Slow performance of WL-HDD - Discussion

[Oleg]

I do not recommend buying wl-hdd, it's VERY slow. Probably wl500gx with external HDD will be faster.

[Styno]

I'm do not recommend buying wl-hdd, it's VERY slow. Probably wl500gx with external HDD will be faster.

Wow, thats a new statement there Oleg! Looks like your IDE trials haven't come up with something better? Thats a big bummer for all WL-HDD owners.

But it makes the WL-500gx more interesting even more.

[blue]

thanx oleg for your advice - i will wait for the wl500gx - i hope it is available soon....

blue

[Oleg]

Wow, thats a new statement there Oleg! Looks like your IDE trials haven't come up with something better? Thats a big bummer for all WL-HDD owners.

But it makes the WL-500gx more interesting even more.

This is not an ide problem - ide works fine, but CPU is too slow to put this data to the network.

[brubber]

This is not an ide problem - ide works fine, but CPU is too slow to put this data to the network.

Why would the wl-500gx be any faster then the wl-500g?

Does anyone have reliable througput figures for wl-500g and WL-HDD?

Searching this forum I see a lot of different figures.

I have only tested wireless connection, and I get an average throughput of around 950 kb/s using NETBIOS over TCP/IP (about 7.5 Mbps) with 36Mbps+ wireless connection. Tested with over 5Gb of compressed data, multiple files ranging from 1kb to a few hundred Mb in size.

[Styno]

Because Oleg's experiments proved that datathroughput is currently CPU bound, and the WL-500gx has a 75MHz faster CPU, the WL-500gx will be faster even if the HDD is connected through USB2.0 and WL-HDD has IDE.

[brubber]

Because Oleg's experiments proved that datathroughput is currently CPU bound, and the WL-500gx has a 75MHz faster CPU, the WL-500gx will be faster even if the HDD is connected through USB2.0 and WL-HDD has IDE.

Is this also true for the Wl-500g compared to the WL-500gx?

Also I'm really surprised that the CPU is the bottleneck in data transfer speed. Is there any technical explanation or is this based on empirical findings?

[Styno]

Yes, its also true for WL500g because its practically the same hardware as WL-HDD.

These findings are empirical but reasonably theoratically proven so you can pretty much take it 'as is'. Ofcourse WL-500g is suffering from USB1.1 bus too.

See this thread:
http://wl500g.info/showthread.php?t=1037

[Antiloop]

Is this also true for the Wl-500g compared to the WL-500gx?

Also I'm really surprised that the CPU is the bottleneck in data transfer speed. Is there any technical explanation or is this based on empirical findings?

speed tests have proven this..

my wl-hdd has an internal speed of ~6megabyte/sec and ~50megabyte/sec cached

after Oleg has manipulated things and now we have achieved a maximum of ~2Megabyte/sec over network

better than 1,3mb/sec maximum, so it's a bit sad if Asus thinks this is a wireless harddisk with 54Mbps of bandwidth..

[brubber]

So is the 54Mbps only a nice feature on box or does it have some real meaning?

I agree with JOCKYW2001 (http://wl500g.info/showthread.php?p=6584#post6584) that this is probably an ethernet issue. It shouldn't be using that much CPU power at such low speeds, certainly not for packets that are destined for the HD

If the CPU speed is really limiting then I would expect only marginal improvements for the wl-500gx (please correct me if I'm wrong)

[Styno]

We're going waaaay off topic here But what the heck, its interesting talk...

If the network performance scales lineair whith CPU speed it will reach about 3 Mbyte/sec with WL-500gx (hardwired). This is not so surprising as the TCP stack is quite large, so lots of code has to be processed before a packed is received/transmitted.

Even modern servers are tested for CPU load in network benchmarks and it shows that even with these advanced processors the system is faily stressed while transferring 1 Gbps, so its not strange that a simple router cannot perform as well. For wireless it will be even worse because of the encryption which has to be done by the CPU as it has no hardware encryption feature.

Full blown network routers have specialized CPU's with hardware handling part of the TCP stack and doing all the encryption.

I wish Asus brings an XScale 400Mhz router to the consumer market, just to please the tweakers and power users

[Oleg]

Several facts about bcm4702. Sad facts.
It's tooks 1.5 cycles for command to complete, so the BogoMIPSes is 125*2/3 about 83.33 (reported 82.94). So, think about 83.33 MHz RISC CPU.

bcm4712, used in the wrt54g v.2 & wrt54gs runs at 200 MHz and has 200 BogoMIPS (one command per cycle, just how normal RISC CPU), so it's 2.5 times faster!

I expect wl500gx will be 2.5 times faster than wl500g.

And yes, CPU is performing much work - copying data, calculating CRC, traversing kernel code...

I've no idea, why ASUS has used bcm4702 in the wl-hdd...

Probably there will be wl-hdd v.2 or something.

[Antiloop]

We're going waaaay off topic here But what the heck, its interesting talk...

If the network performance scales lineair whith CPU speed it will reach about 3 Mbyte/sec with WL-500gx (hardwired). This is not so surprising as the TCP stack is quite large, so lots of code has to be processed before a packed is received/transmitted.

thread splitted for obvious reasons

about the WL500gx this is ofcourse dependent on the architecture used in the processor which will be used.. if we're lucky they will use an CPU which has some optimizations/accelerations for ethernet shit

was a bit late, see also post from oleg above me

[JOCKYW2001]

And yes, CPU is performing much work - copying data, calculating CRC, traversing kernel code...

Looking at the et driver code it seems to me that it can be optimised a lot

[Styno]

bcm4712, used in the wrt54g v.2 & wrt54gs runs at 200 MHz and has 200 BogoMIPS (one command per cycle, just how normal RISC CPU), so it's 2.5 times faster!

I expect wl500gx will be 2.5 times faster than wl500g.

I didn't expect broadcom would be using a different model of their MIPS CPU, I guessed they just boosted the speed a bit. Luckily they used a better design for their bcm4712 CPU. Good news indeed, the 2.5 factor can do much good!