Green Storage for HPC with Solid State Disk (SSD) Technologies

Solid State Drives (SSDs) are emerging as attractive alternatives to traditional HDDs due to there reduced power consumption and potential for increased performance. SSDs have the potential to improve the performance of both Checkpoint Restart (CPR) and the Metadata Server (MDS) for high performance File Systems, e.g., Lustre. However, the write performance of SSDs is relatively poor (compared to read performance), the cost per byte is higher the HDDs, and the failure modes of SSDs are different then HDDs. Therefore, SSDs are not an automatic replacement of HDDs and the benefits and drawbacks must be carefully examined and accounted for in order to integrate SSDs into current applications. RNET and OSU will investigate the failure modes of various SDDs and compare the performance of checkpoint restart (CPR) and Lustre Metadata servers (MDS) with various SSDs and HDDs. In addition, mitigations techniques will be investigated to improve the performance of CPR and MDS with SSDs. Commercial Applications and Other Benefits: The use of SSDs has the potential to greatly improve performance will drastically reducing energy requirements. The use of SSD for checkpoint restart and Lustre MDSs will demonstrate many of the benefits of SSD and investigate the affects of the drawbacks and attempt to develop mitigation techniques. These mitigation techniques can be applied to other applications that would benefit from SSD. Therefore, the techniques developed in this project would not only benefits HPC labs using CPR and Lustre, but would have the potential to benefit a much wider audience of compute users in Academia, Government, and Industry.