Two Enterprise Experiences with All Flash and Hybrid Configurations

Enterprise customers require storage systems that can consolidate a heterogeneous mix of OS platforms and applications and still deliver consistent high performance, non-stop always on operations with the highest level of failover automation.

September 28, 2015

Enterprise customers require storage systems that can consolidate a heterogeneous mix of OS platforms and applications and still deliver consistent high performance, non-stop always on operations with the highest level of failover automation.

Midrange storage arrays cannot deliver on these requirements due to their active/passive dual controller architecture and their limited support for remote replication.

There are application use cases for lower cost midrange arrays where heterogeneous workloads and non-stop operations and remote replication are not required.  Adding flash drives to a midrange storage array will certainly increase performance but that will not convert it into an enterprise storage array.

When one refers to an All Flash Array (AFA) you need to distinguish between midrange and enterprise AFAs.

Hitachi Enterprise Flash Arrays

Hitachi Data Systems builds a family of enterprise arrays, from a low cost 2U “VSP G200” that can virtualize external arrays with no internal disks, to a “VSP G1000” that can support 255 PB of internal and external storage.

Our VSP G200 to G1000 enterprise arrays have an internal switch architecture and a global cache that enables I/O load balancing with partitioning for protection against the effect of noisy neighbors.

The same code that runs in all the VSPs has been optimized for the latencies of flash, just like other flash array vendors. However, Hitachi also optimized the code in the flash module, which increases endurance and performance by removing the flash housekeeping impact out of the I/O path in the device.

As a result an all flash G1000 was able to max out the SPC1 benchmark and set a record of 2 million IOPs with less than 1 ms response time with only 64 flash devices.

These SPC1 performance results are outstanding but what does it mean in real enterprise environments?

At the recent 2015 Flash Summit in Santa Clara, Walter Amsler, our senior director of global technology planning, presented the results of two enterprise customers, The Coop Group with an all flash G1000 and Swiss Re with a Hybrid G1000

VSP G1000 All Flash Array at COOP

The COOP Group is the leading retail company in Switzerland with a total sales revenue of $30 B in 2014. Their core business runs on SAP with SAP HANA for real time business intelligence. With thousands of outlets and production sites, they depend on long distance (120km) Asynchronous Remote Replication for disaster recovery.

The COOP installed an All Flash Array G1000 with 134 x 3.2TB flash modules (432 TB raw capacity) and array based asynchronous replication. The G1000 used Hitachi Universal Replicator for asynchronous replication over 120 km for any server, OS and application. This removed the software and hardware costs that would have been required for host-based replication.

With asynchronous replication they maintained an average response time well below 1 ms for the entire storage subsystem with a daily sequential throughput peak of 6 GB/second. Even in a highly consolidated environment with a variety of OS (AIX, VMware, Linux, Solaris) and a diversity of applications (BI, ERP, Analytics), their consistent low response times indicated subsystem scalability without impact to low latency performance.

The all-flash G1000 coped with substantial latent demand and eliminated hours of elapsed time for critical applications. Response times were reduced by a factor of 10. Faster shopping cart checkout times on their web shopping application improved online customer satisfaction. Their daily consolidation of distribution center data now completes on time and they can run more analytics to enable better planning and decision-making.

The following end user quote showed how satisfied they were with their enterprise all flash solution:

“Habt Ihr über das Wochenende SAP “frisiert”? Geht alles viel schneller…konnte nicht mal mehr ins Kaffee…Die Auswertungen wahren schon auf dem Bildschirm bevor ich aufstehen konnte…“

“What did you do to SAP over the weekend? Everything is so much faster, I could not even find time to have a cup of coffee, the reports were already on the screen before I could get up from my chair…”

VSP G1000 Hybrid Flash Array at Swiss Re

Swiss Re is a leading, highly diversified, global company with 151 years of experience in providing wholesale re/insurance and risk management solutions. They are a pioneer in insurance-based capital market solutions, combining financial strength and unparalleled expertise for the benefit of their clients.

They are challenged with increasing financial pressures, yearly storage consumption growth of 35%, inefficient point solutions (VDI, File servers, e-discovery, etc.), and inadequate high availability and site failover solutions. They needed to reduce site failover time from 8 hours to less than 1 hour.

They chose a G1000 hybrid flash array with a mix of flash, SAS, and nearline SAS devices with dynamic tiering and active tiering.

With only 5% of the total capacity in flash, measurements showed that flash accounted for 82 percent of the IOPs. This was accomplished through dynamic tiering, which automatically moves the most active data to the fastest tier for maximum performance and the least active to lower cost tiers. Another enhancement that has been added to the VSP is Active Flash, which can detect sudden activity in real-time and move data to the flash tier within seconds to sub-second cycle time.

These features working together provide the performance and throughput of an all-flash array for about 82% of the IOPs at a much lower TCO than an all flash G1000.

In order to reduce their site failover time to less than an hour, they used the “Global Active Device feature” of the VSP G1000.

With this feature, applications access data that is virtualized across two G1000’s separated by 20 km. With this active/active configuration, an application can continue to run when one storage systems is unavailable. This eliminates the delays associated with conventional active/passive replication recovery and delivers a zero RTO and RPO. In order to support GAD, an I/O must be written synchronously to both storage systems, so some overhead is required. But even with the GAD overhead and the dynamic tiering overhead, the subsystem average response times were 2-4 ms with < 2ms for key applications. This hybrid approach provided scalable performance for dynamic workloads of hundreds of diverse applications with only 5% of the capacity residing on Flash storage.

Enterprise All Flash or Hybrid Arrays – Your Choice

If you have enterprise requirements choose an enterprise array first before you decide on all flash or hybrid flash. The difference between an all-flash enterprise array and a hybrid enterprise array is that all your data will get flash performance whether they need it or not. On the one hand, I have heard a customer say that he prefers an all flash solution even if it costs more because it simply eliminates all his performance problems and he doesn’t have to worry about those 2:00 AM calls about slow response times. Another customer told me that he bought an all-flash array, and he ended up with an all-flash JBOD, because most of his application data was no longer active.

With Hitachi Enterprise flash arrays you have a choice, and you can non-disruptively change the mix when your workload changes. The smaller models of the VSP also lower the entry cost for enterprise flash arrays. Since our enterprise arrays can also virtualize external storage, you can attach your existing non-flash, flash or hybrid arrays and enjoy all the latest enterprise capabilities of a Hitachi array.


Watch video

In the same category