ApplyIT

VMAX3 Architected for World Record Performance

The flagship EMC VMAX 400K sets a new world record for SPC-2 (bandwidth) performance, delivering more than 30 percent greater performance compared to any other storage array. Achieving a SPC-2 aggregate result of 55,643 SPC-2 MBPS™, an all-flash VMAX 400K leads the pack, beating the nearest competitor, the HP XP7 (the OEM version of the Hitachi G1000 system), by 30%. Not only is the VMAX3 significantly faster, it also provides an attractive price/performance at $33.58 per SPC-2MBPS™.

Architecture Matters

So just how did VMAX3 achieve this record setting SPC-2 performance? Much of its performance and scale is inherent in the impressive VMAX3 architecture. In fact, the original blue print for VMAX3 incorporates several aggressive design-points required to deliver extreme levels of performance. Let’s take a look.

First: Dynamic Virtual Matrix: VMAX3 is designed to handle hyper consolidation and large bandwidth, highly sequential workloads such as data warehousing, decision support, batch processing, and big data analytics. Each VMAX3 director connects to dual, redundant InfiniBand fabrics at 56Gb/s, which provide switched bandwidth for communication among the directors. Within the directors, multi-lane PCIe Gen3 lanes provide full bandwidth to every port on each I/O slot[1]. As average I/O sizes continue to grow, EMC has designed VMAX3 to handle these large bandwidth workloads in addition to the traditional smaller block OLTP workloads.

Second: To use all the bandwidth, VMAX3 has more CPU power and new ways to manage it. Each VMAX3 director has 2 Intel Ivy Bridge CPUs, each providing 6, 8, or 12 CPU cores. VMAX3 systems have up to 384 cores, and hyper-threading provides double that in logical cores. The logical cores are managed in pools, with each pool focused on providing host I/O, disk I/O, remote replication, or other functions. The use of pooling ensures that the CPU power needed for any given workload on a specific set of components is always available. Logical cores can be moved between pools dynamically, providing flexibility to deal with shifts in workload profiles.

Third: VMAX3 continues to provide the cache management that powerful Symmetrix systems are known for. Reads served from cache are faster than from any drive, even flash. VMAX3 does intelligent data placement and pre-fetch to maximize cache read hits. The large cache continues to provide mirrored protection for all write data, allowing the data to de-stage to persistent storage media in the background.

Finally: HYPERMAX OS ties the pieces together to support the industry’s richest data services. For example, customers who are looking for systems that can sustain 50+GB/s of data bandwidth usually have databases that are over 150TB in size, since that would allow them to read the entire database in under an hour. But what about backup and recovery of such a database? As a clear example of the new data services, VMAX3 with ProtectPoint can reduce the restore time for such large databases from days to minutes.

The performance optimized architecture makes VMAX3 the ideal solution for hyper consolidation. Customers can consolidate multiple large data workloads to lower TCO, ease management, and improve reliability, all while delivering industry-leading performance.

What will the future bring?

We expect the future will bring additional world records from VMAX3, not only in the form of SPC results, but also in setting the standard for innovative enterprise data services, advanced replication techniques, and in establishing new ways to stretch VMAX3 data services to the cloud. 

---

[1] Many other architectures use bandwidth over-subscription, whereas VMAX3 has the bandwidth to the I/O cards to drive all ports at full speed simultaneously.

VMAX3: The New Hardware

On July 8, EMC announced the latest new systems in the VMAX family. The new systems are the 100K, the 200K, and the 400K, known collectively as the VMAX³ Family. They are updated versions of the 10K, 20K, and 40K. In addition to all the software enhancements, the systems have all new hardware and packaging. With the new dual engine system bays, the number of bays lines up directly with the names of the systems: the 100K uses 1 bay, the 200K uses up to 2 bays, and the 400K uses up to 4 bays. This post is focused on the hardware and packaging across these new systems.

The VMAX³ Family systems are packaged with one or two engines per system bay, and all bays are 24” wide (19” internal rails). Each bay also contains DAEs holding the disk drives for the engines. These new denser configurations take up much less floor space. For example, the densest drive bay for a 40K holds 400 x 2.5” drives. A new single engine system bay holds up to 720 x 2.5” drives and one engine. A new dual engine system bay holds 480 x 2.5” drives and two engines.

Space Recovery

When migrating data from an old array to a new one, there are many tools to help. EMC VMAX provides a built-in tool called Open Replicator for Symmetrix (ORS), which can push and pull data to and from other arrays. This is a quick review of the benefits one customer has seen using ORS.

Moving Data

ORS makes the VMAX host ports look like a Windows server to other arrays in the SAN (EMC or not). The VMAX needs to be granted zoning access to the original devices on the source array.

Once the VMAX has proper access, the data can be pulled into new devices. The new VMAX devices are usually Virtually Provisioned (thin) devices, as VP devices are much simpler to manage both for capacity and performance reasons.

Recovering Space

As a part of the migration, the VMAX can look for empty areas on the source device. Since VMAX VP volumes allocate space in 768k extents, any aligned 768k of data that is all zeros does not need to be saved on disk, since reads from unallocated space always report zeros. By activating the zero detection with ORS, these extents of all zero data are never written to disk. The use of this relatively small extent size allows for a much finer-grained recovery than some other systems.

Customer Example

I have a customer that has been using ORS for a while to perform migrations. However, they were reluctant to use the zero detection. There was a concern that this was somehow not compatible with all operating systems (it is, always), or that it would otherwise present a risk to their systems. They did some testing in the lab, and became comfortable with the process. Then they tried it in production. Now they can’t get enough of this technology.

Here are their results. For the next 25 servers that they migrated, there was 20.4TB of capacity on the LUNs they were moving. By using the ORS with zero detection, at the end of their move they had 11.7TB of data written on disk – and they had just saved over 8.8TB of capacity that had been all zeros. The graphic here shows the 3 largest systems (each over 2TB of source capacity) and the savings, as well as the overall results. In the end, they averaged over 43% savings by using the zero detection in their data migration.

FAST VP Customer Example

Customers prefer to spend as little as they need to on infrastructure. One of the tricks is to understand what you really need and where you may be able to cut. With the price of disk capacity dropping each year, customers would love to buy the largest/cheapest drives that they can to support their workload. This results in near-term savings when the gear is purchased, and longer term savings from reduced power, cooling, and maintenance bills. But how large can the drives be before they cannot sustain the business workload?

The Problem

Storage tiering is not a new idea. Back in 2003, EMC started a big campaign to help customers save money by tiering their data. It was called ILM - Information Lifecycle Management. The idea was to help customers put the right data on the right storage at the right time, saving them money. There were only two problems: the tiering placements were manual (like this process example), and to place it correctly you needed to understand what the profile for the data was currently and what it would be in the future.

Many customers decided that the manpower was more expensive than just buying additional hardware 'to be safe.' Sure, you could buy a mix of drive types, but buying all the same drives and spreading the workload across all of them is much easier to plan for. And if you did it right, the workload would push the drives hard (to about 70% busy) about the same time you 'ran out' of space (at 75-90% full). Of course, if the workload turns out to have much different needs, you either waste space (the drives get too busy) or waste money (you bought more performance than you need).

Simplified Storage Expansion

Many customers thinking of a new storage array focus on the work required to set up a new server with storage. This is an important item to research.  And some storage vendors can make it a bit simpler than others to connect the host HBA WWNs to the capacity in the array. However, I submit that that there is a much more common storage management challenge, and one that can be made a lot simpler.

The Problem

Customers are constantly adding more capacity to their systems. Application retain more data. New applications are added to servers. Upgrades need more space. It seems that every few months each server is getting some additional storage capacity. So how hard should it be? After all, the servers and storage are already connected.

So let's make this a little more complicated. Many customers are now rolling out VMware ESX clusters of 8 or more servers. Some of these servers have more than 2 HBA ports. So when adding a device to the cluster, there may be 30 or more HBAs that all need to get their access updated.