Tony Pearson is a Master Inventor and Senior IT Architect for the IBM Storage product line at the
IBM Executive Briefing Center in Tucson Arizona, and featured contributor
to IBM's developerWorks. In 2016, Tony celebrates his 30th year anniversary with IBM Storage. He is
author of the Inside System Storage series of books. This blog is for the open exchange of ideas relating to storage and storage networking hardware, software and services.
(Short URL for this blog: ibm.co/Pearson )
My books are available on Lulu.com! Order your copies today!
Safe Harbor Statement: The information on IBM products is intended to outline IBM's general product direction and it should not be relied on in making a purchasing decision. The information on the new products is for informational purposes only and may not be incorporated into any contract. The information on IBM products is not a commitment, promise, or legal obligation to deliver any material, code, or functionality. The development, release, and timing of any features or functionality described for IBM products remains at IBM's sole discretion.
Tony Pearson is a an active participant in local, regional, and industry-specific interests, and does not receive any special payments to mention them on this blog.
Tony Pearson receives part of the revenue proceeds from sales of books he has authored listed in the side panel.
Tony Pearson is not a medical doctor, and this blog does not reference any IBM product or service that is intended for use in the diagnosis, treatment, cure, prevention or monitoring of a disease or medical condition, unless otherwise specified on individual posts.
This week I am in Chicago for the IBM Storage and Storage Networking Symposium, which coincides with the System x and BladeCenter Technical Conference. This allows the 800 attendees to attend both storage or server presentations at their convenience. There were hundreds of sessions, over 20 time slots, so for each time slot, you have 15 or so topics to choose from.Mike Kuhn kicked off the series of keynote sessions. Here's my quick recap of each one:
Curtis Tearte, General Manager, IBM System Storage
Curtis replaced Andy Monshaw as General Manager for IBM System Storage. His presentation focused on how storage fits into IBM's Dynamic Infrastructure strategy. Some interesting points:
a billion camera-enabled cell phones were sold in 2007, compared to 450 million in 2006.
IBM expects that there will be 2 billion internet users by 2011, as well as trillions of "things".
In the US, there were 2.2 million medical pharmacy dispensing errors resulting for handwritten prescriptions.
Time wasted looking for parking spaces in Los Angeles consumed 47,000 gallons of gasoline, and generated 730 tons of carbon dioxide.
In the US, 4.2 billion hours are lost, and 2.9 billion gallons of gas consumed, due to traffic congestion.
Over the past decade, servers went from 8 watts to 100 watts per $1000 US dollars.
Data growth appears immune to the economic recession. The digital footprint per person is expected to grow from 1TB today to over 15TB by 2020.
10 hours of YouTube videos are uploaded every minute.
Bank of China manages 380 million bank accounts, processing over 10,000 transactions per second.
At the end of the session, Curtis transitioned from demonstrating his knowledge and passion of storage to his knowledge and passion in his favorite sport: baseball. Chicago is home to both the Cubs and the White Sox.
Roland Hagan, Vice President Business Line Executive, System x
IBM sets the infrastructure agenda for the entire industry. The Dynamic Infrastructure initiative is not just IT, but a complete end-to-end view across all of the infrastructures in play, including transportation, manufacturing, services and facilities.Companies spent over $60 billion US dollars on servers last year. Of these, 53 percent for x86-based servers, 9 percent for Itanium-based, 26 percent for RISC-based (POWER6, SPARC, etc.), and 11 percent mainframe. Theeconomic downturn has impacted revenues, but the percentages continue about the same.
The dominant deployment model remains one application per server. As a result, power, cooling and management costs have grown tremendously. There are system admins opposed to consolidating server images with VMware, Hyper-V, Xen or other server virtualizaition technologies. Roland referred to these admins as "server huggers".To help clients adopt cloud computing technologies, IBM introduced [Cloudburst] appliances. IBM plans to offer specialized versions for developers, for service providers, and for enterprises.
IBM's Enterprise-X Architecture is what differentiates IBM's x86-based servers from all the competitors, surrounding Intel and AMD processors with technology that provides distinct advantages. For example, to support server virtualization, IBM's eX4 provides support for more memory, which often is more critical than CPU resources when deploying large number of guest OS images. IBM System x servers have an integrated management module (IMM) and was the first to change over from BIOS to the new Unified Extensible Firmware Interface [UEFI] standard.
IBM servers offer double the performance, consume half the power, and cost a third less to manage, than comparably priced servers from competitors. Of the top 20 more energy efficient server deployments, 19 are from IBM. Roland cited customer reference SciNet, a 4000-server supercomputer with 30,000 cores based on IBM [iDataPlex] servers. At 350 TeraFLOPs it is ranked #16 fastest supercomputer in the world, and #1 in Canada. With apower usage effectiveness (PUE) less than 1.2, it also is very energy efficient. This means that for every 12 watts of electricity going in to the data center, 10 watts are used for servers, storage and networking gear, andonly 2 watts used for power and cooling. Traditional data centers have PUE around 2.5, consuming 25 watts total for every 10 watts used by servers, storage and networking gear.
Clod Barrera, Distinguished Engineer, Chief Technical Strategist for IBM System Storage
Clod presented trends and directions for disk and tape technology, disk and tape systems, and the direction towards cloud computing.
Continuing my week in Chicago for the IBM Storage and Storage Networking Symposium and System x and BladeCenter Technical Conference, I presented a variety of topics.
Hybrid Storage for a Green Data Center
The cost of power and cooling has risen to be a #1 concern among data centers. I presented the following hybrid storage solutions that combine disk with tape. These provide the best of both worlds, the high performance access time of disk with the lower costs and reduced energy consumption of tape.
IBM [System Storage DR550] - IBM's Non-erasable, Non-rewriteable (NENR) storage for archive and compliance data retention
IBM Grid Medical Archive Solution [GMAS] - IBM's multi-site grid storage for PACS applications and electronic medical records[EMR]
IBM Scale-out File Services [SoFS] - IBM's scalable NAS solution that combines a global name space with a clustered GPFS file system, serving as the ideal basis for IBM's own[Cloud Computing and Storage] offerings
Not only do these help reduce energy costs, they provide an overall lower total cost of ownership (TCO) thantraditional WORM optical or disk-only storage configurations.
The Convergence of Networks - Understanding SAN, NAS and iSCSI in the Data Center Network
This turned out to be my most popular session. Many companies are at a crossroads in choosing data and storage networking solutions in light of recent announcements from IBM and others. In the span of 75 minutes, I covered:
Block storage concepts, storage virtualization and RAID levels
File system concepts, how file systems map files to block storage
Network Attach Storage, the history of the NFS and CIFS protocols, Pros and Cons of using NAS
Storage Area Networks, the history of SAN protocols including ESCON, FICON and FCP, Pros and Cons of using SAN
IP SAN technologies, iSCSI and Fibre Channel over Ethernet (FCoE), Pros and Cons of using this approach
Network Convergence with Infiniband and Fibre Channel over Convergence Enhanced Ethernet (FCoCEE), why Infiniband was not adopted historically in the marketplace as a storage protocol, and the features and enhancements of Convergence Enhanced Ethernet (CEE) needed to merge NAS, SAN and iSCSI traffic onto a single converged data center network [DCN]
Yes, it was a lot of information to cover, but I managed to get it done on time.
IBM Tivoli Storage Productivity Center version 4.1 Overview and Update
In conferences like these, there are two types of product-level presentations. An "Overview" explains howproducts work today to those who are not familiar with it. An "Update" explains what's new in this version of the product for those who are already familiar with previous releases. I decided to combine these into one sessionfor IBM's new version of [Tivoli Storage Productivity Center].I was one of the original lead architects of this product many years ago, and was able to share many personalexperiences about its evolution in development and in the field at client facilities.Analysts have repeatedly rated IBM Productivity Center as one of the top Storage Resource Management (SRM) tools available in the marketplace.
Information Lifecycle Management (ILM) Overview
Can you believe I have been doing ILM since 1986? I was the lead architect for DFSMS which provides ILM support for z/OS mainframes. In 2003-2005, I spent 18 months in the field performingILM assessments for clients, and now there are dozens of IBM practitioners in Global Technology Services andSTG Lab Services that do this full time. This is a topic I cover frequently at the IBM Executive Briefing Center[EBC], because it addressesseveral top business challenges:
Reducing costs and simplifying management
Improving efficiency of personnel and application workloads
Managing risks and regulatory compliance
IBM has a solution based on five "entry points". The advantage of this approach is that it allows our consultants to craft the right solution to meet the specific requirements of each client situation. These entry points are:
Tiered Information Infrastructure - we don't limit ourselves to just "Tiered Storage" as storage is only part of a complete[information infrastructure] of servers,networks and storage
Storage Optimization and Virtualization - including virtual disk, virtual tape and virtual file solutions
Process Enhancement and Automation - an important part of ILM are the policies and procedures, such as IT Infrastructure Library [ITIL] best practices
Archive and Retention - space management and data retention solutions for email, database and file systems
I did not get as many attendees as I had hoped for this last one, as I was competing head-to-head in the same time slot as Lee La Frese covering IBM's DS8000 performance with Solid State Disk (SSD) drives, John Sing covering Cloud Computing and Storage with SoFS, and Eric Kern covering IBM Cloudburst.
I am glad that I was able to make all of my presentations at the beginning of the week, so that I can then sit back and enjoy the rest of the sessions as a pure attendee.
Continuing my week in Chicago, I decided to attend some of the presentations from the System x side. This is the advantage of running both conferences in the same hotel, attendees can choose how many of each they want to participate in.
Wayne Wigley, IBM Advanced Technical Support (ATS), presented a series of presentations on different server virtualization offerings available for System x and BladeCenter servers. I am very familiar with virtualization implemented on System z mainframes, as well as IBM's POWER systems, and have working knowledge of Linux KVM and Xen, so I was well prepared to handle hearing the latest about Microsoft's Hyper-V and VMware's Vsphere version 4.
Microsoft Hyper-V 2008
Hyper-V can run as part of Windows 2008, are standalone on its own.Different levels of Windows 2008 include licenses for different number of Windows virtual machines (VMs).Windows Server 2008 Standard includes 1 Windows VM, Enterprise includes 4 Windows VMs, and the Datacenter edition includes unlimited number of Windows VMs. If you want to run more Window VMs than come included, you need to pay extra for each additional one. For example, to run 10 Windows VMs on a 2-socket server would cost about $9000 US dollars on Standard but only $6000 US dollars on Datacenter edition (list prices from Microsoft Web site).
Unlike VMware, which takes a monolithic approach as hypervisor, Hyper-V is more like Xen with a microkernelized approach. This means you need a "parent" guest OS image, and the rest of the Guest OS images are then considered "child" images.These child images can be various levels of Windows, from Windows XP Pro to Windows Server 2008, Xen-enabled Linux, or even a non-hypervisor-aware OS.The "parent" guest OS image provides networking and storage I/O services to these "child" images.For the hypervisor-aware versions of Windows and Linux, Hyper-V allows optimized access to the hypervisor, "synthetic devices", and hypercalls. Synthetic devices present themselves as network devices, but only serve to pass data along the VMBus to other networking resources. This process does not require software emulation, and therefore offers higher performance for virtual machines and lower host system overhead.For non-hypervisor-aware OS images, Hyper-V provides device emulation through the "parent" image, which is slower.
Microsoft System Center Virtual Machine Manager (SCVMM) can manage both Hyper-V and VMware VI3 images.Wayne showed various screen shots of the GUI available to manage Hyper-V images.In standalone mode, you lose the nice GUI and management console.
Hyper-V supports external, internal and private virtual LANs (VLAN). External means that VMs can communicate with the outside world over standard ethernet connections. Internal means that VMs can communicate with "parent" and "child" guest images on the same server only. Private means that only "child" guests can communicate with other "child" images.
Hyper-V supports disk attached via IDE, SATA, SCSI, SAS, FC, iSCSI, NFS and CIFS. One mode is "Virtual Hard Disk" (VHD) similar to VMware VMDK files. The other is "pass through" mode, which are actual disk LUNs accessed natively. VHDs can be dynamic (thin provisioned), fixed (fully allocated), or differencing. The concept of differencing is interesting, as you start with a base read-only VHD volume image, and have a separate "delta" file that contains changes from the base image.
Some of the key features of Hyper-V 2008 are:
Being able to run concurrently 32-bit and 64-bit versions of Linux and Windows guest images
Support for 64 GB of memory and 4-way symmetric multiprocessing (SMP) per VM
Clustering for High Availability and Quick Migration of VM images
Live backup with integration with Microsoft's Volume Shadow Copy Services (VSS)
Virtual LAN (VLAN) support, and Virtual and Pass-through physical disk support
A clever VMbus, virtual service parent/client approach to sharing hardware
Optimized performance options for hypervisor-aware versions of Windows and Linux, and emulated supportfor non-hypervisor-aware OS images.
VMware Vsphere v4.0
This was titled as an "Overview" session, but really was an "Update" session on the newest features of this release. The big change appears to be that VMware added "v" in front of everything.
Under vCompute, there are some new features on VMware's Distributed Resource Scheduler (DRS) which includes recommended VM migrations. Dynamic Power Management (DPM) will move VMs during periods of low usageto consolidate onto fewer physical servers so as to reduce energy consumption.
Under vStorage, vSphere introduces an enhanced Plugable Storage Architecture (PSA), with supportfor Storage Array Type Plugins (SATP) and Path Selection Plugins (PSP). This vStorage API allows forthird party plugins for improved fault-tolerance and complex I/O load balancing algorithms. This releasealso has improved support for iSCSI, including Challenge-Handshake Authentication Protocol (CHAP) support.Similar to Hyper-V's dynamic VHD, VMware supports "thin provisioning" for their virtual disk VMDK files.A feature of "Storage Vmotion" allows conversion between "thick" and "thin" provisioning formats.
The vStorage API for Data Protection provide all the features of VMware Consolidated Backup (VCB). The APIprovides full, incremental and differential file-level backups for Windows and Linux guests, including supportfor snapshots and Volume Shadow Copy Services (VSS) quiescing.
VMware introduces direct I/O pass-through for both NIC and HBA devices. While thisallows direct access to SAN-attached LUNs similar to Hyper-V, you lose a lot of features like Vmotion, High Availability and Fault Tolerance. Wayne felt that these restrictions are temporary, that hopefully VMwarewill resolve this over the next 12 months.
Under vNetwork, VMware has virtual LAN switches called vSwitches. This includes support for IPv6and VLAN offloading.
The vSphere server can now run with up to 1TB of RAM and 64 logical CPUs to support up to 320 VM guest images.Each VM can have up to 255GB RAM and up to 8-way SMP.Vsphere ESX 4 introduces a new virtual hardware platform called VM Hardware v7. While Vsphere 4.0 can run VMs from ESX 2 and ESX 3, the problem is if you have new VMs based on this newer VM Hardware v7, you cannot run them on older ESX versions.
Vsphere comes in four sizes: Standard, Advanced, Enterprise, and Enterprise Plus, ranging in list price from $795 US dollars to $3495 US dollars.
While IBM is the #1 reseller of VMware, we also are proud to support Hyper-V, Xen, KVM and other similar products.Analysts expect most companies will have two or more server virtualization solutions in their data center, and it is good to see that IBM supports them all.
Continuing my week in Chicago, for the IBM Storage Symposium 2009, I attended what in my opinion was the bestsession of the week. This was by a guy named Chip Copper, who covered IBM's set of Ethernet and Fibre Channelnetworking gear. Attributes are the four P's:
Power and Cooling (electricity usage)
Equipment comes in two flavors: Top-of-Rack (ToR) thin pizza box switches, and Middle-of-Row (MoR) much larger directors.The MoR directors are engineered for up to 50Gbps per half-slot, so 10GbE and the future 40GbE can be easily accommodated in a single half-slot, and the future 100GbE can be done with a full slot (two half-slots).
While many companies might have been contemplating the switch from copper wires to optical fiber, there is a new reason for copper cables: Power-over-Ethernet (PoE). Many IP-phones, digital video surveillance cameras, and other equipment can have a single cable that delivers both signal and electricity over copper. If you have already deployed optical fiber throughout the building, there are "last mile" options where the signals are converted to copper wires and electrical energy added for these types of devices.
Two directors can be connected together with Inter-Chassis Link (ICL) cables to make them look like a single director with twice the number of ports. These are different than Inter-Switch Links (ISL) as they are not counted as an extra "hop" for networking counting purposes, especially important for FICON usage.
Today, we have 1Gbps, 2Gbps, 4Gbps and 8Gbps Fibre Channel. Since these all use 10-for-8 encoding (10 bits represents one 8-bit byte), then in was easy to calculate throughput: 8Gpbs was 800 MB/sec, for example. Auto-negotiation between speeds is not done at the HBA card, switch or director blade itself, but in the Short Form-factor Pluggable (SFP) optical connector. However, you can only auto-negotiate if the encoding matches. The 4/2/1 SFP can run at 4Gbps or auto-negotiate to slower 2Gbps and 1Gbps. The 8/4/2 SFP can run at 8Gbps, or auto-negotiate down to slower 4Gpbs and 2Gbps. Folks who still have legacy 1Gbps equipment, but want to run some things at 8 Gbps, can buy 8Gbps-capable switches or director blades, but then put some 4/2/1 SFPs into them. These 4/2/1 SFP are cheaper, so this might be something to consider if budgets are tight. Some SFPs handle up to 10km distances, but others only 4km, so be careful not to order the wrong ones.
Unfortunately, there are proposals in place for 10Gbps and 40Gbps that would use a different 66-for-64 encoding (66 bits represent 8 bytes), so 10Gbps would be 1200 MB/sec. These are used today for ISL between directors and switches.In theory, the 40Gbps could auto-negotiate down to 10Gbps, but not to any of the 8/4/2/1 Gbps that use different 10-for-8 encoding.
For those who cannot afford a SAN768B, there is a smaller SAN384B that can carry: 192 ports (4Gpbs/2Gbps), 128 ports (8Gbps) or 24 ports (10Gbps). The SAN384B can be ICL connected to another SAN384B or even the SAN768B as your needs grow.
On the entry-level side, the SAN24B-4 offers a feature called "Access Gateway". This makes the SAN24B look like an SAN end-point host, rather than a switch, and makes initial deployment of integrated bundled solutions easier. Once connected to everything, you can convert it over to full "switch" mode.The SAN40B-4 and SAN80B-4 provide midrange level support, including Fibre Channel routing at the 8Gbps level. In fact, all 8Gbps ports include routing capability. IBM offers both single-port and dual-port 8Gbps host bus adapter (HBA) cards to connect to these switches. These HBA offer 16 virtual channels per port, so that if you have VMware running many guests, or want to connect both disk and tape to the same HBA, you can keep the channel traffic separate for Quality of Service (QoS).
Chip wrapped up his session to discuss Fibre Channel over Ethernet (FCoE), and explained why we need to have a loss-less Convergence Enhanced Ethernet (CEE) to meet the needs of storage traffic as well as traditional Fibre Channel does today. IBM offers all of the equipment you need to get started today on this FCoCEE, with Converged Network Ethernet cards for your System x servers, and a new SANB32 that has 24 10GbE CEE ports and 8 traditional 8Gbps FC ports. This means that you can put the CNA card in your existing servers, connect to this switch, and then connect to your existing 10GbE LAN and your existing 8Gpbs or 4Gpbs FC-based SAN to the rest of your storage devices.
Worried that the FCoE or CEE standards could change after you deploy this gear? Aren't most LAN and SAN switches based on Application-specific integrated circuit [ASIC] chips which are created in the factory? Don't worry, IBM's equipment have put all the standards-vulnerable portions of the logic into separate Field-programmable gate array [FPGA] that can be updated with simplya firmware upgrade. This is future-proofing I can agree with!
Continuing my week in Chicago, for the IBM Storage Symposium 2008, I attended two presentations on XIV.
XIV Storage - Best Practices
Izhar Sharon, IBM Technical Sales Specialist for XIV, presented best practices using XIV in various environments.He started out explaining the innovative XIV architecture: a SATA-based disk system from IBM can outperformFC-based disk systems from other vendors using massive parallelism. He used a sports analogy:
"The men's world record for running 800 meters was set in 1997 by Wilson Kipketer of Denmark in a time of 1:41.11.
However, if you have eight men running, 100 meters each, they will all cross the finish line in about 10 seconds."
Since XIV is already self-tuning, what kind of best practices are left to present? Izhar presented best practicesfor software, hosts, switches and storage virtualization products that attach to the XIV. Here's some quickpoints:
Use as many paths as possible.
IBM does not require you to purchase and install multipathing software as other competitors might. Instead, theXIV relies on multipathing capabilities inherent to each operating system.For multipathing preference, choose Round-Robin, which is now available onAIX and VMware vSphere 4.0, for example. Otherwise, fixed-path is preferred over most-recently-used (MRU).
Encourage parallel I/O requests.
XIV architecture does not subscribe to the outdated notion of a "global cache". Instead, the cache is distributed across the modules, to reduce performance bottlenecks. Each HBA on the XIV can handle about 1400requests. If you have fewer than 1400 hosts attached to the XIV, you can further increase parallel I/O requests by specifying a large queue depth in the host bus adapter (HBA).An HBA queue depth of 64 is a good start. Additional settings mightbe required in the BIOS, operating system or application for multiple threads and processes.
For sequential workloads, select host stripe size less than 1MB. For random, select host stripe size larger than 1MB. Set rr_min_io between ten(10) and the queue depth(typically 64), setting it to half of the queue depth is a good starting point.
If you have long-running batch jobs, consider breaking them up into smaller steps and run in parallel.
Define fewer, larger LUNs
Generally, you no longer need to define many small LUNs, a practice that was often required on traditionaldisk systems. This means that you can now define just 1 or 2 LUNs per application, and greatly simplifymanagement. If your application must have multiple LUNs in order to do multiple threads or concurrent I/O requests, then, by all means, define multiple LUNs.
Modern Data Base Management Systems (DBMS) like DB2 and Oracle already parallelize their I/O requests, sothere is no need for host-based striping across many logical volumes. XIV already stripes the data for you.If you use Oracle Automated Storage Management (ASM), use 8MB to 16MB extent sizes for optimal performance.
For those virtualizing XIV with SAN Volume Controller (SVC), define manage disks as 1632GB LUNs, in multiple of six LUNs per managed disk group (MDG), to balance across the six interface modules. Define SVC extent size to 1GB.
XIV is ideal for VMware. Create big LUNs for your VMFS that you can access via FCP or iSCSI.
Organize data to simplify Snapshots.
You no longer need to separate logs from databases for performance reasons. However, for some backup productslike IBM Tivoli Storage Manager (TSM) for Advanced Copy Services (ACS), you might want to keep them separatefor snapshot reasons. Gernally, putting all data for an application on one big LUNgreatly simplifies administration and snapshot processing, without losing performance.If you define multiple LUNs for an application, simply put them into the same "consistencygroup" so that they are all snapshot together.
OS boot image disks can be snapshot before applying any patches, updates or application software, so that ifthere are any problems, you can reboot to the previous image.
Employ sizing tools to plan for capacity and performance.
The SAP Quicksizer tool can be used for new SAP deployments, employing either the user-based orthroughput-based sizing model approach. The result is in mythical unit called "SAPS", which represents0.4 IOPS for ERP/OLTP workloads, and 0.6 IOPS for BI/BW and OLAP workloads.
If you already have SAP or other applications running, use actual I/O measurements. IBM Business Partners and field technical sales specialists have an updated version of Disk Magic that can help size XIV configurations fromPERFMON and iostat figures.
Lee La Frese, IBM STSM for Enteprise Storage Performance Engineering, presented internal lab test results forthe XIV under various workloads, based on the latest hardware/software levels [announced two weeks ago]. Three workloadswere tested:
Web 2.0 (80/20/40) - 80 percent READ, 20 percent WRITE, 40 percent cache hits for READ.YouTube, FlickR, and the growing list at [GoWeb20] are applications with heavy read activity, but because of[long-tail effects], may not be as cache friendly.
Social Networking (50/50/50) - 50 percent READ, 50 percent WRITE, 50 percent cache hits for READ.Lotus Connections, Microsoft Sharepoint, and many other [social networking] usage are more write intensive.
Database (70/30/50) - 70 percent READ, 30 percent WRITE, 50 percent cache hits for READ.The traditional workload characteristics for most business applications, especially databases like DB2 andOracle on Linux, UNIX and Windows servers.
The results were quite impressive. There was more than enough performance for tier 2 application workloads,and most tier 1 applications. The performance was nearly linear from the smallest 6-module to the largest 15-module configuration. Some key points:
A full 15-module XIV overwhelms a single SVC 8F4 node-pair. For a full XIV, consider 4 to 8 nodes 8F4 models, or 2 to 4 nodes of an 8G4. For read-intensive cache-friendly workloads, an SVC in front of XIV was able to deliver over 300,000 IOPS.
A single node TS7650G ProtecTIER can handle 6 to 9 XIV modules. Two nodes of TS7650G were needed to drivea full 15-module XIV. A single node TS7650 in front of XIV was able to ingest 680 MB/sec on the seventh day with17 percent per-day change rate test workload using 64 virtual drives. Reading the data back got over 950 MB/sec.
For SAP environments where response time 20-30 msec are acceptable, the 15-module XIV delivered over 60,000 IOPS. Reducing this down to 25,000-30,000 cut the msec response time to a faster 10-15 msec.
These were all done as internal lab tests. Your mileage may vary.
Not surprisingly, XIV was quite the popular topic here this week at the Storage Symposium. There were many moresessions, but these were the only two that I attended.