ORIGINALLY POSTED 15th October 2008

10,490 views on developerworks

It seems like only a few weeks ago I was covering the release of SVC 4.3.0 with support for Space-efficient Vdisks and Vdisk Mirroring, and here we are again with today’s announce of SVC 4.3.1 software.

This time round however its more than just a software release. Until today, mid-sized businesses wishing to deploy virtualized storage have been forced to choose between more costly solutions designed for large enterprises or less robust offerings with limited capability and lower growth potential.

IBM System Storage™ SAN Volume Controller Entry Edition

Today IBM has announced the IBM System Storage™ SAN Volume Controller Entry Edition, (2145-8A4) which offers the same SVC enterprise-class storage virtualization functionality in a more affordable package optimized to meet the needs of mid-sized businesses.

Its important to understand that the SVC Entry Edition (which I will refer to as SVC EE from herein) is running the same software base that is provided with the standard SVC nodes. This means mid-sized businesses can now benefit from the same feature rich functionality provided by the SVC software, packaged in a more affordable solution. SVC was designed to help simplify management of storage while also helping to improve storage utilization, which helps improve an IT departments return on investment. SVC EE also helps improve administrator productivity so allowing more efficient management of storage today, and as a business grows. SVC EE is an affordable way to increase the flexibility of existing infrastructures and respond to constant demands for more storage while remaining in control of investments and improving efficiency.

Along with SVC EE hardware, a new license model has been added specifically for SVC EE software users. This model is designed to meet the needs of mid-sized business. The licensing model for SVC EE clusters is based on the number of physical drives being virtualized, rather than the total capacity. This makes the licensing itself more-affordable, needless to say it also makes the case for large capacity HDD. The SVC EE license is per cluster. (This differs from standard SVC, where the licensing is per site – and can cover multiple clusters).

The challenges facing mid-sized businesses are the same as enterprise businesses. Data growth. One customer recently summed this up by explaining they were “drowning in data”. SVC EE can grow with your business and can easily be converted to the full SVC offering for growth up to large-enterprise-scale environments. This is a license conversion only, the SVC EE hardware runs the same SVC software as existing SVC hardware. The conversion from an EE to normal SVC changes the license to a capacity model and allows for future growth and extension of your SVC cluster, again improving your return on investment by expanding the SVC cluster itself.

If you have converted the SVC EE to a capacity based license you can intermix the SVC EE (8A4) node models with existing SVC node models (4F2, 8F2, 8F4 and 8G4).

For existing SVC users with older hardware models, such as 4F2 and 8F2, you can non-disruptively upgrade your node hardware to the SVC EE (8A4) hardware and continue to use your existing capacity license. This provides a more-affordable hardware upgrade root for existing customers as well as benefiting from the performance gains of moving to the 8A4 node. As an example the 8A4 hardware can provide twice the throughput of a 4F2 or 8F2 node.

What’s in the SVC EE hardware. Well it uses a single socket variant of the Intel Xeon platform that all other SVC hardware uses. Inside the node is a single Dual-core 3.0GHz Xeon CPU and 8GB of cache memory (per node). Like other node hardware the SAN interface comprises 4x 4Gbit Fibre Channel ports. With all the work we did back in the 4.2.0 software release to optimise the SVC software to make best use of the multi-core Xeon architecture, the Dual core CPU allows the same benefits, with the same binding, lock elimination and performance enhancements that provide more than enough performance capability for the SVC EE software. There is plenty of headroom for when your business grows and you need that extra performance. The major difference in hardware when comparing the SVC EE with the existing 8G4 node is internal memory bandwidth. Due to less memory channels with the single socket SVC EE (vs Dual socket 8G4), the 8A4 provides around 60% of the bandwidth of the 8G4 hardware… at 60% of the cost…

---

SVC 4.3.0 – Space-efficient – SPC-1 Benchmark

The second SVC related announcement today is the publication of our recent 4.3.0 Space-efficient Virtual Disk Storage Performance Council benchmark.

What I found interesting last month was 3Par’s announcement of their new T-class controller systems which achieved a respectable 224K SPC-1 IOPs, asked for who the grandfather of Thin Provisioning is, most would answer 3Par. However their published result was not performed using their Thin Provisioning technology… I also found it amusing that their press release made much of their new ASIC, with zero-detect and XOR and that these were functions that could only reside in an ASIC to get the performance benefits. I guess its not enough of a benefit to warrant using them in the SPC-1 submission, as not only was it lacking in Thin Provisioning but it also used mirroring (RAID-1), not RAID-5.

I was thinking that therefore IBM was the first vendor to submit an SPC-1 result using its “Thin Provisioning” technology, but I guess that NetApp and the IBM branded N-series could well be considered as a Thin Provisioning solution by virtue of their WAFL filesystem which implements a form of Log Structured Array (LSA). So maybe IBM isn’t the first to submit a Thin Provisioned result to the SPC.

However, since there are surprisingly few Thin Provisioned solutions with submitted SPC-1 results, I’m sure that those following the SPC-1 horserace will appreciate that SVC is adding an important new entry today. It may also come as a surprise that our SVC 4.3.0 Space-efficient SPC-1 result actually surpasses our current SPC-1 benchmark, clocking in at 274.9K SPC-1 IOPs. No mean feat considering ALL of the benchmarked capacity was run to virtual disks that were Space-efficient. You may wonder why this has actually improved on our previous result – that was the immediate question the performance team were asked when we reported the results internally. During the 4.3.0 development phase I worked closely with the SEV team to perform early internal benchmarking, highlight areas that were unduly suffering and prod the developers to fix them. As a result of this work one or two key performance tweaks were found not only in the SEV code itself, but some in the base I/O path code that resulted in gains in certain write based workloads for non-SEV and SEV volumes alike. Although we haven’t re-submitted the base SVC for SPC-1, given that our SEV result surpassed our previous result…

The configuration used for this benchmark was identical to the existing 272K SPC-1 result, with the addition of the SEV feature being enabled across all vdisks. Since SPC-1 simulates an OLTP environment, and runs a set of tests that setup the ‘database’ infrastructure and verify I/O operations etc, the data infrastructure was created and allocated before the results were taken. Thus this gives a true comparison of an SVC cluster where Space-efficient Vdisks are used and simulates reads and writes over the existing customer data-set. I know our EMC friends always mock the SPC as un-realistic, however as a yard stick, even Chuck agreed it was perfectly reasonable to compare one SVC result with another, thus this shows that in such an environment the impact of directory lookups (and indeed the quality of the directory caching algorithms) mean there is little or no impact to running our fine grained SEV. If an application writes tonnes of new data every day, there there will be an additional impact as new grains are allocated and directory writes are committed to disk, as is the case with all “Thin provisioning” solutions (Of course in the latter case there maybe a question if SEV is the correct solution, and there are always trade-offs between performance and capacity).

Ah well. Maybe they, and others will step up and post similar Thin Provisioned results in the future. It will be interested how they stack up to stock results.

I’ll cover some more detail of the other code enhancements and interop additions over the next few days, and if you are at StorageExpo or SNW be sure to stop by the IBM stands to get more details on the new SVC Entry Edition.

---

Details of the SVC Results for 4.2 are available at:
http://www.storageperformance.org/results/benchmark_results_spc1#a00052
Details of the SVC SPC-1 Results for 4.3 SEV are available at:
http://www.storageperformance.org/results/benchmark_results_spc1#a00072
Deatils of the 3Par T800 SPC-1 Results are available at:
http://www.storageperformance.org/results/benchmark_results_spc1#a00069