How many streams at the same time ?

[Bee]

can somebody give the rule for knowing the maximum of concurent streams with one server (Intel IMac for example) over a 100BT ethernet, with mp3 files (256kbps) ... or other configurations ?

[ceejay]

can somebody give the rule for knowing the maximum of concurent streams with one server (Intel IMac for example) over a 100BT ethernet, with mp3 files (256kbps) ... or other configurations ?

Well, at the top end: a 100Mbps LAN should give you about 50Mbps of usable throughput, so divide 50,000 by 256 to get 195 theoretical streams.

Other bottle necks include the capacity of your hard disc to deliver bits (should be more than 50Mbps so not the limit) and the ability of your slimserver to keep up (which will probably be the limiting factor in practice - so depends on processor speed, system memory and other tasks running).

This of course is assuming no transcoding on the server side.

This is a long way of saying there is probably no magic formula which will tell you what you really want to know. Why not say how many you were thinking of, and see if anyone can offer evidence of matching that?

There was also a thread a while ago on how many streams people had managed, and I seem to remember the answer was tens.

HTH
Ceejay

[rudholm]

Well, at the top end: a 100Mbps LAN should give you about 50Mbps of usable throughput, so divide 50,000 by 256 to get 195 theoretical streams.

50Mbps?

I'm curious why you don't expect 100Mb/second from your 100Mb/second Ethernet.

[funkstar]

error correction and protocol overhead.

You will *never* get 100mbit of through put on a 100mbit network. The speed quoted is the wire speed as opposed to the data transfer speed.

Gigabit gets round this to some degree by using jumbo frames. Basically this means that your data to overhead ratio is better as each frame being sent contains more data.

[pfarrell]

Re: How many streams at the same time ?

rudholm wrote:
> ceejay Wrote:
>
>>Well, at the top end: a 100Mbps LAN should give you about 50Mbps of
>>usable throughput, so divide 50,000 by 256 to get 195 theoretical
>>streams.
>
> 50Mbps?
>
> I'm curious why you don't expect 100Mb/second from your 100Mb/second
> Ethernet.

Because Ethernet uses Carrier Sense Multiple Access with Collision
Avoidance to control traffic. No Ethernet can deliver its rated
speed, even getting 50% of its rated speed is hard if not impossible if
there is any sharing of the network.

And then you have to add overhead, addressing, error correcting, ack/nak
messages, etc.

For a practical example for Slim users, a WiFi 'b' network is
rated as 11 megabits/second but it rarely can actually
delivery uncompressed audio. This is a problem for SB1 users,
since the SB1 only does WiFi B, and can't do Flac on the fly

The RedBook specifies that stereo music is to be recorded in 16-bit PCM
sampled at 44.1 kHz. A little arithmetic shows that the data stream of
CD audio must be at least:

Data Rate = "X"
= 16 bits * 2 channels * 44.1kHz
= 16 * 2 * 44100
= 1,411,200 bits/second
= 176,400 bytes/second
= 172 Kbytes/second

So while Red Book (or uncompressed audio) is only 1.4 mb/s and that
is only 12% or so of a WiFi "b" link speed, you can't get it.

Assuming that you can get 10% of the rated speed is probably good
engineering.


--
Pat

http://www.pfarrell.com/music/slimserver/slimsoftware.html

[Jacob Potter]

Re: How many streams at the same time ?

On 6/4/06, Pat Farrell <pfarrell (AT) pfarrell (DOT) com> wrote:
> Because Ethernet uses Carrier Sense Multiple Access with Collision
> Avoidance to control traffic. No Ethernet can deliver its rated
> speed, even getting 50% of its rated speed is hard if not impossible if
> there is any sharing of the network.
>
> And then you have to add overhead, addressing, error correcting, ack/nak
> messages, etc.

Only hubbed Ethernet uses CSMA/CD; switched is full-duplex. I've
benchmarked over 95 Mbps using generic consumer cards and switches,
although that was card-to-card (netperf), not disk-to-disk. It's not
that hard.

- Jacob

[radish]

rudholm wrote:
> ceejay Wrote:
>
>>Well, at the top end: a 100Mbps LAN should give you about 50Mbps of
>>usable throughput, so divide 50,000 by 256 to get 195 theoretical
>>streams.
>
> 50Mbps?
>
> I'm curious why you don't expect 100Mb/second from your 100Mb/second
> Ethernet.

Because Ethernet uses Carrier Sense Multiple Access with Collision
Avoidance to control traffic. No Ethernet can deliver its rated
speed, even getting 50% of its rated speed is hard if not impossible if
there is any sharing of the network.

And then you have to add overhead, addressing, error correcting, ack/nak
messages, etc.

For a practical example for Slim users, a WiFi 'b' network is
rated as 11 megabits/second but it rarely can actually
delivery uncompressed audio. This is a problem for SB1 users,
since the SB1 only does WiFi B, and can't do Flac on the fly

The RedBook specifies that stereo music is to be recorded in 16-bit PCM
sampled at 44.1 kHz. A little arithmetic shows that the data stream of
CD audio must be at least:

Data Rate = "X"
= 16 bits * 2 channels * 44.1kHz
= 16 * 2 * 44100
= 1,411,200 bits/second
= 176,400 bytes/second
= 172 Kbytes/second

So while Red Book (or uncompressed audio) is only 1.4 mb/s and that
is only 12% or so of a WiFi "b" link speed, you can't get it.

Assuming that you can get 10% of the rated speed is probably good
engineering.


--
Pat
http://www.pfarrell.com/music/slimse...msoftware.html

Whist I agree that you don't get 100%, I think your numbers are somewhat on the pessamistic side. I get over 80% of theoretical on my gigabit lan without any tweaking, without jumbo frames, and with the cheapest lan cards I could buy ($10 each). Not bad. As for your 802.11b - again the myth that you can't run pcm to an sb1 is just that - a myth. I was doing it for a long time before I finally replaced my sb1 with a later model. Sure there was the odd glitch because that buffer was so small but in general it worked fine. This was with a normal consumer access point through several walls and with less than 80% signal strength in a noisy (rf wise) apartment building.

[pfarrell]

Re: How many streams at the same time ?

radish wrote:
> pfarrell Wrote:
> - again the myth that you can't run pcm to an sb1 is just that - a myth.
> I was doing it for a long time before I finally replaced my sb1 with a
> later model. Sure there was the odd glitch because that buffer was so
> small but in general it worked fine.

My experience is different. I found PCM over wireless to my SB1 to be
unlistenable, hence unusable.

The fix was easy, drag an Ethernet cable.


--
Pat

http://www.pfarrell.com/music/slimserver/slimsoftware.html

[rudholm]

With 802.11b or g, yeah, I'd be happy with 50% of advertised bitrates, but the topic at hand is 100Base-TX (i.e. wired) Ethernet.

Even with the preamble byte, the SFD byte, the source and destination MAC addresses, the frame type, and the checksum, you're still only talking about about 20 bytes of overhead per frame, which isn't much even with 100Base-TX's MTU size of 1500 (i.e. not the "jumbo frames" that Gb Ethernet often uses).

None of that could possibly account for a 1:2 ratio between user data and Layer 1 data.

And my experience bears this out. I normally see about 10.5 megabytes per second transfer rates on 100Base-TX links, which is quite close to 100 megabits per second.

So my question to ceejay stands --I'm curious why he expects only 50 megabits per second from 100Base-TX.

[Mark Lanctot]

Not to beat up on poor ceejay, but I tested this once for a review of a router:

http://www.abxzone.com/forums/attachment.php?attachmentid=17528&d=1107717661

See page 8. I found the "100 Mbps" connection averaged 82.17 Mbps. Not quite 100 Mbps, but not 50 either.

BTW that "RacoonWorks Speed Test" is a pretty cool little free program: http://www.snapfiles.com/get/speedtest.html You can test server -> client on your internal network, wired or wireless, or between a web page and a local PC.

A more configurable command-line speed tester that can be incredibly powerful and accurate is Iperf: http://dast.nlanr.net/Projects/Iperf/

[ceejay]

Not to beat up on poor ceejay, but I tested this once for a review of a router:

Well, fortunately I have quite a thick skin so don't feel terribly beaten up!

The point I was trying to make in a very short post was that you probably won't get 100% usage from a LAN. I offered 50% as a rule of thumb that has worked well for me.

I'd agree that a short point to point will get much closer to 100% because you really only have to worry about some overhead.

But when you start to run long cables, or introduce multiple devices on the network, that figure drops. In my case, I know that if I'm doing a sustained long copy across my network (when backing up my music library, for example), 50% is pretty much what I get.

Let's also not forget the context of the original question - "how many streams can you have?" Which implies multiple devices and contention as being relevant....

YMMV !

Ceejay

[rudholm]

Well, fortunately I have quite a thick skin so don't feel terribly beaten up!

The point I was trying to make in a very short post was that you probably won't get 100% usage from a LAN. I offered 50% as a rule of thumb that has worked well for me.

I'd agree that a short point to point will get much closer to 100% because you really only have to worry about some overhead.

But when you start to run long cables, or introduce multiple devices on the network, that figure drops. In my case, I know that if I'm doing a sustained long copy across my network (when backing up my music library, for example), 50% is pretty much what I get.

Let's also not forget the context of the original question - "how many streams can you have?" Which implies multiple devices and contention as being relevant....

YMMV !

Ceejay

Multiple devices can drive down your aggregate bit rates when there are collisions but I wouldn't expect to see collisions in a switched environment.

Neither would I expect to see long-ish cable runs costing much, unless you're pushing the rating limit (100M) or have out of spec cabling. If you're seeing 50Mb on your backups, you might have some other issue like bad cable/jacks or I/O limitations on your hosts. If your network is really bottlenecking you at 50Mb, something ain't right...

[Bee]

thank you for the discussion about the security margin.
I am sure that we can exceed 80 mb/s on a switched environment
but counting on a 50mb/s seems to be a safe limit

... which in fact is not a limit in this case

I worry more about the ability of the server and of the hard drive to deliver 6 to 8 signals at the same time.

Of course I assume that no transcoding will be done on the server side,.

Did somebody tried this successfully already?

[rudholm]

thank you for the discussion about the security margin.
I am sure that we can exceed 80 mb/s on a switched environment
but counting on a 50mb/s seems to be a safe limit

... which in fact is not a limit in this case

I worry more about the ability of the server and of the hard drive to deliver 6 to 8 signals at the same time.

Of course I assume that no transcoding will be done on the server side,.

Did somebody tried this successfully already?

I've had about four or five playing at once with no problem on a 2.4GHz Pentium 4 running a 2.4 linux kernel. I've never had occasion to try to run more than that many players simultaneously.

I think if you avoid transcoding you should be able to support a lot of players simultaneously.

[mherger]

Re: How many streams at the same time ?

> I worry more about the ability of the server and of the hard drive to
> deliver 6 to 8 signals at the same time.

6-8 clients isn't a number you should worry about. I'm running five
devices on a low-power Via C6/1000 with absolutely no problem at all.
Sometimes I'm even using AlienBBC (only on one at a time, though).

--

Michael

-----------------------------------------------------------
Help translate SlimServer by using the
StringEditor Plugin (http://www.herger.net/slim/)