Screamernet PC's Hard Disk

[yaschan]

I'm thinking to set up this screamernet PC to use as a render node for Screamernet.
I'm concerned about the electricity consumption of this machine.
Spinning plate hard disk will consume more electricity than a flash disk doesn't it?
Would it make a sense (or is it even possible, for that matter) to install Windows 7 (64 bit) to a memory stick and run from it? Can this cause any potentional bottlenecks what comes to Screamernet rendering?

[Lightwolf]

I'm concerned about the electricity consumption of this machine.
Spinning plate hard disk will consume more electricity than a flash disk doesn't it?
That should be the least of your worries though. A SSD uses roughly 1W, a hard disk between 6 to 15W or so (depending on the make and speed).
However, the rest of the system will use at least 30W and that's only when it runs idle, as soon as you're rendering you'll be around 150-180 easily (on a modern machine).

Cheers,
Mike

[3dWannabe]

Possibly a more interesting question would be the speed increase one might get from using SSDs.

This article is a bit Mac-centric, but it's comparing the speed of the drives against SSD drives at the end:

http://www.barefeats.com/hard135.html

I'm really interested in how to setup a low cost render box.

What gives the most bang for the buck?

I also wonder how a small RAM disk could be used to speed things up?

I used to compile using a RAM disk years ago, and the speed increase was just incredible.

[Hopper]

I'm really interested in how to setup a low cost render box.
Then you won't be doing it with those... high speed SSD's are running $600+ for ~200GB.

And for a render farm, your bottleneck isn't your disk I/O.. it's CPU/RAM.

[3dWannabe]

Then you won't be doing it with those... high speed SSD's are running $600+ for ~200GB.

And for a render farm, your bottleneck isn't your disk I/O.. it's CPU/RAM.
Compared to what I spent on RAID cards and drives, that actually is low cost.

But, very good point about CPU and RAM.

Curious whether bridging the NICs gains much speed, or whether a small RAM disk would be useful in some way.

[Lightwolf]

Curious whether bridging the NICs gains much speed, or whether a small RAM disk would be useful in some way.
Any remotely complex scene is going to tax your CPU more than anything else.
Unless you really have a massive amount of nodes (100s as opposed to 10s) i/o isn't much of a problem - especially as lwsn tends to loads static items (lwos and images) only once and then re-uses that for all frames of the same scene rendered on the same node (lwsn -2 that is. -3 does as well, but you have to be able to render a bunch of consecutive frames on the same node).

And a fast disk isn't going to help you either as well. If LW has to swap out due to memory then you have a massive problem anyhow, you're better of getting more RAM.

Small disk, fast CPU, tons of RAM, 1 GBit NIC...
On the server side you may want to either team NICs or even switch over to 10GB depending on the size of the network though. That's your first i/o bottleneck.

Cheers,
Mike

[3DGFXStudios]

I'm thinking to set up this screamernet PC to use as a render node for Screamernet.
I'm concerned about the electricity consumption of this machine.
Spinning plate hard disk will consume more electricity than a flash disk doesn't it?
Would it make a sense (or is it even possible, for that matter) to install Windows 7 (64 bit) to a memory stick and run from it? Can this cause any potentional bottlenecks what comes to Screamernet rendering?

No I wouldn't make any sense to do that.

[Hopper]

Curious whether bridging the NICs gains much speed, or whether a small RAM disk would be useful in some way.
Bridging two NICs will not increase your speed, however it will certainly increase the rate at which your headaches occur.

If you bridge via LACP, you're only increasing your overall available bandwidth, but not on a per connection basis. A single connection will never use more than one interface. Layer 4 of the OSI model (i.e. transport layer) won't allow it. With LACP, your maximum group count is 4.

If you want to use ICS or a TAP interface, just don't. If someone attempts to convince you that it's a good idea, just run. You're not gaining anything but an administrative nightmare.

If you are running a server OS, you can use Teaming, but the only thing that will get you is a throttled/dedicated input and output path for traffic. It would be the same as setting up a dedicated input (or output) at 50% using QoS on a switch or router.

[3dWannabe]

Bridging two NICs will not increase your speed, however it will certainly increase the rate at which your headaches occur.

If you bridge via LACP, you're only increasing your overall available bandwidth, but not on a per connection basis. A single connection will never use more than one interface. Layer 4 of the OSI model (i.e. transport layer) won't allow it. With LACP, your maximum group count is 4.

If you want to use ICS or a TAP interface, just don't. If someone attempts to convince you that it's a good idea, just run. You're not gaining anything but an administrative nightmare.

I think you know a good deal more about networks that I do.

Are you saying that if I have a switch on each bonded NIC (2 networks), that I won't get any speed increase, but will increase my headaches?

Or is that only if you connect both NICs to the same switch?

By 'single connection' I think you are saying that it really doesn't matter, you're not going to increase your bandwidth per connection.

But, wouldn't file sharing be using multiple connections?

Thanks! An increase in speed does nothing if it increases my hassles, and less than nothing if it won't even increase my speed.

[Hopper]

I think you know a good deal more about networks that I do.
In a not so recent previous life, I was a developer for Cisco. There's nothing like being surrounded by super-geeks to humble you into a corner. There was so much to learn from those guys on a daily basis, it made my brain hurt.

Are you saying that if I have a switch on each bonded NIC (2 networks), that I won't get any speed increase, but will increase my headaches?
In short, yes. Any given connection will still use only one NIC. Mult-homed systems are fine, but you have to be very mindful as to which NIC any given piece of software will bind to. Multi-homed system + NetBIOS = instant problems if you're using Windows. If you're using strictly IP based applications (not using conventional UNC naming), it's very do-able, but still requires a bit more planning and administration.

Or is that only if you connect both NICs to the same switch?
Two NICs connected to separate switches is usually not an issue because each NIC will be on a completely different subnet. You will still have to specifically bind any server type apps (i.e. services) to a specific NIC (some apps allow binding to multiple networks but not many).


By 'single connection' I think you are saying that it really doesn't matter, you're not going to increase your bandwidth per connection.
Yessir. You increase your bandwidth limitation, but not speed.

But, wouldn't file sharing be using multiple connections?
File sharing can use as many connections as your OS will allow, but any one connection will only use one NIC unless your Teaming (and that's an entirely different can of worms).

Thanks! An increase in speed does nothing if it increases my hassles, and less than nothing if it won't even increase my speed.
I just need to clarify that as the subject relates to ScreamerNet, I think that is a fair statement. If you were hosting a single system requiring many user connections and maximum bandwidth as to not limit response time and throughput, multiple NICs is certainly a normal path to take.

[3dWannabe]

In a not so recent previous life, I was a developer for Cisco.
You probably saved me some unnecessary hassle! Thanks!