Removing the wall between system and storage administrators
Sometimes we simply don’t know what we don’t know. Arguably in many IT shops, departments are either extremely siloed or very broad. Some environments are “towered,” employing separate departments for managing storage and systems administrators who look after the operating system. Other shops have one person responsible for storage and server configuration with little opportunity to develop deep skills on either pillar. Organizations need a cross-pollination of information to burst open the silos.
In this article, our aim is to facilitate communication across organizations with separate OS and storage-administration groups, but hope others will find the information useful as well. This article slants toward AIX* but you can extrapolate high-level concepts for other OSs.
Share and Share Alike
We’re not trying to advocate turning systems administrators (SAs) into expert storage administrators (StorAdmins) or vice versa. But each side requires some base knowledge of the other. Sometimes it’s a matter of speaking the same language. If the SA knows enough about storage, requests will be phrased in a way that helps avoid obvious mistakes—such as placing data and logs on the same spindles so concurrent access becomes a problem, impeding good performance. Likewise, the StorAdmin should know enough of the SA role to query for complete information that avoids double mirroring, incorrect application of RAID or similar problems.
We wish we could tell you that SAs need never know the physical device characteristics of the I/O subsystem nor involve themselves in understanding the RAID configuration. We can’t. A basic understanding of the infrastructure will help avoid many of the performance issues that so often surface with misconfiguration, not fully exploiting redundancy, disk isolation vs. storage pooling and improper striping.
SA knowledge that must be communicated to the StorAdmin includes access requirements (e.g., sequential vs. random access and read/write characteristics), backup schedules, allocation of data and log pairs (accessed concurrently) and if the data is mirrored or striped at the OS level. Working together can minimize job-scheduling conflicts and prevent problems that can arise from simultaneous requests, such as writing backups during heavy online periods. In short, it’s imperative that the advanced IT shop share information across towers.
Virtualization adds another dimension that must be made transparent. When a Virtual I/O Server (VIOS) and high availability (HA) are involved, or anything that needs redundant allocations, such as FlashCopy* Manager (FCM) or global mirroring, considerations must be made for keeping configurations in sync and allocating LUNs not only for efficiency but also to meet application, availability and replication requirements. Further, if an HA solution is running, visibility must be retained from either HA pair and changes must be cluster aware. The SA will be in a unique position to understand the typical tuning knobs the OS must specify, such as queue depths, device driver levels (which are controlled by the OS) and VIOS code levels.
If the storage expert and the OS expert share information and produce vital documentation together, the environment is less vulnerable to performance woes.
Search our new 2013 Buyer's Guide.
E-Newsletter | Removing the wall between system and storage administrators
E-Newsletter | IBM enhanced storage services take care of the details