IBM recently released new SAN Volume Controller (SVC) and Storwize V7000 storage system hardware and software.
New SVC DH8 (Gen8) hardware
IBM has introduced a new, 2U version of their SVC controller node hardware that provides more than twice the CPU compute (6-cores to 16 cores), 3 times the I/O slots, 6 times the local SSDs, and 2.5 times the memory of previous SVC controllers. The DH8 sports 8-core Xeon 2.6Ghz E5-2650v2 CPUs with up to 64 GB of IO cache per node. Each node offers Gen3 PCIe with up to 3 host interface cards per node with multiple quad-port 8Gpbs FC or a single quad port 10GbE CNA card for iSCSI or FCoE.
As you may recall, the previous hardware (CG8) supported real-time data compression (RtC). The DH8 now supports optional hardware Compression Acceleration cards using Intel QuickAssist Technology. The DH8 can have up to 2 Compression Acceleration cards per node. IBM suggests using 2 Compression Acceleration cards when the normal data working set > 32TB. We presume IBM has some tools to help determine the normal data working set. To use RtC one must install the optional 2nd CPU with an additional 32GB of cache bringing the total CPU cache to 64GB.
The new DH8 also uses 12Gpbs dual-ported, SAS for its dedicated SSDs. Moreover, the DH8 now supports up to 192 dedicated SSDs, (up to 48 attached to each pair of nodes). The SSDs are available in 2U drive expansion enclosures with 24 SSDs each.
The item that received the most cheers at IBMEdge2014’s review of new SVC updates was the new integrated redundant battery in the DH8. This eliminates the need for a UPS. With the new DH8 being a 2U controller node, the elimination of a UPS ends up with the same rack space for the SVC configuration (previously, the node and UPS were each 1U).
New SVC DH8 performance
With RtC active and best case compression (pseudo random workload, 65% compression), the CG8 was able to perform 50K read miss IOPS, 16K write miss IOPS and 40K 70:30 read:write miss IOPS while the new DH8 can do 175K, 115K and 149K IOPS respectively. For throughput, the previous generation CG8 provided 1400 MB/sec read miss and 720 MB/sec write miss while the new generation DH8 supports 2,266 MB/sec read miss and a surprising 2,650 MB/sec write miss. When asked why the DH8 write miss throughput was so high, the answer the CG8 was waiting on data compression and with the new Compression Acceleration cards on the DH8 this is no longer an issue.
With no compression active the Cache read IOPS (read-hits) went from 900K IOPS on the CG8 to 1150K IOPS and cache write went from 300K to 500K IOPS. As for miss activity read misses went from 240K IOPS to 700K IOPS and write misses went from 50K IOPS to 200K IOPS.
New Storwize V7000 (Gen 2) hardware
The new V7000 Gen2 hardware was also updated and doubles the processing cores with a new 8-Core Xeon 1.9GHz E5-2628L v2 CPUs with 32GB of cache per node. The new V700 also has an integrated, on-board Compression Acceleration card but also offers another optional Compression Acceleration card. The V7000 Gen 2 provides two host interface card slots which can support two quad-port FC 8Gbps host interface cards or one FC card with a new quad-port 10 GbE card for FCoE, iSCSI or IP replication. Also the integrated battery in the Gen2 V7000 is now contained in the PSU. The new V7000 also includes Gen 3 PCIe and 12Gbps SAS for faster system throughput.
Storwize V7000 performance has improved significantly as well. Without compression, read hits went from 700K to 900K IOPS, write hits went from 200K to 300K IOPS, read miss went from 130K to 240K IOPS and write miss went from 29K to 50K IOPS.
New 7.3.0 software enhancements
Aside from all the new hardware above, new software was also released for SVC and all Storwize systems. The new software includes Easy Tier3 with support for 2 or 3 storage tiers and automatic IO balancing within a storage pool. The 7.3.0 software includes new dual layer cache support that better manages cache to improve sequential look ahead, destage and other caching IO activity. The new Storwize software also offers more drive expansions for both the V5000 and V3700, now supporting up to 9 drive expansions each.
All in all, we would have to say a pretty extensive update to SVC and Storwize system hardware. The improvements in performance are also impressive across the board.
The new RtC hardware is pretty interesting as it removes almost all the burden of data compression from the IO path. We have long been a proponent of data compression but only if it doesn’t impact IO access or latency. In the past this could only be done with special purpose data compression hardware in the data path. With the help of Intel, IBM seems to have solved this problem nicely by using standard off the shelf hardware and their proprietary real-time compression software.
The software changes aren’t nearly as impressive but there’s probably more than meets the eye with the new dual layer caching enhancements, and the update to the latest Easy Tier functionality will help as well.
[This storage announcement summary dispatch was originally sent out to our newsletter subscribers in May of 2014. If you would like to receive this information via email please consider signing up for our free monthly newsletter (see subscription request, above right) and we will send our current issue along with download instructions for this and other reports. Dispatches are posted to our website at least a quarter or more after they are sent to our subscribers, so if you are interested in current storage announcement and/or storage performance results please consider signing up for our newsletter.]
Silverton Consulting, Inc. is a Storage, Strategy & Systems consulting services company, based in the USA offering products and services to the data storage community.