Automate Space Reclamation on the SAN: Interview with Greg Hayes, Product Manager for Raxco Software
What exactly is SAN space reclamation?
GH:Everyone is familiar with the fact that deleted data can still be read on a PC or server disk and the same is true in a SAN. When a user deletes a file on a guest system Windows deletes the pointers to that file in the index file on the appropriate volume. With Windows 2003/2008 Server nothing tells the SAN that the blocks the file occupies on-disk are no longer needed. The space consumed by the deleted file is now “dead” space since it no longer has a valid file pointer. This space cannot be reused by the storage controller until it is reclaimed to the storage pool.
Don’t the storage vendors provide a mechanism to reclaim this space?
GH: Yes and no. Most modern SANs support one of two mechanisms for reclaiming space: Zero Detection or UNMAP recognition. Zero Detection is built into the storage hardware and reclaims space when zeros are written to the dead clusters. UNMAP is a command the storage recognizes and frees up the space. However, before any space can be reclaimed on the storage the candidate space needs to be identified in the guest systems. The storage vendors do not provide any tools that run on the guest to identify the reclaimable space; this is the gap PerfectStorage fills.
How does PerfectStorage identify and reclaim dead space?
GH: Let’s say you are using PerfectStorage with the Zero Fill option. PerfectStorage runs on each Windows guest system and compares what Windows says is free space to what is actually on the disk. If Windows says a group of clusters are free and on-disk those clusters have data then PerfectStorage will write zeros to that space. If the space on-disk is already a zero, then PerfectStorage will refrain from re-zeroing the space. All of this can be managed from the PerfectStorage console; the reclamation runs online while the servers are up and running and the entire process can be scheduled.
Aren’t there free tools available that reclaim space?
GH: Command line tools like Microsoft’s SDELETE are available but they have some downsides. For example: SDELETE creates a file consuming ALL the free space on the disk. As a result, reclamation can only be done safely with the server offline and it can’t be scheduled. For organizations with hundreds or thousands of virtual machines (VMs) this isn’t really a viable reclamation option. Many SANs only support the UNMAP recognition mode and SDELETE only supports Zero Fill.
How does an organization benefit from a space reclamation program?
GH: There are several benefits. First and foremost there is a savings in storage costs. Enterprise storage can be very expensive, especially Flash storage. The ability to recycle dead space means that storage that was previously inaccessible can now be used again. This allows the enterprise to defer future storage buys.
Secondly, with PerfectStorage the reclamation can be done online with no impact on operations.
PerfectStorage also estimates how much space can be reclaimed; this helps management decide which disks will yield the most space. It’s difficult to do an effective job of storage management if you don’t know how much dead space is out there since its space you can’t use.
Finally, many guest OS VMDKs are thin-provisioned and subject to bloat since they don’t get any smaller when files are deleted. Reclamation provides the opportunity to shrink the VMDK back to a manageable size.
How much space can a site expect to reclaim?
GH: That depends on the user volatility on the disk. A file server with a lot of file create/delete activity would likely have a lot of reclaimable space. Database servers seem to have a lot of reclaimable space too. We had one site that reclaimed 25TB and now makes PerfectStorage part of their routine maintenance.
Do users have a preference for Zero Fill or UNMAP?
GH: This is largely determined by what the storage supports. HP 3PAR and Nutanix do very well using Zero Fill. I believe 3PAR developed Zero Detection so their users are very familiar with that mode. Nutanix has a nightly cleanup process that reclaims any zero-filled clusters. EMC’s XtremIO users have used both modes to reclaim on servers and from VDI environments. The Zero Fill is straightforward and reliable. UNMAP works well with 2003/2008 Server but with 2012 Server you need to be on VMware 6.0 to take advantage of that feature.
Does PerfectStorage work with thin-provisioned disks?
GH: PerfectStorage was specifically designed to work with thin-provisioned disks. Thin provisioning allows disks to grow as data is added but they don’t shrink when data is deleted. PerfectStorage identifies those deleted files as reclaimable space and gets them out of there. If a site has the right versions of VMware and the virtual hardware level they can then shrink the VMDK “on-the-fly.” PerfectStorage also works with data drives that are mapped to a LUN.
What’s the most important thing to know about space reclamation?
GH: Space reclamation needs to be an ongoing disk maintenance effort in a virtual environment just like backup or guest defragmentation. Otherwise, a steadily increasing percentage of your storage will be unavailable and that just wastes space. Automating the reclamation process allows space to be reclaimed on a near-continual basis keeping maximizing the usable space. The ability to schedule and manage the reclamation of an expensive resource like storage is just a good idea. The cost of reclamation has to consider not just the cost of the storage but also the cost of labor and system downtime that accompanies manual reclamation with tools like SDELETE. If you look to manually reclaim 3-4 times a year these costs can really add up.