Video Surveillance and IP Storage - PDF

Please download to get full document.

View again

of 31
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report



Views: 5 | Pages: 31

Extension: PDF | Download: 0

Related documents
Video Surveillance and IP Storage A scalability study with OnSSI NetDVMS and Intransa Scalable, External IP Storage 1 Contents Introduction 3 Part One: Video Surveillance Solution Overview
Video Surveillance and IP Storage A scalability study with OnSSI NetDVMS and Intransa Scalable, External IP Storage 1 Contents Introduction 3 Part One: Video Surveillance Solution Overview 4 Video Recording and Frame Loss 4 Technology Used 4 Intransa Storage Architecture 5 Video and Storage System Terminology 7 Frame Sampling Method and Cumulative Percentage 10 Part Two: Test Configuration 13 System Setup and Configuration 13 System Performance Evaluation 14 Single LUN Using a 14 Disk RAID 5 Diskgroup 14 Two LUNs Using Single 14 Disk RAID 5 Diskgroup 15 Four LUNs With Two 7 Disk RAID 5 Diskgroups 16 Disk/File Level Fragmentation 16 Number of Cameras or FPS 17 Frame Loss Due to Disk Failure and RAID Rebuild 18 Part Three: Guidelines for Configurations 18 NVS Best Practices 19 Networking Best Practices 21 Storage Best Practices 22 About Intransa 29 2 Introduction Used to thinking about recorded video from the days not so long ago when tape-based VCR recording was the primary media, many security practitioners are finding new challenges confronting them today. Requirements for longer retention periods, better video quality, more coverage, and increased reliability are vying for attention while budgets are stagnant or under pressure. IP storage is a key to resolving these seemly opposed issued. Unfortunately, when it comes to disk storage as a component to video surveillance systems, most security practitioners only think about storage capacity in terms of DVRs and NVRs. And worst of all, most of the time, storage is thought of only after the rest of the surveillance system has been designed and purchased. This can be a costly mistake, not only because DVRs and NVRs typically use fixed, captive storage instead of more cost effective, scalable and reliable shared, external IP storage, but also because not all storage performance is equal. What many security practitioners may not know is that the storage system that is critical to the long term viability of their surveillance system. This report looks at the issues, shows performance limitations and results, and then demonstrates how a powerful video management system such as the OnSSI NetDVMS application can leverage Intransa Shared, External IP Storage much more effectively than DVR/NVR fixed, captive storage. Testing was performed on a real, functioning video surveillance system with components from OnSSI, Intransa, AXIS Communications, Dell, and other vendors, a strong example of an IP based system with multi-vendor interoperability. The report closes with recommendations on best practices for enabling a video surveillance system with OnSSI NetDVMS. 3 Part One: Video Surveillance Solution Overview Video Recording and Frame Loss Unintended video frame loss is common in many video storage systems today, typically caused by one of just two factors. First, the storage system may not be able to keep up with the video data rate due to the number of cameras, compression, resolution and frame rate settings employed. Or the problem may be because the network video recorder (NVR) can not process the all of video frames in a timely manner. Dropping frames in video applications may have been tolerable in some application environments, exhibited in behavior such as a nearly unnoticeable picture freeze. However, more and more security practitioners and their end user customers are finding continued frame loss unacceptable. Often, frames drop more as more cameras with higher resolution are added to the surveillance system. From storage system prospective, physical security system designers should consider this and the impact that the design will have on performance when it comes to system scalability and stored video quality. Technology Used OnSSI NetDVMS is a perfect example software platform to represent all VMS vendors. NetDVMS is a very modern, powerful digital video management system that is fully IP enabled. This whitepaper will look at NetDVMS to illustrate how storage affects stored video quality. We will also look at how videos are actually recorded to storage systems. Finally, we will define parameters for quantifying stored video quality, and then present a series of useful performance measurements. From a performance perspective, an Intransa mid-range system will be used as an appropriate storage solution example of an IP storage area network. Intransa StarterBlock occupies the middle of the company s shared, external IP storage family, with more system throughput and capacity than either the edge recording platform StarterBlock or the departmental EdgeBlock series. A single Intransa StorStac BuildingBlock system is able to support 880 cameras with 30FPS 4CIF MJPEG compression [26400 FPS]. The high end of the Intransa product line, PerformanceBlock, can support 3,500 cameras at similar compression rates [100,000 FPS]. With such capability for scaling to huge online video repositories, Intransa enables OnSSI NetDVMS for large scale video surveillance deployment. 4 Intransa Storage Architecture Intransa StorStac BuildingBlock systems consist of two independently scalable components, integrated form a complete IP-based storage area network (SAN) platform for recording of video and data. These components are: Performance Controller Units or PCUs Storage Capacity Enclosures or SCEs Performance Controller Units handle all storage management and virtualization. PCUs are clustered to form one or more realms, managing the performance scalability (each adds additional system throughput), and to support high availability (should one PCU fail, the system fails over to another PCU in the cluster; up to four PCUs may be clustered together). Storage Capacity Enclosures hold the disk drives used in the system. All are hotswappable so as to be replaceable while the rest of the system continues to function in the event of a disk failure. Multiple SCEs can be added, modularly increasing storage capacity for retention up to 1,500TB. Such independent scalability of both processor throughput and retention capacity allows video surveillance system administrators to manage the number of cameras and varied resolution, independent of the retention period. Table 1 below summarizes optimum storage requirements for video surveillance applications. 5 Table 1: Summarize Video Surveillance Storage Requirements Cost / Low Entry Price & Manageability Retention & Capacity Scaling Resolution & Performance Scaling (PFS) Reliability Microsoft SQL Server/VMware ESX Support Tight Integration Distance Scaling From single storage enclosure to multiple, managed as a single entity Easy setup and simple ongoing administration Managed through a single GUI interface, with easy provisioning Manage major functions from choice of physical security or IT perspective Make efficient use of disk capacity Support current capacity needs and future requirements Scale from a few TB to 100, or 1000 and beyond modularly Add capacity on demand, without re-cabling Optimized for video surveillance I/O workload Dynamically add cameras / NVRs / IP devices on demand, from a few to thousands Minimum performance impact by: - Fragmentation at file or disk level - RAID rebuild - Controller failure / link failure No single point of failure Clustered storage RAID support HA configurations Server and storage consolidation through virtualization. From storage planning, provisioning, monitoring to performance Load balancing, high availability. Support LAN and WAN IP-based The Intransa Scalable, External IP Storage Architecture is ideal for surveillance deployments as the primary video storage system. All Intransa storage platforms in the product line are Security-Grade IP Video Storage certified, supporting nearly 100 applications from physical security and IT vendors. OnSSI NetDVMS is also Security-Grade IP Video Storage certified though the Intransa StorAlliance Lab program. 6 Intransa IP storage includes design features that deliver outstanding capacity and performance. These are: Improved video storage reliability and capacity, allowing increased retention, higher resolution and maximum frame rates while decreasing maintenance and administration burden and costs Scalable IP Storage optimized for video surveillance IO workload, including heaving near-constant writes Certified for enterprise-grade IT storage for varied applications including consolidation and virtualization with VMware ESX, & Microsoft operating systems, Microsoft Exchange (2003 & 2007) & Microsoft SQL Server, plus applications such as VTL and disk-to-disk backup and data warehousing Certified for physical security applications through the Intransa StorAlliance Labs, the GSO2010/GSI2010 labs, and the IPVS Magazine lab program Affordably grow capacity from 4 to 1,500TB or increase performance from 220MB/sec to 880MB/sec modularly with BuildingBlock, without re-cabling Advanced RAID 0, 1, 5, 6 and 10 protection, high availability configurations, and hot-swap disk drives, fans, power supplies and major components Video & Storage System Terminology Physical security and information technology (IT) are very different disciplines, and terminologies demonstrate this. A quick review of these terminologies and how they are applied is required to ensure a level understanding of the discussion. In a typical surveillance system, there are four components: surveillance camera, networking infrastructure, Network Video Recorder (NVR) and video storage system. Figure 1 shows how video frames are recorded onto storage devices. 7 Camera vendors usually define performance as a combination of the number of channels, image resolution, frames per second, and compression method. Camera Channel - Typically represents a single camera feed that can support multiple subscribers view the video independently). Image Resolution - Usually measured in CIF, 4CIF, megapixel, 10 megapixel, etc. Frames per Second or FPS - The higher the frame rate, the better the video quality. Compression Method - MJPEG, MPEG-2, MPEG-4, and H.264 are common. As more intelligence is added to the cameras for things like motion detection and video analytics, the resources of the device become less available. This can lead to video quality becoming degraded. In the IT networking world, performance is based on a combination of available network bandwidth plus how fast switches/routers on the network are able switch and route packets to their destinations from individual devices. Network bandwidth ranges from as low as 10Mbps to 100Mbps (fast Ethernet), 1,000Mbps (1Gbps) and more recently 10Gbps. Note that the Mbps unit is Megabits per second. Typical IP network interfaces are 1GbE (1 Gigabit Ethernet) or 10GbE (10 Gigabit Ethernet). 8 For switches/routers, performance is related to how many packets per second the device can switch toward its destination and what is the latency or delay for each packet to transverse the switching device. The NVR is the engine of the modern video surveillance system. The NVR manages the cameras as well as storage system. The NVR receives frames from various cameras, converts the frames into IOs (Input/Output operations), and then writes the IOs to the storage system. There are two ways NVRs write to storage: DVRs can directly write to a storage block device (a disk drive), or they can write to the storage device (disk drive) through a file system. For Windows-based NVRs, the NTFS (Net Technology File System) file system is typically used. The ext3 (Third Extended File System) file system is common for Linux-based NVRs. Many NVRs also offer advanced video functionality, such as video analytics, access control and event management. For more definitions and specific details of video surveillance, IT and storage terms, Intransa provides an extensive Glossary of Terms at useful for those interested in either IT or physical security terminology. Storage is the last mile of the video surveillance system, where the rubber meets the road. Video frames from cameras are passed from the NVR to the storage system as IOs. This translates in the IT storage world to IO workload. The pattern used by IOs to access the storage system has a huge effect on the storage system performance. The list of parameters for IO workload includes: request block size (typically in the unit of KB or KiloByte), random or sequential actions, read write ratio, etc. When it comes to the storage IO transport, the terms most pertinent are SCSI (Small Computer System Interface), FC (Fibre Channel), SAS (Serial Attached Storage) and iscsi (SCSI over IP protocol). The device that actually stores the video is disk drive. Common current disk drive types are SATA (Serial ATA or Serial Advanced Technology Attachment), SAS (Serial Attached SCSI) and FC (Fibre Channel). Disk drives are differentiated along the lines of performance, capacity and reliability. Storage systems provide virtualized storage to the NVR, and provide data loss protection through RAID (Redundant Array of Independent Disks). The performance terminology for storage is typically measured units of MBps (MegaBytes), for throughput and IOPS (IOs per second). For protection, review, or analysis, video also needs to be archived or backed up. When designing and implementing a video surveillance system, practitioners need to 9 be aware that the storage requirements for a backup/archive application are very different from the requirements for live video. As a result, many NVRs use the concept of separating live video storage from storage used for archived video, yet are able access both through the same management interface. OnSSI NetDVMS is an example of a system with this advanced capability. Frame Sampling Method and Cumulative Percentage How do we define the recorded video quality for a video surveillance application? For many applications, only recorded video is meaningful, so measuring frame losses is a measure of the recorded video quality. The recorded video quality can be considered a total system scalability measurement. It can be impacted by the number of cameras, FPS, resolution, NVR configuration and the underlying storage system. Since video is a continuous streaming application, we will need to use a statistical sampling method to quantify the recorded video quality. The cumulative percentage of recorded frames fits nicely into describing video quality, since applications have different tolerances and requirements for video quality. For this paper, we will use two parameters: Total Frame Loss (TFL) Frame Cumulative Percentage Total Frame Loss is the percentage of fame loss over a period of time. Frame Cumulative Percentage as a way to describe the smoothness of the video stream. As an example, Appendix A shows a measured FPS sampling result. The results were sampled every 30 seconds over a period of 1 hour. These results are based on 60 cameras, at 30FPS and 4CIF with MJPEG compression. The storage is based on an array of 14 RAID 5 SATA disk drives. Quick math would lead you to expect that the storage system would therefore record 60*30FPS = 1800 FPS or (cameras x frames per second = total frames per second). Figure 2 shows sample history, and you can see there are few drops on frame rate. 10 This shows that average FPS is 1737 FPS, so the total frame loss is actually ( )/1800 = 3.5%. To better see how frames get dropped, Figure 3 shows a histogram of the above samples. The majority of samples are around 1800 FPS as expected. However, 11 there are also some instances where the recorded FPS is quite lower than the average 1800FPS. To see what percentage of samples fall below the FPS threshold, we plotted the cumulative percentage of the FPS sampling as shown in the figure below. Figure 4 is very useful to defining acceptable recorded video quality. Reading from the graph, you can see that 50% of the samples fall below 1750 FPS. 9% are less than 1650 FPS, and 3% are less than 1600 FPS. Therefore, if you assume your recorded video quality requirement has a zero tolerance for FPS dropping below 50% of required frame rate (if 1800 FPS is the expected frame rate, sampling instances must be 1800*50% or 900 FPS). The graph indicates that this is possible. If your require is that no more than 2% of samples have a frame loss exceeding 10% (1800*(1-10%) = 1620 FPS), the graph indicates this cannot be achieved with the current system. 12 Part Two: Test Configuration System Setup and Configuration Figure 5 shows the system setup configuration. The Intransa BuildingBlock system consists of two independent scaling components: Performance Control Units (PCU) that handle all storage management and virtualization, and the Storage Capacity Enclosure (SCE) married with the Storage Expansion Enclosure (SEE) - these are basically disk drive enclosures. PCUs are connected to SCEs through standard Ethernet network. Figure 5: System setup topology System Performance Evaluation In this example, our test system is based on a OnSSI NetDVMS running on two Dell 2950 servers, a typical platform used in this market. The Dell servers are connected to an Intransa StorStac BuildingBlock through a typical Ethernet switch. The devices and software in the configuration are as follows: Axis Communications virtual camera version OnSSI NetDVMS version 6.0e Microsoft Windows 2003 server R2 32bit OS Intransa StorStac Microsoft iscsi software Initiator 2.05 Intel Pro MT 1000 Dual port NIC card Dell PowerConnect 4948, jumbo frame enabled Ethernet switch 13 System Performance Evaluation A LUN or Logical Unit Number is a network storage term. A LUN is a method of linking multiple disk drives together to form a single volume. From a Windows perspective for example, instead of 14 drives appearing as separate drive letters each, A: through N:, the LUN can be set to appear as a single drive or volume. A LUN layout is how individual LUNs are presented to or are accessible by the NVR system. Getting the LUN layout correct can go a long way to achieving optimal performance. Single LUN Using a 14 Disk RAID5 Diskgroup We will first evaluate how many cameras can be supported using 14 disk RAID 5 equipped disk array with a single LUN. Figure 6 shows the total frame loss percentage for various FPS and camera combinations: With a single volume supported by a single, 14 disk RAID 5 diskgroup, the Intransa StorStac BuildingBlock system supports more than 55 cameras at 30FPS with 4CIF and MJPEG, resulting in less than 5% total frame drop. This corresponds to about 57MBps throughput from the cameras to the storage system. 14 Overall performance can be monitored through the Intransa StorStac Graphical User Interface, or from the NVR. Examples are shown below in Figure 7. Two LUNs Using Single 14 Disk RAID 5 Diskgroup With 2 LUNs per 14disk RAID 5 configuration, Figure 8 shows the cumulative percentage of frame distribution from 2 NVRs 15 With 2 LUNs from 14 RAID5 STAT disks, the storage system is able to support 64 cameras (30FPS, 4CIF MJPEG) with 2 NVRs (each with 32 cameras, or 960 FPS). The total frame loss is ranging between % and 10% cumulative percentage is ranging between % (less than 3% of samples have more than 10% frame losses). This translates to about 1920 total FPS, or about 60MBps total throughput. You can also view performance from StorStac GUI as shown below. The performance statistics includes throughput (MBps), IOPS and latency. The quality of the video is reflected on the smoothness of the performance. Figure 9 below shows an example of the StorStac performance per LUN. Figure 9: Intransa StorStac shows performance statistics for 2 LUNs. Four LUNs With Two 7 Disk RAID 5 Diskgroups With 2 diskgroups, each with 7 disks and RAID 5, in a total of 4 LUNs were able to deliver 72 cameras (30FPS, 4CIF MJPEG) with 2 NVRs. Disk/File Level Fragmentation Disk and file level fragmentation is another factor that must be considered for performance. 16 In the live environment, the Test Engineering team ran out
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks