How Copy Program Products Benefit from Compression Accelerator Module

By Dang Luong posted 15 days ago

  

Adaptive Data Reduction (ADR) is a feature on the Hitachi storage systems that reduces the size of data to store, effectively increasing the usable capacity on the array. There are two data reduction methods: 1) Compression and 2) Deduplication and Compression. These methods can be enabled heterogeneously on a per volume basis.

Before the introduction of the VSP 5200 and VSP 5600 models, ADR was performed by software in the storage system controllers.

New technology was introduced on the VSP 5200/5600 platforms to move this operation to dedicated hardware: Compression Accelerator Module. The new module offloads compression processing from the storage system controllers. This frees up CPU cycles, which enables the storage system to deliver higher IOPS and/or lower latency.

Another important benefit of the Compression Accelerator Module is that it performs compression at 32 KB data units. Compression done by the storage controllers operates at 8 KB data units. As a result, host I/O with larger block size data benefits more from the new compression feature than host I/O with smaller block size data. Of course, all I/O profiles benefit from the offloading to some degree!

And it is not only host I/O that improves with the larger 32 KB compression data unit. Replication performance is also boosted for the same reason. This includes Initial Copy and Resync with ShadowImage, Volume Migration, TrueCopy, Universal Replicator, and Global-Active Device. These copy operations work at 256 KB to 512 KB data units so there can be significant performance improvements when compared to compression done with storage controllers. Note that Hitachi Thin Image was not included above because as a snapshotting technology, its initial copy operation does not actually copy data.

Let’s take a look at a study case focusing on GAD Resync.

First, some background details: a resync operation involves replicating data that has been added or modified while replication was suspended from the primary storage system to the secondary storage system, with the goal of making the two copies exactly the same (AKA consistent). More often than not during such resync operations, data on the primary storage system is also actively being read and written to by application servers. At a minimum, the primary storage system is servicing host I/O, replicating data and performing ADR tasks. The secondary storage system is receiving replication data and compressing and/or deduplicating this data. All of these demands make Resync a taxing operation on the storage system’s processors. That is why it makes for a good study case to demonstrate the benefits of the Compression Accelerator Module.

The following charts compare the response time from the application servers during resync between VSP 5500-2N with compression done by the controllers versus VSP 5600-2N with compression done by Compression Accelerator Modules. Note that lower response time is better. One can see the obvious improvement in response time on the VSP 5600-2N: lower line overall and smaller variance in the line. In fact, the max response time decrease by 36% on the VSP 5600-2N during Resync, thanks to the Compression Accelerator Modules.

Chart of VSP Response Time (ms)

The table below breaks down the specific response times observed on the two platforms.

Table of VSP Response Times

Final data point: the storage processors’ peak utilization decreased by 12% on the primary storage system and 43% on the secondary storage system when comparing VSP 5600-2N to VSP 5500-2N. Why is there such a large discrepancy between the two numbers? The primary storage system was servicing application I/O of about 9 KB units. In contrast, the secondary storage system was servicing resync I/O (from the primary storage system) of 256 KB units. Since the Compression Accelerator Module operates on 32 KB blocks, larger data units, irrespective if they are host I/O or replication I/O, enjoy a bigger boost.

The end result is that with the additional CPU cycles, the VSP 5200/5600 can deliver more IOPS and at lower latency!

For more information on the VSP 5200/5600 platform, go to https://www.hitachivantara.com/en-us/products/storage/flash-storage/enterprise/vsp-5000-series.html


#HitachiVirtualStoragePlatformVSP
#FlashStorage
#ThoughtLeadership
#Blog
0 comments
19 views

Permalink