-
SAN discussion – volume vs file based – interesting question at the end
I’ve been using SanMP for about 4 months now, and have a few things on my mind. In this discussion I’m referring to SanMP since it is the only SAN I have personal experience with. It should be pretty close in operation to other volume-based SAN’s like Commandsoft’s fibrejet.
SanMP is a high performance SAN system using volume based management. This means all users can mount all volumes for read access at the same time, while only one user can write to each volume at a time. This model has some advantages: It does not need a metadata controller, and does not have a metadata controller’s overheads. Once the volume is mounted, the client is directly reading and writing from the array with no delays.
This system is quite stable provided you synchronise, or in most cases, unmount / remount the SAN volumes about once a day. Herein lies the kick: When a client mounts a volume read/write, it takes control of the file allocation tables and can write files and delete files at will. Great.
However when it mounts the volume in read-only mode, it reads a copy of the file allocation tables, and keeps it. Meanwhile, another client has write access and is merrily changing the volume: adding new files and deleting old ones. The read-only client does not know this, and can even continue reading and using files after the writer has trashed them… Once the writer overwrites a critical part of the data, the reader reads crap and crashes.
The reader has two ways to get an updated version of what is really happening on the volume: it can “synchronise” or unmount and re-mount a volume. Synchronise is SanMP’s process of reading an updated version of the volume without unmounting – it sometimes does not work without closing all apps (especially FCP) using the volume, and unmounting always needs all apps using the volume closed. There is also a setting to automatically sync every X minutes, but that process always leads to dropped frames and is thus unusable.
If the reader stays reasonably up to date (in our case about once a day seems fine), the system is really stable. It is also resilient to data transfer spikes: You all know that occasionally after using FCP’s “capture now”, FCP feels the need to read back the complete capture: It can capture merrily away at 20MB/s uncompressed SD, but as soon as you hit escape, it reads back from the drive at 185MB/s for about 30 seconds solidly. Sometimes it writes at that speed as well. I don’t know why this happens, I just see it happen occasionally on the performance monitor. The best part of this is that the other sanMP clients don’t know about it. They can continue editing their uncompressed work without a hiccup – something Im not sure a six seat XSAN installation on the same hardware – a single ADTX LH 15 drive array – would handle as well. Any system using gigabit ethernet would fall over.
I have used this SAN for a large project with tight deadlines and it has helped us a lot – digitising in four suites at the same time; playing out a 50 minute broadcast master and protection copy from two FCP suites at the same time 2 hours before broadcast, etc.
Now for some negatives: We are running four FCP suites and two Pro Tools final mix suites on this SAN. We often need to send a quicktime video file and OMFI audio file from the video suites to the audio suites instantly – this is one of the main reasons for getting the SAN. Using a volume based SAN, the audio suites must always synchronise, and often, when that fails, close all open apps using these volumes, unmount/remount and then re-launch all apps before they can start working. This also happens when we export a reference quicktime movie for immediate playout on another suite – we need to close FCP on the second suite before we can play out.
There is no workaround on a volume based SAN – it causes frustrating delays that we would not have on a file-based SAN.
The second issue is more about thinking ahead: We do quite a fair number of DVD’s for approval and final deliveries. And in many cases the client comes in for a viewing, likes what he sees and wants a DVD made immediately for further approval back at his office. Well that’s what I would like to use compressor 2’s distributed rendering for. But using our SANMP there is no way.
I currently use compressor like this: Inside FCP I export the FCP timeline as a reference quicktime movie. I switch to finder and drag the exported reference file onto compressor, choose a mpeg-2 preset and submit.
The dual G5 machines can then actually continue editing in FCP while compressor is chugging away in the background – the only problem being that it still takes a while (about 1.5 x real time) to do the compression to mpeg 2. On a single machine.
My thinking is this: I have 4 dual G5 machines – if I can let them help compressing in the background, I can cut the compression time by at least two thirds.
My problem with sanMP would be this: After exporting the quicktime reference movie, all other machines that want to help with the render would need to synchronise or re-mount the source volume – impractical already. That takes care of reading the file, but what about writing? All the machines that help with the compressing would need to write to one volume at the same time – and that is not going to happen with any volume based SAN soon.
The reason it’s on my mind like this is that I suspect the next major revision of FCP will include network rendering in the same fashion as compressor 2. And I don’t want to miss out on that because I’m on a volume based SAN.
So after baring my sould like this, I have two questions:
– Anybody using Apple XSAN or especially Tiger Technology MetaSAN on a similar setup: six seats doing 70% DV and 30% uncompressed SD work. Please discuss general operations, and what advantages or disadvantages you see in everyday operations? I would also like to hear from a MetaSAN user – how does Metasan run real-world loads without a dedicated metadata controller?
– Pro Tools will not run on any file-based SAN. If I decide to go to XSan or Metasan, with metadata controller, would it be possible to run SanMP on some LU’s and XSan (or metasan) on other LU’s on the same box?Regards
Francois