Technology Blog

LeoStor PACS film archiving storage system configuration scheme

LoongShine Network Copy Link
Abstract: With its excellent small-file data throughput and easy to use horizontal expansion, LeoStor surpasses the traditional NAS over FC-SAN, IP-SAN and open-source Ceph, and can be widely used in PACS production and archive storage systems. Its excellent parallel file read and write performance meets the needs of AI imaging analysis, making the system more cost-effective.

Overall requirements

According to relevant national laws, PACS film data needs to be kept for 15 years and can be accessed by clinicians at any time. Therefore, PACS archive storage has the following characteristics:

  1. Horizontal expansion capability: because PACS data is growing every day, based on the consideration of construction cost, a simple, effective and non-stop horizontal expansion scheme is required;
  2. Massive file retrieval: It is possible to generate tens of billions of files in 15 years, and it is necessary to find the required PACS files in the massive files;
  3. Storage access protocol: Due to the rapid development of storage protocols, unified storage of NAS, objects and future protocols needs to be comprehensively considered;
  4. Fault self-management: the whole system should self-diagnose, self-heal and reduce human intervention;

Still, taking a Class III hospital as an example, the daily incremental data of 500G and 2 million files need to meet the simultaneous failure of at least two nodes or hard disks.

Item Data 15 years
Amount of data written 500GB/day About 2.7PB
Number of documents 2 million/day About 6 billion
Number of directories 40000/day About 120 million

MDS node configuration

Increasing the number of metadata nodes can significantly improve the efficiency of files quickly retrieved from 10 billion files. Based on the characteristics of PACS film data, it is recommended to add a pair of MDS metadata nodes for each additional 1 PB of data; Other, improving the CPU's dominant frequency can shorten the execution cycle of each instruction. It is recommended to configure a CPU with a dominant frequency of more than 3.0 GHz; The use of U.2 NVMe SSD can improve the read and write of concurrent data.

Item To configure
Number of nodes 2 sets
CPU 2 * Intel 4215R CPU
Memory 128GB
Metadata disk 2 * 1.92TB U.2 NVMe SSD HBA
Network 2 * 10GE SFP+ / 25GE SFP28

MDS nodes need to be added in pairs with consistent storage capacity; Upgrade the MDS disk. The capacity of the new disk is no less than that of the old disk. When the system works normally, pull out the old disk and insert the new disk to complete the upgrade; At the same time, when the system works normally, cut out the old node and add a new node, that is, update the node.

OSS node scheme

Calculating the number of nodes from the target capacity is a common calculation method for archive storage. In this case, it is recommended to use the LeoRaid 4+2:2 (N+M: M) redundancy mode, with the availability of 66%, and the raw capacity=2700TB ÷ 66%=4090TB; LeoStor has an excellent redundancy technical solution. Based on the consideration of construction cost, DF1200 2U12 disk device can be used at the initial stage of construction. When the data increases to a certain amount, DF3600 4U36 disk high-density storage server can be used for expansion or replacement. In this case, 10TB ST SATA disk is used, and the redundancy model LeoRaid4+2 is used. We implement the strict 8-node model, so the system is initially configured with 8 storage nodes.

Single hard disk capacity Number of hard disks 12 drawer device 24 drawer device 36 drawer device
8TB 512 43 22 15
10TB 409 52 18 12
12TB 341 29 15 10
14TB 293 25 13 9
16TB 256 22 11 8

Note: The green area represents optional, and the Seagate X16 SATA enterprise disk is recommended for mechanical disk.

In this project, it is recommended to configure eight storage nodes at a time, which does not mean that there can only be one redundancy. For all redundancy policies that meet the requirements of eight nodes, the directory can be configured separately; At the same time, it is not necessary to fully configure all hard disks, but only to configure them on demand; Because the price of hard disk will continue to decline, it can generally be based on all the original data capacity plus the increment of the next year, which can lower the overall investment. In addition, there are hard disks in different periods in the system, and the probability of failure of more than M hard disks is reduced.

If you need to update the hard disk, you can unplug the hard disk or cut out the number of nodes at the same time, which is the lowest M value in multiple redundancy strategies. Before updating, you need to confirm that the system is in normal working state. The capacity of the updated hard disk can be less than the capacity of the old disk, but it is recommended that it is not less than.

  • Standard

    1. Scheme model: DF3600-E
    2. Storage node: 4U 3.5" 36 drawer device
    3. Number of storage nodes: 8
    4. Storage hard disk: 24 * 10TB ST X16 SATA disk
    5. CPU: 2 * Intel 4210R
    6. Memory: 64G
    7. Network: 2 * 25GE SFP28
    8. LeoStor capacity licensing
    9. LeoStor Object storage software license
    10. LeoStor LDM Archiving Software License
  • AI Analytical

    1. Scheme model: DF3600-E
    2. Storage node: 4U 3.5" 36 drawer device
    3. Number of storage nodes: 8
    4. Storage hard disk: 24 * 10TB ST X16 NL-SAS disk
    5. CPU: 2 * Intel 4215R
    6. Memory: 256G
    7. GPU:NVDIA V100/RTX
    8. Network: 2 * 25GE SFP28
    9. LeoStor capacity licensing
    10. LeoStor LDM Archiving Software License

Support 25GE switches such as Huawei, Huasan, Maple, and Suntech. It is recommended to select Suntech E680-48Y8C, which has 48 10GE/25GE and 8 100GE ports.