Ilja Coolen \ ICSS

I just read StorageBod‘s post on “Your Data, Your Responsibility” and I totally agree with him. You should really be very aware of the risks you are potentially facing when deciding to store all or most of your data in a cloud, no matter who’s cloud it is.

The remark on storing it in two different places triggered me to rethink something else in this context.

When you have decided you will not be open for the risk of storing the data with one cloud provider only, you might even consider having a different cloud provider for holding the second copy. Because if you store two copies at the same provider, you do not mitigate the risk of this single provider having a massive outage and possibly losing your data or even going “belly up”.

Selecting two cloud providers is most likely even more difficult than just one. You have to negotiate two Service Level Agreements, pricing models and such. If you consider all this effort and risk mitigation you are doing, is storing your stuff in the cloud still a lot cheaper or more convenient than providing for your own storage? Is the cloud still providing what you are looking for, such as reducing administrative labor, footprint, cost? I could only be sure about reducing footprint, the rest is dependent on a lot of other things.

I myself am very paranoid about losing data, so I always store all data I consider valuable in two different places.

I don’t have all the current prices for cloud storage services, so I cannot make a good comparison, but I am surely curious.

Well,

It has been quite a while since my last post. I’ve been very busy working on a number of storage proposals. Which I guess is good in this time of financial crisis.

I’ve been working on IBM San Volume Controller for a number of years, and I even had a contract as a Storage Instructor for IBM San Volume Controller.  I have had many questions or remarks as to why the SVC still had no support for role based access and centralized account administration like LDAP or Active Directory. And honestly, it did surprise me how long it took IBM to incorporate it into the code.
Only recently, the SVC code 6.3 has been released sporting AD and LDAP user authentication. I have not yet have any change to test LDAP, but I did get Active Directory working. The IBM documentation isn’t all that detailed as to how to setup AD authentication. Neither has the AD support team on site been very helpful (accept maybe for one guy who seems to know what AD is all about).

There are a couple of things you should be aware of;

• The SVC is an appliance that is not trusted by AD, it is not part of a Active Directory domain. Therefore you might need to provide an AD account that has sufficient credentials to use AD to authenticate.
• The account you will use to query AD with (not the account a user uses to attempt a login) has to be configured with it’s complete AD location path.an example: you are provided an AD account that is privileged  to query AD that is called ‘ibm2145queryad’ and sits in the AD tree at “DOMAIN\Accounts\ReadOnly” (<-fictional).You will need to enter the ldapconfiguration in the SVC console as;

  chldap -username 'CN=ibm2145queryad,ou=ReadOnly,ou=Accounts,dc=DOMAIN,dc=corp' -password 'XXXXXX' -security none -userattribute sAMAccountName -groupattribute memberOf -auditlogattribute name -type ad

• Even if AD authentication fails because the query user is not authorized to query AD, the ‘testldapserver’ will result in a success. If you are testing authentication of a valid user login, this will result in a message that the user is not correct or the password is mismatched. This suggests the user trying to login is invalid, but actually, the account used to query AD is invalid. So this might throw you of a bit.
• Only supply AD servers or LDAP servers, do not mix AD and LDAP servers. This will not work as desired.

Your SVC administrators have to be members of a specific security group in AD, separating them from ordinary users that should not have SVC access. This group name has to be defined in the SVC as a remote group. Only users that are member of that specific security group in AD will have access to the SVC. The user group on the SVC has to be configured with specific access rights on the SVC, like “administrator” or something else that matches your company requirements. This role is also assigned to AD users that are part of that security group at the moment the successfully log in to the SVC.
When using LDAP or AD user authentication, local users still are usable. You will need a local account with the “Security Administrator” privilege to update LDAP/AD settings on the SVC. I guess you do not want to make all authenticated users have the highest privilege on SVC.

Once you know these quirks you will  be able to configure and use LDAP or AD services to authenticate your SVC administrators with. Two things I have to comment about are the fact you cannot enter LDAP or AD servers by their hostnames. You need to enter the IPv4 addresses. This also shows the second annoyance. There is no support for IPv6 yet. I guess it is on it’s way, but who knows how long this will take IBM.

The HDS Blogger Day or Geek Day 2011 was a blast. It was my first HDS Blogger day and when I entered the venue, I was kind of surprised by the number of HDS representatives in the room. At first it felt like a hearing committee. The HDS to blogger ration was about 2 to 1. After the introduction though, it already felt comfortable. HDS had assembled exactly the right sales and engineering staff, who were there clearly to listen to the feedback the bloggers were giving, and to respond thoroughly to whatever we could ask. HDS clearly knows how to handle a Bloggers event.

Corporate cultural change towards openness and feedback.

From the recent Hitachi Data Systems events a certain message can be distilled. If you know a bit about Japanese culture, than you will probably also know that in general Japanese people keep to themselves, as in they will not brag or tell you what they have or can do. This attitude has kept the rest of the world from knowing what is going on in Japan, and especially within Japanese IT land.

The earthquake and tsunami of early March would have been disasters that the rest of the world would have known little of, if they would have happened a few years back. This indicates that Japan is opening up a bit more to the rest of the world.

The same is true about the Hitachi philosophy. Hitachi Data Systems is now reaching out to the  IT crowd in the world to tell them what they are up to, and ask the customers/end-users what it is they want or need. This is a major change from what they have come from.

Hitachi now has to get that message and philosophy out into the world. Events like the Blogger Day and the Information Forums contribute a lot to that message, and I honestly have the impression Hitachi stands behind their message. So I would sincerely advise my readers to have a new and refreshing talk with your HDS representatives about this new approach.

Hitachi Servers

Lynn Mclean, Vice President Sales, gave an introduction to the server portfolio offered by HDS. Yes, servers. Hitachi is selling servers for some 50 years now. From mainframe type servers to blade chassis and servers today. The Hitachi Storage Division (HDS) is no longer limited to storage. As of 2009 Hitachi Ltd. has decided that it is up to the Hitachi Data Systems group to enter the server market in the US and EMEA region. If you consider that servers process data, that will make sense. Otherwise Hitachi should change the HDS acronym to “Hitachi Datacenter Systems”, to encapsulate all products they are covering in the datacenters. That would even work on their networking gear. Yes, HDS even has networking gear, but at this moment (Sean Moser gave away a small teaser here though), this is no fancy revolutionary stuff, and is not sold outside Japan. From my local HDS rep I also understood Hitachi even has a DBMS and Unix of their own.  I am completely flabbergasted. I would definitely love a peek into that stuff.

So in case you are now thinking whether or not you have been living under a rock for 50 years, because you didn’t know HDS is selling servers, the answer would be no, you’re probably fine. Hitachi actually did not sell outside Japan up until 2004, so they are fairly young in the US and EMEA server market, but not young enough to get away with existing completely unnoticed. Therein lies the actual problem . Hitachi has a huge marketing flaw that has allowed the Hitachi servers to go unnoticed for this long.

The gear we have been introduced to, actually seems nothing less to what other vendors have to offer. It’s decent and high-tech equipment of at least the same standard as all Hitachi equipment.

Blade failover, blade stacking are features supported by the majority of blade vendors. The one thing that does stand out is the fact the blade chassis has 8 internal PCIe busses available for expansion, along with the somewhat standard mezzazine slots.  If those PCI slots aren’t sufficient, you can externally expand with a 16 PCI slot expansion unit, which interconnects to the main chassis on two of the internal PCI slots using PCI-Express connections. This expands connectivity massively for a blade chassis, but the performance is not enhanced, as you are limited to the bandwidth of the PCI lanes of the two busses in the chassis.

Hitachi already recognized that this needs more work. Hitachi is able to deliver a completely filled 42u rack with 320 high density micro servers. The total rack would consume less than 12 Kilowatt. Whether or not this is a great accomplishment, actually depends on the total processing capacity this rack would have. I need to dig deeper into this to make a comparison. There are bloggers out there that know server stuff better than I do, so I will be collection a number of related posts later on. For now, you will have to do with some HDS documentation I could find that contains a reference to the Hitachi servers.

They have no other differentiators I know of, but that could also proof that they potentially could have been right up there with the server market leaders, if they would have done marketing right.

Should we worry about Hitachi trying to enter the server market to take on vendors currently ruling the datacenters? According to Lynn, the approach will not be to take on the other big dogs, but include the servers in a complete solution package, which I consider to be comparable in concept to UCS or VCE type solutions.

It is good that Hitachi is not arrogant enough to pick a fight.  I think they earn respect for that attitude. Searching the HDS website gave me no results on servers except some white papers, so I  think Stephen Foskett’s quote is spot on.

“HDS is smart enough not to wander into the blade server saloon and pick a fight with the big guys at the bar #HDSday “

More to come from this event on “Hitachi Content Platform” and “Storage Economics” when I get time to write….

In the mean time, make sure you read these related posts (updates will follow):

• Chris Evans :  Hitachi Bloggers Day 2011 – Part I (focus on VAAI)
• Hu Yoshida : An Amazing Week of Content and Context
• Bundled search result by HDS
• Juku IT : Hitachi’s new VMware strategy (focus on VMware)

Disclaimer:
HDS has invited me and a range of well known and respected bloggers to visit HDS EMEA Headquarters at Sefton Park UK for  the 2nd HDS Annual Blogger Day. Travel expenses and meals have been paid by HDS and there we are not obligated in any way to write about what we have seen or heard. My time has not by compensated, except for the good company I was in.

Tomorrow I will be attending the HDS 2011 Geek / Blogger Day. In light of this event and the recent Calvin Zito (@HPStorageGuy) post on the HP APEX features I had some thought on this.

We all know HP OEM’s the HDS VSP product, as they did (still do actually) for the XP (HDS USP) series. In the third quarter of last year, HDS announced the new VSP product line, instantaneously followed by HP and the P9500. HP is differentiating itself from HDS by having some additional features that are not available from HDS.

The HP Application Performance Extender (APEX) enables certain host operating systems (currently Linux, Windows and HP-UX) to prioritize their host I/O over other systems. This enables IT departments to set performance characteristics to their core applications to ensure the optimum services to their (internal) customers.

In the post by Calvin Zito I mentioned at the beginning, you can read the APEX software now also allows for LDEV ownership transfer to a less busy controller in the P9500. This way you can also optimize the array usage without depending on APEX host agents.

For this to work properly, you might want to make sure you do not set conflicting values for the APEX host agents and the settings for DLOM. My question, will the DLOM micro-processor ownership changes also change preferred LUN/ALUA settings in respect to the host, if they notice it at all? I believe not all native host MPIO drivers like this while being actively used.

Anyway, one of my questions for the HDS Geek Day will be whether HDS thinks an APEX type feature is not needed at all, or if they are working on a similar feature to be used by the VSP.

Since a few days (as of March 10th), Sony has a new PS3 firmware available. Version 3.60 to be exact.

As of this version, you can copy a number of your savegames, not all,  to your PlayStation^tm Network (Plus only) account. This way you can download all your “cloud stored” savegames to another PS3 console (with the same firmware obviously) and resume your gaming there. Be sure to copy them to the cloud again after you’re done gaming.

Note, it is a manual process. You have to navigate the crossbar menu to the Games options and then copy the savegames to your PlayStation^tm Network account.

I have been trying to figure out if I have a Plus account or not, but so far, all I got from the PlayStation^tm Network Plus site is a 404 error. It is also really annoying to navigate the website because it is cluttered with banners and game info, and to me it far from informational. I have been trying the website out since I got my PS3 a couple of years back, but it hasn’t improved much over time.

I do think it is a good move to enable “online” storing of your save games, but is it what we need? I think not. The Nintendo Wii did a better job I guess, by enabling you to store your game info on one of the controllers and bring that along to your friends place or wherever.
The problem to me is that it still is a manual process. Sony should make it an automatic synchronization (maybe a la Dropbox) feature you can enable in the settings menu.

The 150MB online storage space isn’t much, but it is more than enough to store a bunch of savegame data. Game progress information is only small in size. Sony has limited the online storage to 150MB or 1000 files, which seems like more then enough for me, but I couldn’t hardly determine the required space for hardcore gamers (which I am not).

Sony’s move is displaying a trend though. A lot more devices and not limited to game consoles or personal computers, are moving towards the online storing, or storing in the cloud if you wish.
My home cinema soundbar (HTS9140) has upgradable firmware (it is Busybox Linux under the hood), and it is quite possible that future releases enable me to save or read cloud saved configuration data. Heck,  I could even make this feature myself, because Philips offers the source code on request.

But for what purpose would I want all my consumer electronics connected to and storing in the cloud? We can all think of a number of things, like a real-time fridge inventory for my friendly neighborhood grocery store, so he or she can deliver new beer if my fridge tells him or her my stash has dried up.  I’d rather have  my frensh-fry cooker tell me on my PlayStation mid-game that the fries are done, or maybe even force a pause mid-game  to prevent a fire in the kitchen. Not speaking from experience here by the way !!  Will it come to that? Who knows…. it will only depend on the demand we can all generate.

I suspect most cloud-stored consumer electronics information will be vendor controlled and mined and might not interact with other devices in the same home, unless they are from the same vendor.
As a data center technology guy I have encountered numerous compatibility issues. This might all become true for home electronics as well if we don’t set the demands and requirements from a consumer standpoint instead of  vendors perspective.

I am really looking forward to what is yet to be invented…

Recently I have been reading into filesystems a bit more. Currently, I am interested in WORM filesystems. I can see the potential of WORM filesystemens for long term archiving of data that is only used for reference. For me, long term archiving automatically means performance is of less importance. With the large capacity disk drives out there currently, neas line storage could be cheaper then putting files in an offline (I mean outside the storage system it is initially stored in) hierarchical storage management (HSM) system. For an HSM solution, a fairly costly infrastructure has to be maintained as compared to a single (large) WORM filesystem on high capacity disks drives like a 2TB SATA drive.

The widespread discussions repeatedly warned us about the risk of losing more than one SATA drive during the rebuild of a previously failed drive because of the duration of that rebuild with the large capacity drives, also known as a double disk failure. This could potentially be very harmfull for your data, depending on the protection you have chosen. RAID10 and RAID6 can go a long way but are still limited. This is a clear example of a potential WORM filesystem corruption.  Or isn’t it?

There might be various reasons a filesystem can get corrupted.  What I am particularly interested in, is how the well-known vendors have tackled this problem.

So despite all the cloud storage buzz going around on the internet, there is quite a few places or companies that would still prefer a privat solution for various reasons. If you wish to call that a privat cloud, that is up to you.

So to all vendors offering WORM filesystems (by means of NAS or whatever other way you see fit) :

• How does your WORM filesystem withstand the dangers of filesystem corruption?
• Isn’t your WORM filesystem susceptible to filesystem corruption? How come?
• When corruption should occur, how can you recover?

So if you have any information you can and would like to share with me and the rest of the world, please respond in the comments, or shoot me an email, and I will post the updates here.

I have been working with a BlueArc NAS head for some time and just recently the requirement for NFSv4 and ACL’s had come up.

For a while I have been looking for some more info on how the NFS usermapping and ACL’s works, and especially for NFSv4. We were looking for a solution to the problem where the mapping of the Unix usernames and userid’s did not align with the NFS Server’s configuration, although we could not determine why.  When the user id and usernames cannot be matched with the NFS server, the active users credentials are squashed to anonymous. That makes it hard to enforce some decent ACL policies. The NFS client log clearly stated that it could not map a user to a domain.

Feb 25 14:24:26 nfsclient rpc.idmapd[30316]: nss_getpwnam: name 'username' does not map into domain 'domainname'

It was clear the usermapping differs between NFS3 and NFS4, but I did not quickly find what the difference was. The vendor’s administration guide wasn’t very elaborate on this topic, so I reverted to the beloved Google searches to try my luck. Unfortunately most hits were related to a bug in libnfsidmap.so in utils-nfs-lib which was supposed to be solved in versions which seem old to this time. So that could not be the cure to my issue. I accidentally stumbled across the solution by reading the command line man pages for the NAS head. It is just another example of an administration guide or command reference that is not as elaborate or at all complete as the man pages supplied with the management interfaces supplied by vendors. This BlueArc example isn’t on its own here.

The NFS server config wasn’t like the NFS clients configuration, because the NFS server was a NAS head. The NFS client uses the idmapd to match unix names with unix ID’s when connecting to NFS. The client uses domain information stored in a file /etc/idmapd.conf. Look at the section [General].

[General]
Verbosity = 0
Pipefs-Directory = /var/lib/nfs/rpc_pipefs
Domain = domainname

The value behind “Domain” should match the configuration of your NFS server. In the case of NFSv4 user and group mappings, this is the value that should follow the @ behind the corresponding unix user  name.
An example;
On the NFS server, there has to be a mapping between Unix user “username” with  id “2010″ and in /etc/idmapd.conf the domain name “domainname” is set. On the NFS server, you use the NFS specific utility to configure a mapping for NFSv4 for “username@domainname”.

At this point I can only supply an example for a BlueArc HNAS usermapping.

BLUEARC CLI#> user-mapping-add --unix-name username --unix-id 2010 --nfsv4-name username@domainname

After succesful configuration, the NFS client log should display the following.

Feb 25 14:24:51 nfsclient rpc.idmapd[30316]: Client 15: (user) name "username@domainname" -> id "2010"

The same method applies to the group names, which obviously will need a mapping corresponding to the group names and ID’s.

So for all the people looking at the same problem in the near future, this post should help you a bit.

This morning, I saw a tweet about a x86 blade being shoved into a zSeries frame.
It appeared to be an IBM Press release introducing new developments in IBM zSeries land. Like tape, mainframe is dead for years (so they say). There are not many analysts that believe this statement, and mainframe is long from being dead. I do see fewer of them however. Only the very large (mostly financial) companies seem to be able to run zSeries workloads because of the expertise and cost involved with running zSeries.

Over the last years, IBM has done a lot of development in mainframe equipment and really has brought down cost of running mainframe gear. For most IT folks, the mainframe has lost its sexiness (if it ever was sexy), and it has gotten really hard to find decent staff to operate mainframe gear and workloads. So in a technical and financial sense, the mainframe might be long from dead, but without good staff, who can run mainframe gear in their shops? I have been seeing a lot of new faces in the IT industry, none of which seem to be developing skills in the mainframe arena.

The open systems world seems to be more exciting because the development is done much faster and cheaper (although I myself might not agree with the cheaper part). Many new developments in the various IT stacks like networking, storage, systems and software are solely targeted at open systems worlds, completely ignoring the mainframe world. The vendors we spoke last week at Tech Field Day also have no plans on developing for mainframe. Part of which is the mainframe vendors own fault, since they have really closed down access to mainframe development resources for everyone.

zEnterprise (z196)

The new zEnterpise will be available later this year, and will hold 96 of the worlds fastest CPU’s running at 5.2 Ghz. It has water-cooling enablement. Funny, because water-cooling was removed around the time I was introduced to the mainframe world, back in 1996. This system is going to have 60% more capacity then its predecessor “System z10″, while consuming about the same amount of energy.

Introducing the BladeCenter Extension.

IBM is also developing gear which is supposed to simplify the data-center. The’d be stupid not to obviously. The BladeCenter extension is a frame that can be attached to the new IBM zEnterprise “main”-frame which will be able to hold a number of “open systems” blades.

” The IBM zEnterprise BladeCenter Extension allows supports purpose IBM POWER7 and System x BladeCenter systems as well as blades optimized for specific workloads, such as analytics and managing Web infrastructure.”

• Later this year IBM will be introducing the Power7 blades to run IBM AIX
• Next year, xSeries blades will be running Linux OS in this extension.

Using the new Unified Resource Management software, IBM claims to be able to run over a 100,000 virtual machines on a fully configured zEnterprise system.

The mainframe software has a very well deserved reputation of being extremely manageable and configurable and is well known for its stability and predictability. My life in IT once started as a MVS operator, so I always have had a weakness for mainframe environments.

What this will hold for us in the future, who knows, but if IBM manages to gets the virtualization part of the ground including Microsoft Windows workloads, this might be another player in the UCS and VCE arena worth watching, although I sure hope there is a way to run this zEnterprise system without the need of mainframe system engineering skills. If these skills are required to operate this system, I think the market is limited to the current mainframe shops and will pose no threat to the UCS and VCE solutions.

My opinion on this is, unless IBM manages to run this system with the server virtualization features a la VMware or Microsoft Hyper-V they will have a hard time selling this. Even in the shops that already deploy mainframe gear.

The data center convergence question I have for IBM is; when will you join in convergence with “Ficon over Ethernet (FioE)” or in accordance with recent Tech Field Day developments FCoTR?

But I love to be educated on the markets IBM is targeting and how they would be doing that.

This post actually is more of a summary of notes I took during Tech Field Day in Seattle 2010.

• NEC is very environmentally aware and has some very strickt rules about every innovation has to be more environmentally friendly than the product it replaces.
• NEC is active in very many industries, from space engineering to construction.

Mini HYDRAstor

The Mini HYDRAstor is the same as a regular HYDRAstor, but the storage and accelerator nodes are both in one chassis.

• GA’d June 30th 2010
• Same code base as it’s bigger brother.

HYDRAstor

• Two tier approach
  • Accelerator nodes (XEON based arch)
  • Storage nodes
• Unrestricted expandability.
• Each tier is independently expandable, so we have scale-up and scale-out.
• Each node can be independently replaced by new equipment if available, without disruption. NEC calls this Adaptive Grid Storage.
• All the modern techniques are available, like Thin provisioning (not sure about zero-page-reclaims though.
• All nodes are interconnected through 1Gbs interfaces. The SN have 4 ports each.
• Global Inline De-Dupe
• Application aware De-Dupe (based on profiles selected during creation of the share). The SN knows what part of the incoming data is user data and what part is meta data. After separating this data, the dedupe only is done on the user data. This enables better De-Dupe ratios.
• HYDRAstor comes with 1 Storage Node for every 2 Accelerator Nodes.
• Filesystem size is 256PB by default.
• Shares can be on NFS or CIFS (SMB1 though).
• CIFS and NFS mountpoints can not be the same filesystem. They have to be separate filesystems.
• WORM FS (HYDRAlock)
• Licencing is based on the feature and the number of accelerator nodes it is configured on.
• RepliGrid – allows for data to be sent offsite for DR purposes (aSYNC).
• HYDRAstor architecture features: 1) support multiple generations of node hardware in one grid.
• HYDRAstor architecture features: 2) non-disruptively add multiple generations of nodes to existing grid.
• HYDRAstor architecture features: 3) capacity automatically discovered WITHOUT provisioning.
• HYDRAstor architecture features: 4) existing data auto load balanced across nodes.
• HYDRAstor architecture features: 5) data resiliency level automatically maintained via Distributed Resilient Data.
• There are no virus scanning features.
• Many-to-one and many-to-many replication (at this time, there is a 16:1 fan-in ration).
• In-flight encryption.
• HYDRAstor-Netbackup OpenStorage Integration.
  • Dynamic IO        (free of charge)
  • Optimized Copy
• Snapshot and replication are not charged extra.

Storage Nodes

• Automated aggregations of scattered fragments.
• storage nodes scale from 10TB to 20PB with all storage managed as 1 logical pool of capacity.

Accelerator Nodes

• 20 AN’s achieve 10GB/s
• 500MB/s per NEC Accelerator Node
• 36TB/Hour on a 4 rack system (20 AN / 40 SN config)
• An 11-rack HYDRAstor delivers over 25 GB per second or 90 TB per hour of throughput.
• Takes care of the chunk based De-Duped replication, based on a filesystem level.

Erasure Coding

• User determines the level of resiliency. You can enable protection to up to 6 disk failures.
• Data chunks are variable in size.
• No RAID, therefore no penalty in RAID rebuilds after failing disks.
• Chunk is broken down in fragments, this technique is called Distributed Resilient Data (DRD).
• Recovery of failed disks is always performed on the remaining storage, and is not postponed until the failed disk is replaced.
• Minimum of 9 chunks are required to reconstruct the fragments into user data. The first 9 received chunks are used, therefore not being impacted by busy SN having a high latency/delay.

Availability

• Europe / EMEA region is not yet on the roadmap.
• Just Japan and Northern US.

Pricing

• Smallest config :
  • $40,000.- Starting at 4TB (listprice)
  • $25,000.- per additional 4TB (max 12TB raw cap)
• Largest config :
  • Minimum $120,000.-

What’s missing

• Cloud-> add REST interface to enable cloud services.
• SMB2.0 or higher would greatly enhance performance for CIFS enabled backup applications or end-users storing files on it directly.
• Virus Scanning (for customers that would like to store files on it directly)
• For long term archiving a MAID implementation would be the GREEN option.
• EMEA availability.

References

• Reference customers get support costs discount.
• Case Studies

The TechFieldDay success must be huge, when a company decides to use TFD as a platform to announce it’s launch. The delegates are all witnessing this launch. It is a great experience to be able to be part of an event like TFD, especially when you also get to be part of a new companies launch.

The introduction

The new company is called NimbleStorage, was founded in 2008 and is based in San Jose.  Nimble Storage offers a hybrid of flash and SATA storage array. The 3U high box services iSCSI storage and has a fixed size, no scale-up. Nimble Storage claims to achieve 60% cost reduction than existing solutions. Nimble’s storage architecture is “log-structured file system” which was created by Mendel Rosenblum (VMware founder).

• Varun Mentha (CEO & Co-Founder) kicks off by introducing his crew.
• Umesh Maheshwari (CTO & Co-Founder and filesystem expert)
• Dan Leary (VP Marketing)
• Ajay (Former Netapp)

Nimble Storage will be selling through VAR channels exclusively. At first they will be selling in the US only, but expansion to Europe will be in the works for 2011.

The technology

• (MLC) Flash and Low-cost High-capacity SATA disks iSCSI based storage targeted at the mid-sized enterprises.
• Point in time snapshot primary Replication based DR
• Capacity optimized snapshots in stead of traditional backup to eliminate backup windows.
• Listpricing < $3/GB
• Cache Accelerated Sequential Layout (CASL Highlists)
  • LZ-ish based inline compression reduces data 2-4x (no dedupe)
  • Flash caters to high-performance for all active data
  • SATA disk cost-effectively stores all primary data and 90 days worth of snapshots
  • WAN efficient offsite replication
• Application aware snapshots/backups (Microsoft VSS and VMware integration)
• Nimble Storage says they are 35x time more space efficient than leading vendors in this market (eg:Dell/Equalogic)
• Different retentions periods for local and remote data
• Bi-directional replication
  • System many-to-one replication.
  • Volume is one-to-one replication. This means many systems can replicate to one, but a single volume only has a single replication relationship.
• Rapid fail-over between sites (including flexible iqn identities)
• Version 1.0 is not cascaded replication, but it will be there in future releases
• Application templates
  • Predifined application aware storage and data protection configuration
  • LUN Blocksizes are matched to the application
  • LUN Caching is matched to the application
• Zero-copy hypervisor integrated cloning (included in the package)
• Web based GUI, and SSH based full featured CLI interface
• Full autosupport feature built-in (Real-Time Phone Home Support)
• MPIO is used for fail-over, no network based LACP

The flash storage is used as an intelligent cache that holds all the active data. What is active data is determined by the use frequency (and more). The cache is indexed. All data is written to SATA disks, so the flash disks are really only used as cache. All incoming data is compressed inline. Due to compression, the actual blocksize of the written data can vary. Because all the data is written sequentially to the SATA disks, the various blocksizes pose no real issue, and they are all supported. This also enables an application specific blocksize. By using templates in the definition for volumes, you can match the blocksize to match the blocksize to for example an Microsoft Exchange database volume, and another volume for it’s logs where both have different blocksizes.

All volumes can have their own snapshot schedules, or they can be grouped together in Protection Sets, which can be considered consistency groups (volume groups, not hosts groups).

You might be affraid that the SATA disk would provide bad performance, but the sequential reads and writes are actually something SATA disks can do pretty well. So this performance risk is mitigated by the compression-sequential-write (full blocks) part of the array’s code.

The flash cache is made up out of SSD’s, and are hot-replaceable and are shared between the controllers. All data is already on disk and therefore there is no need for any means of protection for these cache disks.

Products

• 3RU Units, large flash layer, multicore Intel Xeon processors
• Comes with 2 x quad GbE NICs
• Everything is redundant (controllers are active/passive)
• All drives are hot swapable
• peer-to-peer clustering
  • CS220: 9TB primary + 108TB backup
  • CS240: 18TB primary + 216TB backup ($99.900.-)
    • 1.3TB flash capacity based on 2x compression
    • 12 x 2TB disks (1x hot-spare/2x parity)
• Annual maintenance between $4000 and $6000 .

Roadmap

Although NimbleStorage wasn’t going to give us any formal roadmap intel at the moment, the following features are surely being introduced in upcoming upgrades/updates.

• Cascaded replication                        
• VMware SRM integration
• 10GbE NICs
• V1.1 Scale out, LUN’s can be striped across arrays.
• Role based access.
• QoS for replication sessions (including time of day based policies)
• SNMP alerting
• FCoE

Overall impression

Curtis W. Preston asked (and I was pondering on it) “why not NAS?”. The midsize customer segment doesn’t use a lot of NFS and for CIFS, the tend to use a regular Windows file server with (iSCSI) block storage from the SAN. The context to me was actually that it was not a need-to-have feature for the product launch. There might be a different view on the file services option in the future.

I am very impressed by these guys. They bring a ton of experience into the company which is transfered into their products. It is clear the products are functional and quite complete, but a couple of relevant features are still missing. The relevance is dependent on the size and level of operations of the client looking into this product. Smaller customer might not be depending on SNMP alerting or 10GbE interfaces. These features and the aforementioned features are sure to be introduced shortly after this launch.

The Nimble Storage guys presenting at Tech Field Day are brave in my book. They come in to present a new company and new products to a group of tech guys that could give them a really hard time, but they stood tall, and gave us a very great presentation. They definitely believe in their product, and at the same time respect their competition.

I will be watching them closely.

Resources

• Nimble Storage
• In The News (Gestalt IT)
• Twitter @NimbleStorage
• Facebook