Why virtualization is the best for improved scalability, performance

Last time, we discussed “white space,” unused pool of assets, which when it comes to storage, means idle disk storage. It is what we used to term “headroom” – that extra mass storage we keep on hand for when our users unexpectedly feel an urge to store a few extra terabytes of data.

In this context, virtualization technology allows users to see a single – and perhaps limitless – supply of storage.  In the simplest terms, it does at the enterprise level what the C: drive on your desktop would do if its capacity extended out to the horizon.

Virtualization is the best way yet to achieve improved scalability, capacity utilization, availability and performance, the usual values we look for from this technology.  Additionally, it offers the advantages of superior storage administrator productivity (due to centralized storage management) and provides excellent opportunities for companies to both consolidate their storage assets and improve their approach to disaster recovery. 

Virtualization can be server-based, array-based and storage-area network (SAN) based.  One size does not fit all, and the one you want depends on your requirements.

Need to have a single host (or a single cluster of hosts) address multiple arrays?  Look to a server-based implementation, where the intelligence resides on the host and virtualized capacity can be extended across all the contents of multiple heterogeneous arrays.

Array-based virtualization works best when multiple hosts need to address a single array.  With the intelligence on the array controller, multiple hosts can access a data pool that scales to the limit of the array’s capacity.

Both the above examples represent one-to-many implementations.  Utility computing however, must assume multiple many-to-many relationships in order optimize use of its various components: many users sharing the same resource, many resources providing services to multiple processes, and so forth.  If you anticipate a topology with multiple hosts addressing multiple arrays, the virtualization intelligence must reside in the storage network.  Assume this to be a requisite of any storage utility.

FalconStor’s IPStor, HP’s CASA, Sun’s StorEdge 6900 and Veritas’ Volume Manager are examples of products that have successfully implemented storage virtualization technology across a heterogeneous mix of products. Additionally, IBM earlier this month announced its first homegrown virtualization offerings, the TotalStorage SAN Volume Controller and the SANIntegration Server. 

Now comes the interesting part. The larger the enterprise, the greater the likelihood that any SAN worth investing in will be a heterogeneous environment.  Heterogeneity means more than mixing and matching different vendor’s arrays, however; it can also mean mixing and matching disk technology… so expect to find Fibre Channel, SCSI and ATA on your SANs in the near term.

Yes, even in the enterprise IT shop ATA likely has a roll to play.  Clearly it will not be used for primary storage of business-critical information, but as almost all of us know, all data in the data center is not of equal value.  ATA disks will prove to be a cost-efficient mechanism for primary storage of our relatively unimportant data, an inexpensive way for managers to offer their users near-line capacity, and a staging area for use in many implementations of improved backup services.  Furthermore, as storage resource management software increases in intelligence, expect to see data allocated to devices within the storage pool automatically, with the data distributed according to its value to the company.

Most vendors understand this (although one significant player clearly does not) and at least one major storage provider, EMC, has even introduced a box that accommodates both Fibre Channel and ATA drives within the same chassis.

This means that the single management console that provides your view of the storage pool will soon offer qualitative as well as quantitative management capabilities.  All you will have to do is establish policies that assign higher-value services and storage to higher-value data, and the system will do the rest. 

All this may be coming to a competitor near you soon.  Probably you should look at it as well.