Search
Close this search box.

Another Look at What Is the Best Snapshot Method

A couple of months ago I wrote a blog entry commenting that, based upon the research that DCIG had done to date, allocate-on-write (or redirect-on-write) had emerged as the “best” snapshot method that a midrange array could support. That opinion was formed after looking at snapshot functionality on over 20 midrange arrays and the pros and cons of each snapshot type. But yesterday I had a lengthy conversation with HP 3PAR’s Director of Product Management, Sandeep Singh, that suggested the copy-on-write algorithm used by HP 3PAR’ is as good or better than any other vendor’s allocate-on-write snapshot method.

The reason as to why DCIG initially concluded that allocate-on-write was a better way to do snapshots than copy-on-write essentially boiled down to one simple reason. A copy-on-write snapshot incurs a performance penalty anytime previously written data is changed. In these circumstances, the previously written block of data first has to be read, re-written to another block of storage and then the new data written when the other data previously resided.

Allocate-on-write conversely “freezes” data that was previously written and writes changes to that data elsewhere on other disk. This eliminates the three step “read, rewrite and write” process associated with copy-on-write and reduces snapshots to just one step – a write.

That sounds good on the surface but there are two main downsides associated with allocate-on-write that need to be taken into account.

Since new data is written to other disk, the data can become fragmented over time. This can impact performance later on, especially when it comes to doing sequential reads such as occurs when backups take place.

New data generally has to be written to the same tier of disk as the initial production set of data. This potentially creates a need to move the inactive or frozen data to lower tiers of disk in order to recapture that space on which it is currently residing

Yet even with those downsides associated with allocate-on-write, I still favored allocate-on-write over copy-on-write since allocate-on-write avoided the performance penalty associated with the initial write.

That opinion changed somewhat yesterday after talking at length with Sandeep. I am not saying that I have universally changed that opinion for all midrange array vendor’s implementations of copy-on-write. I have not.

But HP 3PAR has taken an innovative approach to overcome the limitations of copy-on-write snapshots such that HP 3PAR’s implementation of copy-on-write can arguably said to be as good or better than allocate-on-write.

HP 3PAR avoids this copy-on-write performance issue with its wide striping technology which distributes data across the majority of the disk drives in an HP 3PAR system. This becomes important in terms of how HP 3PAR avoids the performance hit normally associated with copy-on-write.

HP 3PAR production volumes will reside on one or more disk drives so when data changes, multiple disks are called upon to read the data that needs to move. But just as important in HP 3PAR’s architecture the disk drive or drives to where the data is copied will likely not be the same disk as where the source data resides. Further, HP 3PAR’s ASIC is called upon to do the calculations associated with the moves in a copy-on-write and it will only move non-duplicate data which further accelerates the copy-on-write snapshot on the system.

So even though HP 3PAR does not eliminate the need for a “read, rewrite and write” nor copy-on-write’s performance penalty, HP 3PAR argues that its wide striping technology coupled with the calculations in its ASIC so thoroughly mitigate the performance overhead associated with copy-on-write that its implementation of copy-on-write is now just as fast as allocate-on-write without allocate-on-write’s downsides.

So has my opinion about allocate-on-write being a better way to do snapshots than copy-on-write changed? I would say no. But I now see and am willing to recognize and admit that there are implementations of copy-on-write such as what HP 3PAR offers that rival and arguably overcome the limitations of allocate-on-write such that organizations should no longer automatically view copy-on-write as being inferior to allocate-on-write.

Share
Share

Click Here to Signup for the DCIG Newsletter!

Categories

DCIG Newsletter Signup

Thank you for your interest in DCIG research and analysis.

Please sign up for the free DCIG Newsletter to have new analysis delivered to your inbox each week.