Comment Re:Maybe a Network Appliance Netfiler instead? (Score 2) 149
Netapps have prove to be very reliable in my experience -- I've used a few of their filers at an ISP -- grinding 24x7 for web, mail, shell, etc.
That said, I wouldn't necessarily recommend them for this application. NFS does *NOT* scale for most applications. It's great for distributing low-to-medium workloads, but I'd question the ability of the box to perform for such a large data set. (mpg streams) One of the major benefits of NetApps is the internal caching, which works very well for read-intensive applications. In some benchmarks, and occasionally in real life, I've seen data return from a netapp faster than from local 7200 rpm disks. However, with multi gigabyte streams, the cache will be worthless.
My only personal experience with FC-AL has been with Sun E4500's and MTI Raids.. similar to EMC, I'd imagine. They work well, and using Veritas products you can do high-availablilty and easy data relocation.. something to think about long term. (i.e. the cost for a mid-range quad proc AXmp, ~$35-45k could be offset in the future when you need to scale..)
I guess to summarize: for workloads like web serving, Netapps work great. For email, locking presents issues -- its not bad with Solaris, but its a joke in linux. For the initial application of large mpeg streams, I'd go with an FC-AL raid array, using dual loops on dual fail-over hosts. If availability isnt an issue, then 1 host, 1 loop. (And if performance is much more important than cost, use switched Fibre Channel)
-lynch
That said, I wouldn't necessarily recommend them for this application. NFS does *NOT* scale for most applications. It's great for distributing low-to-medium workloads, but I'd question the ability of the box to perform for such a large data set. (mpg streams) One of the major benefits of NetApps is the internal caching, which works very well for read-intensive applications. In some benchmarks, and occasionally in real life, I've seen data return from a netapp faster than from local 7200 rpm disks. However, with multi gigabyte streams, the cache will be worthless.
My only personal experience with FC-AL has been with Sun E4500's and MTI Raids.. similar to EMC, I'd imagine. They work well, and using Veritas products you can do high-availablilty and easy data relocation.. something to think about long term. (i.e. the cost for a mid-range quad proc AXmp, ~$35-45k could be offset in the future when you need to scale..)
I guess to summarize: for workloads like web serving, Netapps work great. For email, locking presents issues -- its not bad with Solaris, but its a joke in linux. For the initial application of large mpeg streams, I'd go with an FC-AL raid array, using dual loops on dual fail-over hosts. If availability isnt an issue, then 1 host, 1 loop. (And if performance is much more important than cost, use switched Fibre Channel)
-lynch
