realized? Are SATA drives less reliable that SCSI drives were, or is it a function of size? Lots of questions, and I suspect if we get the right people involved, we can finally get some meaningful answers. I hate the vitriol and ad-hominem attacks, but I for one will be following closely to see if we can learn something beyond vague generalities in the months to come. Good luck to everyone! :D

  29. Ravi’s avatar

    The world of storage is changing drastically with new technologies and the best practices has to change accordingly.
    My primary focus as part of my day job at Pure Storage is to articulate how Flash storage challenges and changes the general prevailing storage practices and in particular a virtualized data center. Storage in the new world would be so much simpler and the traditional RAID may become part of the system itself and users just create LUNs and use it with ease.
    To that end, I have done (and continue to do) extensive work on our all flash storage and have documented them on the blog site here –
    http://www.purestorage.com/blog/virtualization-and-flash-blog-post-3/
    http://www.purestorage.com/blog/vm-performance-on-flash-part-2-thin-vs-thick-provisioning-does-it-matter/
    http://www.purestorage.com/blog/say-goodbye-to-vm-alignment-issues-and-poor-performance-with-pure-storage/

    I encourage reader to review this and comment on it.

  30. Russell’s avatar

    The problem I see with applying “consumer level” thinking with determining RAID levels is that it suddenly isn’t applicable when you bring an enterprise storage array into the equation.

    Since one can write hundreds of pages on the subject it seems the best way to sum it up in a book that is absolutely not about storage is to provide some basic guidance with the “it depends” caveat.

    When I read statements like this:
    “RAID 5 has been tested and the reliable answer is that the niche for it appears to not exist with the point where RAID 5 starts to become viable being so far out of norm that RAID 0 actually starts to pass it in reliability.”

    It just blows my mind. There are plenty of use cases where RAID 5 is appropriate and depending on vendor implementation you may not even need to worry about things like UREs making your life hell.

  31. mjbrender’s avatar

    As an early EMC Engineer to join Spiceworks and advocate for its adoption internally, I’m embarrassed to see such a childish attack. It could not have been more unprovoked by a nicer human being.

    From my year-or-so exposure, I don’t think this behavior is representative of the millions of Spiceworks members out there or the few dozen I’ve met in person.

    It is, though, notably disappointing. I expect more from any community.

  32. Russell’s avatar

    “Scott, understood. I think that being a CTO and specializing in storage makes us assume that that implies specialist. As an EMC customer, I expect that the EMC CTOs are the world’s absolute authorities on storage. I’m a little confused that you would not be considered an expert?

    I appreciate that you do not speak for EMC directly nor consider yourself a storage expert. But as a member of EMC’s C-suite, I think that there is a certain implication that you are a major component of EMC’s interface to the world.”

    I wanted to touch on this because I think it’s important to understand that EMC has a great deal of products. Some of these products are storage-centric and some of the products are not.

    EMC may make a good chunk of revenue doing storage but they’re also a solutions company. Storage is only a single component to a given solution and it’s important that EMC has expertise in other domains besides storage products. Scott’s technology background makes him an ideal fit to sit at the top and drive solutions because he knows enough about each domain to intelligently communicate with those on his team with a deeper knowledge of the subject matter.

    That said you’re still driving home a point that I don’t think is relevant in the context of a book that is absolutely not about storage by any stretch. You yourself have linked dozens of pages of your own writing on the subject. Clearly a pretty significant volume can be written on the subject.

    So what do you do when you’re writing a book that is about a specific virtualization product? Do you detract from the core subject and spent some chapters on just storage? Or do you put a caveat in there with what’s going to be a starting point for the reader to take and do further research?

  33. nate’s avatar

    For me it’s less about vmware and more about the applications running on top of it, right? I’ve noticed in my clusters which are good ‘ol thick ESX 4.1U2/U3 that each node does about 30-40 write IOPS to it’s own volume per server.

    My current organization’s workload, which was migrated from a major public cloud player turns out to be about 90% write on the front end, closer to 50/50 on the back end. Others may be different, we didn’t have the i/o information up front(cloud metrics were crap, oh look there’s 4,000 IOPS! with a 0.1kB I/O size!), most of the reads are done from in memory in the servers, our MySQL DBs pretty much fit 100% of their reads from memory, so fancy flash acceleration that only accelerates reads really would have no benefit to us.

    A key thing for me has been for years – flexibility. Have a storage solution which has the ability to change the configuration after the fact without having to do massive data migrations. Everyone makes mistakes, and even if your perfect up front, the I/O profile may change down the road. So to be able to dynamically change the level of RAID of data volumes, even when all physical spindles are provisioned to me is really great. More power to you if you can run multiple levels of RAID on the same spindles. I remember a couple years ago a discussion with HDS where they claimed hot conversion between RAID levels, but I had to drill them on it and they finally admitted that you needed blank disks to migrate to, which sort of killed it for me.

    For me with only very few exceptions build my RAID so that I can survive a full shelf failure(quite rare but can happen). The current storage system my small organization has runs on 4 shelves – 4 half populated ones. So we run RAID 5 3+1 for the most part (some RAID 10, not much). On the system (which is HP 3PAR – my main background on the storage side is 3PAR) that gets us to about 90% of the performance of RAID 10, with significantly less overhead(vs RAID 10). I wouldn’t advocate 3+1 on another storage architecture that doesn’t leverage sub disk RAID because the number of spindles you’d lose to parity is a large number, with 3PAR of course there are no dedicated spindles for RAID or hot spare. The parallel RAID rebuild is also somewhat unique and negates a lot of the need of RAID 6 with current drive sizes, but if you really wanted to go all out you could go RAID 6 (I prefer RAID 5), and if you have 8 shelves the system will automatically protect you against the loss of up to 2 shelves(in the big systems this is up to 80 drives if fully populated). Your I/O may tank(depending on workload) but your data will remain online. With 8 shelves I would suspect most of my hot data would stay in 3+1 (two sets of 3+1, one per 4 shelves of disks), really hot data would go to RAID 10, and colder data would go to 7+1. All of these workloads of course running on same disks, there would be no dedicated spindles/raid groups.

    Though I am a long time 3PAR customer across many companies, user etc. I learned more about Compellent earlier in the year and really do like their approach (would be especially useful with my 90% write profile) where they send all writes to the highest tier by default and then auto tier it down later(this function is included in the base software license – it’s not part of the auto tiering software addon). I’d almost kill to have that on a 3PAR box. Though that is going beyond RAID and into tiering(I haven’t seen any other auto tiering solution that gets me excited even the one from 3PAR), then there is SSD caching(EMC FAST looks great on paper, never used it and haven’t talked to anyone who has) etc which too is beyond RAID.

  34. Robert Pelletier’s avatar

    I find it wierd that there is only one comment on RAID-50. Why go RAID-10 instead of RAID-50? Why lose 50% of your available space instead of N-2 or N-3 or N-4?

  35. VMTyler’s avatar

    In the words of the late Steve Jobs – “Just one more thing..”

    One thing I didn’t mentioned that I wanted to include here as well. Solid state (flash) drives have some very interesting properties about them, one of which is that (depending on the flash controller chipset) UREs are basically nonexistent. With the large amount of IOPS available per drive and (currently) the small amount of capacity, RAID5 is commonly used for this growing slice of the storage community as well.

  36. mjbrender’s avatar

    And to end on a fun note, here’s a Spiceworks post I can get behind:

    Scott Lowe > Chuck Norris

    http://community.spiceworks.com/topic/217819-scott-lowe-chuck-norris