We quite often get asked questions about what levels of hardware need to be spec’d out for various institutions to setup their Podcast Producer systems – and the answers really boil down to how much can you afford, and what do you want to prioritise? It’s also a little like asking how long is a piece of string.
Podcast Producer systems can be run on any Mac hardware that supports OS Server 10.5 or 10.6. I’d say just focus on the 10.6 requirements. To put this into perspective, we’ve run systems on both a single Mac Mini, to a small cluster of Mac Minis and graduated up the scale to our present setup of a small XSan with 4 Xserves. Recently at a workshop we setup a two node cluster using a Mac Pro and and Macbook!
The following notes are mostly observations from our own experiences and those of others we’ve talked to. At the end of the day, what you need compared to what you want, compared to what you can afford will determine what you get.
Single Box solutions
We’ve run a complete PcP system on a single Mac Mini (Core 2 Duo 2Ghz, 2Gb RAM, 160Gb HDD, Intel GMA 950 with 64Mb VRAM), and it works fine, though very very slowly. For each hour of submitted content we would need to allow 3 hours of processing time. Moving up to bigger systems doesn’t necessarily equate to greater gains in turnaround time, but does improve overall throughput (a factor to consider when you’ve got multiple jobs being submitted simultaneously). To that end, the newly released Mini Servers are a nice starting point and it is good to see Apple begrudgingly accepting that these small devices are very practical.
Typical PcP workflows may have 4 or so tasks that *can* run in parallel (most of the time it is a serial list of tasks), so anything over 4 processing cores will see improvements in terms of concurrent processing and therefore throughput. Higher CPU ratings will be a significant factor here in terms of turnaround rate (how long it takes to get something out from something put in) as will copious quantities of RAM, and also a good spec GPU will help (some workflows use Quartz transformations which are rendered on the GPU rather than the CPU).
Mac Pros can be an option here, especially if you’re wanting to cram in as much (cheapish) disk storage as possible (they hold more disks than any other Mac system). However, they are rather pricey, and don’t really scale easily/cheaply beyond the single box setup, nor fit into server room racks.
If you’re looking at machines for the server room, then even the base level Xserve (single Quad Core Nehalem 2.26GHs, 3GB RAM) is going to be fine for handling light loads (e.g. 12 hours of content per day – though really, this is a crude approximation and metric because this is about batch processing, and as long as disk space allows, everything will get processed eventually). However, I’d recommend: 2 * 160Gb + 1 * 1TB HDDs, 6Gb RAM, and 2 PSUs. More RAM is good, though more than 12Gb is probably excessive at current prices. I’ll touch on this again in a moment.
Storage
Storage configurations will need considerations. If you’ve got a SAN, then you need to know can it be effectively accessed via OS X platforms, and by what mechanisms (Ethernet? Fibre Channel?). The data throughput you can get from your storage system to your processing system needs to be understood, and if not sufficient can be a significant bottleneck to processing rates.
Likewise, have you got a backup solution that is compatible with your OS X setup? Keeping podcasting data separate from the system data (e.g. OS and logs on one disk and PcP shared filestore on another disk) will help in terms of backup strategies. This topic alone is worth several long reports, but in short:
Find a way to back up your system disks. OS X’s ability to clone their drives to Disk Image files on another system makes for a simple and easy way to both backup and restore systems in cases of disk failure or system corruption.
Are you using your server as a hosting platform as well as a processing platform? Then you’ll want to back up the data partition/disk (shared filestore) to ensure your hosted outputs are not lost. If you’re having difficulties getting funding for this, remember to factor in the wages costs of everyone involved in recreating lost content, both in terms of restoration
of backups, or worst case, creating the content again from scratch. Now weigh this against the cost of various disks, tapes and backup solutions (and these range from the weak consumer level approach of disposable hard disks on shelves, through to multiple tape copies in many locations and tape system, etc).
This is a topic that really tells you how much your content is worth to your institution.
XServes again
For single xserve solutions, there’s really two choices to pick between in terms of disk configurations: 2 * 160Gb + 1 * TB, or 1 * 160GB = 2 * 1TB. Your choice here is likely to be influenced by your backup options and where you consider the most risk to your setup to exist. I’d strongly avoid considering a 3 * 1Tb setup, perhaps thinking that you can RAID them all into one large (more) resilient setup. This will adversely affect system performance as a single volume is not well suited to both many small read/write operations (e.g. log files, system configs, database accesses) at the same time as large read/write operations (multi GB data files, i.e. videos). Our preference is the 2 * 160 setup, with the two smaller disks in a RAID 1 setup for the system partition, and the 1 * 1TB drive as a scratch disk (PcP Shared volume).
Whilst I think about it, also consider:
The SSD option is a luxury that’s probably not worth it at current prices. Likewise the RAID card (the onboard RAID is fine).
Dual Power supply is nearly always a good idea – though only if you’ve got a separate power feed to take it from (e.g. one from Raw mains, one from a UPS, or two separate Raw supplies). Having a spare to hand is good if you’ve only got the single PSU in the box, or when you’ve got multiple machines. Whilst XServe hardware is pretty damn reliable, the most common (of rare events) is for either a PSU to die, or one of the HDDs. Oh, and the HDDs Apple use are not regular desktop drives, hence the seemingly obscene costs compared to consumer hardware.
Ensure you’ve got the relevant video adaptor dongle – you’ll need to be able to plug it into a monitor occasionally.
If you can, take out the Apple Care Extended warranty. It at least covers parts for 3-4 years and provide some technical support.
Thinking ahead
If you’re looking at serious infrastructure, you’re probably working towards an XSan and multiple Xserve setup, and in that instance you might want to have that future in mind when you get your first system. As always, your milage may vary, and my answers to your specific setup needs may likely vary significantly from these notes. I hope though that you might gain a little insight into some aspects to consider about the physical infrastructure for an OS X Server based system.
Carl
Permalink
| Leave a comment »
