FalconStor flying high
Much more capacity, availability and integration.
By Chris Mellor | Published: 23:00, 20 June 2007
FalconStor is run by ReiJane Huai, a bespectacled, clear-sighted CEO who is technically very knowledgeable and might as well be FalconStor's chief technologist or chief engineer, he is that adept and au fait with the software technology layers based on IPStor, the basic FalconStor platform.
FalconStor is a publicly-traded company an has been since 2001. It has been profitable since 2004. All sales are indirect with OEM sales accounting for 40 percent of its business and value-added reseller (VAR) sales the remaining 60 percent.
FalconStor provides two products for data protection: a VTL; and CDP (Continuous Data Protection). Huai said: "VTL sales represent half of our business," and business is good: "In Q1 07 we grew 70 percent over Q1 06."
A lot of significant storage suppliers OEM FalconStor's virtual tape library software such as EMC, Sun and Hitachi Data Systems (HDS), whose customers for it include General Electric and Radio Shack. A Diligent VTL was replaced by the Hitachi/FalconStor one at Radio Shack for speed reasons.
Copan is another FalconStor OEM. HP OEMs FalconStor's iSCSI software but not its VTL; HP gets that capability from Sepaton.
Concerning Sun, Huai said: "We have a very good relationship. We work with Sun on a daily basis in Europe. Huai said his company's VTL software is chosen because: "We hide the complexity behind the VTL."
Together EMC and Sun account for ten percent of FalconStor sales.
VTL and tape and discovery
Huai's of backup and archive is that data travels from a primary store to a VTL and then on to tape. He says: "Backup and archive are the same. The final stage is tape." He is dismissive of the idea of having additional data stores for legal discovery reasons: "You don't need a separate discovery database."
The reasoning here is that FalconStor's VTL can index the data it holds and so obviate the need for a discovey database: "We can have variable granularity. We can index the words in a file so it's quite, quite suitable for discovery."
The traditional VTL is a VTL appliance mated to a modular drive array but this is changing. Huai said: "VTLs are becoming bigger and faster." The traditional VTL is a VTL appliance mated to a modular drive array but this is changing. What's happening is that "the VTL is getting deeper into the data centre. Now a VTL exists using enterprise-class monolithic storage," referring to EMC's introduction of a VTL using Symmetrix disks.
We talked about de-duplication, the removal of repeated sub-file level elements from data stored on a VTL to reduce the space it takes up, and so extend the period of time for which a VTL could hold backup data. For FalconStor de-duplication is to be done after the data has been acquired by a VTL: "If you don't post-process de-dupe you can't have speed."
De-duplication is a CPU-intensive activity and doing it as backup data is fed to a VTL causes the ingestion speed to slow markedly and so lengthen the time needed for the backup. Huai is pretty scathing about ingestion time de-dupe: "Acquisition time de-dupe is propaganda. It will not be valid."
Huai predicts that Copan is going to use FalconStor de-dupe technology: "In the next few months Copan will introduce a FalconStor de-dupe capability for its VTL."
The cluster story
Huai says: "It's all about recovery now. Our business is recovery. We provide it; we facilitate it," meaning: "We facilitate recovery by increased reliability and management, and we optimise the storage facility with de-duplication."
"We'll continue to improve our VTL, giving it even greater capacity." He discussed the idea of clustering. "Fibre Channel and iSCSI exist today for a host to talk to a VTL and send it data. We will cluster the VTL nodes easily and we'll introduce a faster transport. Infiniband at 20GB/sec is faster than Fibre Channel at 4GB/sec. We have an Infiniband partner and we'll incorporate Infiniband as the main transport."
Huai said: "Architecturally a cluster is a cluster is a cluster." FalconStor will implement its clustered VTL on separate physical servers, connected by Infiniband, or on virtual server nodes under VMware and Solaris.
A physical cluster will be up to eight server nodes with sixteen node support in the future. In the VMware implementation there will be a number of virtual VTL servers, four or more depending upon the I/O bandwidth available. Under Solaris the VTL nodes will run in four or more Solaris containers, again depending upon the I/O strength of the machine.
With clustering extra nodes can be added if a VTL's workload increases. Clustering will also increase availability as the workload of a failed node can be spread across the remaining nodes.
Better tape library integration and LTO-4
There will also be greater integration of the VTL product with tape libraries. Huai said: "We'll scale connectivity and capacity, and services will have a tighter integration with the VTL. LTO-4 will become the de facto standard for enterprise tape libraries. The VTL's virtual tape drives will move to LTO-4. A VTL could have up to 64 drives, but doing 64 backups at once means the virtual tape will be under-utilised. "
"The VTL manages jobs to increase capacity per slot (virtual drive). We can stack the (virtual) tapes and, in effect, compress them onto two physical tapes." When hosts back up to a virtual LTO-4 drive and there are up to 64 such drives the individual virtual cartridges on the virtual drives don't fill up. FalconStor's VTL consolidates this content and puts in a packed state - 'stacked' in mainframe terminology - on one or two real tape cartridges which are then fully utilised. This avoids having lots of partially filled tape cartridges and saves wasted tape cartridge space.
With LTO-4 reels able to hold 800GB of raw data this feature could avoid lots and lots of empty space on LTO-4 cartridges.
FalconStor's Continuous data protection (CDP)
This is a parallel stack to the VTL. By using CDP, Huai says, you don't need to do daily incremental backups at all. You snapshot primary storage to disk and completely eliminate the backup window.
The image can be offered to NetBackup Server over a storage area network (SAN) and it will backup faster than over a LAN, four times faster according to Haui.
FalconStor aims to increase VTL capacity and speed through de-duplication and clustering, and integration with tape libraries through tape stacking. Better indexing will enable a VTL to provide discovery database functions avoiding the need for a separate discovery filestore purchase. The CDP product will also be extended.
There is no reason not to suppose that FalconStor will continue to fly high and extend its grip on the market. Its software will become more capable and more complex, thus increasing the barriers to entry of any FalconStor OEM that was thinking about possibly rolling its own VTL software.