And then suddenly, I noticed something incredibly fishy.  In addition to being a CLARiiON only solution, RecoverPoint/SE also only supports Windows. (My reference material was outdated… RPSE apparently supports most major OSes now, my apologies)

It make sense that SMB shops used to run only Windows and CLARiiONs, so the “Value Edition” of RecoverPoint supports that particular combination.  But SMB shops nowadays are venturing out to FOSS solutions such as LAMP based Alfresco to save some $. But if that translates to having to fork over a lot more money for the RecoverPoint “Deluxe Version”, it’d probably make SMBs look into some other storage vendors with more FOSS friendly price quotes. (However, RPSE does only limit you to CLARiiON-only storage, so for those wanting to save money by using less expensive boxes from a competing vendor, EMC will make you pony up for the full veresion of RP) After all, not everyone needs a Mercedes C class when a Toyota/Honda/Mazda works just fine   Remember, Linux is your friend.

• NuView StorageX (acquired by Brocade, OEM’ed to NetApp and IBM, EOL’ed this year)
• Acopia ARX (acquired by F5)
• Neopath File Director (Acquired by Cisco, EOL’ed soon after)
• Rainfinity(?) (Acquired by EMC, sales force seems to have stopped pushing it, hinting the pending FAST for Celerra release)

Out of the above 4 major players, 1 survived, (1.5 if counting Rainfinity)

What happened ?

NAS virtualization solves the scalability problem of NAS devices and provides an easy way to migrate files and/or volumes between dissimilar NAS devices. Most NAS devices back then were simple X86 file servers running code optimized for file serving. Although highly tuned, most of these NAS devices only scaled up to a active/passive fail-over cluster for HA, which improves the availability, but does nothing for scalability. The 32bit filesystems used in these NAS devices limited the volume size to ~ 16TB and can only hold less than a hundred K’s of inodes (file/directories). While the limitation is adequate for most organizations, it is a nightmare for those with large data sets (ie. HPC), or large number of small files (ie. Web based image serving sites). NAS virtualization broke the limitation by creating virtual volumes that span multiple NAS systems, effectively raising the capacity/inode ceiling.

The migration service lets customers move volumes around  NAS devices from same or different vendors. This alleviates the pain of running long robocopy or rsync sessions and reduces the cutover window.  Some of the migration services also comes with policy based file movements that moves  files around automatically  based on user defined terms (such as file age, size, last access time, file type) between faster ($$) tier and slower ($) tier.

(The net result is a relatively easy way to implement something like looks like distributed filesystem of the early late 90s … think IBM DCE/DFS, AFS and HPSS, minus the tape part and all the nerd quotient mandatory for implementation/support)

NAS virtualization hype seems to have completely deflated, considering that major NAS players are now selling devices with either built-in single name-space capability (NetApp), online migration mechanism (EMC, NetApp) and massive scalability (NetApp, Pillar). The only buyers of NAS virtualization would be organizations that have not standardize their NAS on a single platform and wants to move files around easily – A pretty small crowd.

However, NAS virtualization still has a chance, and this life line is Open Source – Cloud Storage.
Similar to the way SAN virtualization controllers (IBM SVC,  HDS USP-V) improves the performance of the physical arrays by adding more cache and spreading out volumes across more spindles, NAS virtaulization device can create high performance virtual NAS devices by distributing a virtual volume onto a number of inexpensive Open Source file servers. The NAS virtualization in this sense then provides replica service to allow users to access their files from any of the replicated filesets that is either less busy or closer to the user. Since the filesets are replicated can also exist in multiple sites, all the data is always available – instant BC/DR function.  Imagine – an easy and inexpensive way to build your own gfs (google, not RH’s). Slapping in a Amazon S3 or EMC Atmos interface, you now have both internal and public cloud all in a nifty device.

Maybe it’s time for a major linux vendor to venture into storage business ?

Curiously, HDS is the only big 3 vendors not listed, mainly because I’ve NEVER been able to get hold of their sales force.

Perhaps it’s because I’ve lately been concentrating on the public sector, where HDS had not been paying too much attention. Whenever I tried to call the fed sales number on the HDS site, it either goes directly to a voicemail box or answered by some receptionist who redirects me to a voicemail box. Either case, I’ve never gotten call-backs.

Initially I just had some unlucky time reaching HDS sales, but after talking to a few storage pros in the sector, I realized the mutual concerns about HDS’s responsiveness on both sales and tech support (post-sales) for the public sector.

HDS.. please wake up! EMC had already stolen your market dominance. Seeing IBM is completely losing the battle on the high end monolith storage devices, and that not many tier-1 buyer taking NetApp’s pseudo high-end FAS6080 seriously, I’m afraid if this trend continues, I can only recommend EMC SYMMs to my customers with Tier-1 needs.
I’ve heard so much about HDS’ innovation and technology (not to mention the user friendliness of their boxens, compared to the SYMMs) and I really hate to see the aforementioned scenario from happening.

Seagate had just made an announcement (KB207931) acknowledging the problem with firmware and is now providing firmware fixes.

Owners of recent Seagate drives, please download the software to verify if your drive is on the affected firmware list. If so, email Seagate for firmware upgrade instructions before it’s too late.

For those with 7200.11 drives that just died, contact Seagate for free data recovery service.

Too bad my bad drive had already been RMA’ed a month ago :-/ no free data recovery service.

Here are some of my early observations (rants)

1. DocApps had been replaced by DAR (Documentum ARchive). Old DocAPPs created by Application Builder can be migrated to DAR by using Composer

2. TaskSpace 6.5 + Composer 6.5 = Broken
I cannot, for the life of me, figure out how to import a new  Taskspace applications from repository into Composer 6.5 without it yelling at me about some dm_ type issues.  It seems to be a bug that slipped through QA

3. Even more virtualization lameness for index server
In addition to VMWare ESX guests, now AIX LPARs had also been added to the list of forbidden platform for the FAST index server.
Hello ?! I thought EMC *OWNS* VMWare now ?! Not allowing LPARs will upset a lot of IBM shops (Especially those unfortunate ones using GPFS)

4. 64bit-ness is arriving, at last!
Some of the WDK based applications now says they also support running in 64bit native mode. It is a late, but welcomed change. (However, the thing that really mattered – Content Server, is still 32bit only)

5. JBOSS
To my surprise, JBOSS had replaced (Oracle) Bea Weblogic as the embedded J2EE App server. In my first few lab installs, JBOSS seems to reduce the memory usage somewhat. It may be possible to create a usable Documentum installation including Content Server, Index Server, and Webtop/DA on a laptop like the pre 6.0 days using Tomcat.  (I have yet to try laptop install myself, but judging from the fact that my 6.5 content server install doesn’t even use SWAP while the same VM was swapping with no load under 6.0)

It doesn’t support creating new categories offline nor does it have spell check capabilities, but it is a very good start .

For the uninitiated -
Paid RH users get their OS updates/patches through the Red Hat Network (RHN), which is similar to Microsoft windows update website but with more management features that allows admins to manage various package channels on multiple systems. (Windows update is only designed to patch the system directly connected to it and nothing else, unless you implement your own SMS/SCCM services).

In a way, Satellite server is pretty similar to SCCM, both are installed locally on your own network, and you synchronize the patch contents with the upstream provider; both polls hardware inventory and uses some sorts of database backend to store all the client/patch dependency information(Oracle for Sat srv, MSSQL for sCCM). Up until now, they were both closed source.

One thing that irritated me the most during my prior experience with Sat server was its closed source-ness.
The backend database was oracle and a lot of magic happens in there. After knowing about the database, I immediately set out to write scripts that can pull machine information out of the database. After talking to RH support a few times, I was told they would not give out specifics of the schema and they’re likely to change frequently and advised me not to interact directly with the backend db. That was a major bummer. The Sat server was basically a hugely expensive black box. You know it’s running but you can’t see inside it.

After I figured out how to centrally admin all the kickstart scripts in Satellite server, I migrated all the previously hand maintained ks files into it. Everything was fine for a while, until one day, the anaconda (redhat installer) was updated to support more commands such as a flag to suppress HBA driver from loading. It was very imporant feature because it prevents anaconda from prodding SAN devices when you don’t want it to. (For example, re-imaging a GFS cluster node without having to disconnect the FC cable or temporarily re-zone the SAN). Unfortunately the Satellite server did not support that particular flag and I was told the only way to do it in Sat server would require going into the oracle db and add things in the schema. (I was also given the warning that once I do that, I’d void the support ..eeeeek!)

Needless to say, I gave up on that idea and moved all the GFS cluster kickstart off the satellite server into where it came from, so I can do hands-free cluster node reimaging. (back to square one)

The opening of the satellite server code is a very good thing for RH. It lets CentOS users experience what paid users get and may convince those who depends heavily on CentOS to migrate to a pay model, it also allows industrious SA/System engineers to see what’s in the black box and grant them the ability to do more amazing things with it.

Good job, RH!

(And btw, I’ve been told the oracle server will be replaced by Postgres soon… back in FOSS 2006! Now that you’ve done opening the code, perhaps the OSS community can help RH speed up the divorcing process with oracle )

I really didn’t pay much attention to them since I got sidetracked by other projects that took me away from the storage group.  After I resumed working with the storage folks, I found out the management interfaces are now dropping out of network sporadically, and the only way to get it back online would be to cycle the disk controller where the management interface is .. OUCH!
Obviously the clients, especially the management levels aren’t happy ( Those who did not support venturing out of EMC solutions and the EMC onsite reps were thrilled … )

Our IBM rep had recommended firmware upgrades, setting the interface to 10 auto neg off , half/full duplex, but none of these had helped.

I’ve seen some pretty big DS4000 series installation at some other sites but I’ve never seen any controller with management interface that drops dead so easily, even after multiple attempts of firmware upgrades. This should never happen on a storage system, regardless whether it’s enterprise class or SMB class!  I can tolerate my Linksys router at home going belly-up every now and then, but a storage controller should never do that.

the Zip was great during its heydays when its only competitor was the Panasonic SuperDisk that came too late. Even though the SuperDrive has a great potential as a drop-in replacement for existing floppy drives because it was bootable and back-ward compatible to old floppies, its 3 year lag was its fatal misfortune.
Nevertheless, both of these products died quickly when inexpensive CD-R/RW became more available.

Following Cisco’s foodstep on acquiring Linksys to get into the consumer market, I guess EMC can use iomega in the same way. Hopefully EMC would be more agressive on leveraging its newly acquired consumer brand than its networking counterpart.
As new computers come with more and more HDD space, and lots of households have more than a couple desktops, where most of the systems did not come in any kinds of protected data, nor usable backup solutions ( try backing up 200G HDD using DVD-R’s ….), EMC/iomega has a potentially HUGE market if it plays the cards right.

(Iomega’s existing portfolio such as REV and external HDDs should be canned, since blueray-R/RW will soon take their place, just like the way CD-R/RW swallowed the market share for Zip and Jaz)

What EMC/iomega really needs to do, is to deliver an expandable Windows Home Server(WHS) alternative that does not eat the precious data for lunch.

A modular, stackable NAS system with 2 drive bays that comes with 2 drives, pre-configured as RAID1 for data protection. This system can run some types of embedded Linux/FreeBSD with all sorts of media servers running.

Once the space runs out in the main unit, an expansion unit wit 2 pre-populated drives can be added by stacking it with the main unit. Upon seeing the new drives, the main unit concatenates the additional raid-1 volume to the existing one, therefore doubling the size of the storage capacity, while still providing data protection.  (Think adding PVs into a VG on Linux)

Meanwhile, the embedded OS would run some sorts of backup server software, say – retrospect ?
and lets backup client store backups to it.

Now the final touch: Leveraging SaaS by using embedded Mozy client, coupled with the $4.95 monthly unlimited plan they have, you have a consumer friendly data protection solution that also provides remote data bunkering.

Will EMC/iomega make this ? I don’t think so, but I sure hope so.