Block Storage

Attached disk

First and foremost, streaming VOD is an I/O challenge: the faster the disk the faster the origin can push content into the network.

Typically a combination of SSD (or even RAM disk), attached disk and NFS is used. Please see the Cloud Storage and Cloud Storage Providers sections for an overview of HTTP backed storage.

For an overview of in-use setups including storage behaviour please have a look at this whitepaper.

Network filesystem (NFS)

NFS is a distributed file system protocol originally developed by SUN, many storage system offer NFS connectivity - please refer to your storage system manual.

If for some reason the NFS server-client setup is not consistent/trustworthy (for instance a sub-standard driver implementation), memory-mapped I/O can be turned off. Please see the Apache All virtual host related options section for details.

In general the read/write blocksizes should not be set too high: 8, 16, or 32k are possible good settings. 1M has been found as too large, leading to a lot of unnecessary traffic between NFS client and server.

Dell EMC Isilon / OneFS

EMC Isilon is an advanced storage platforms with settings for the most common, but different use-cases.

The access pattern for streaming video is 'random', so it is recommended to set the storage to the recommended settings for this pattern.

The administrator can choose from the access patterns below at a per-file or directory-level.

Access pattern

Description

Concurrency

Optimizes for current load on the cluster, featuring many simultaneous clients. This setting provides the best behavior for mixed workloads.

Streaming

Optimizes for high-speed streaming of a single file, for example to enable very fast reading wi th a single client.

Random

Optimizes for unpredictable access to the file, by adjusting striping and disabling the use of any prefetch cache.

More information can be found in the EMC whitepaper.

NFS Settings

Secondly, it is also important to look at the NFS client settings (when using NFS). The read/write blocksizes should not be set too high (8, 16, or 32k are possible good settings, 1M has been found as too large, leading to a lot of unnecessary traffic between NFS client and server).

For example, in /etc/fstab:

10.10.40.112:/ifs on /mnt/filers/isilon type nfs (rw,noatime,tcp,bg,hard,intr,nfsvers=3,rsize=32768,wsize=32768,addr=10.10.40.112)