MFS Soapbox Derby: aka what is your download speed?

[ercdvs]

Nope. USB1.1 is limited to 12mbps, and it isn't very efficient at packing data into that bandwidth, so in practice you are lucky if you can get 6mbps. 600-800KB/sec is typically the peak rate you'll see from a usb1.1 device. With the asix based devices like the FA-120, you should get at least 2MB/sec, and 3-3.5MB/sec if you are willing to do all the tweaks (custom kernel, etc). gige+jumbo can get you up to 5-6MB/sec These are extraction speeds. insertion is a little slower.

understood.. but, how many of the custom tweeks are still around that have not already been incorperated into the drivers ? Am i reading the threads linked on the first post & others wrong, or are these tweaks only effective on dtivos ?

[Jamie]

understood.. but, how many of the custom tweeks are still around that have not already been incorperated into the drivers ? Am i reading the threads linked on the first post & others wrong, or are these tweaks only effective on dtivos ?

See the Soapbox Derby thread. Things that help: use the current mfs-utils, use a custom kernel with CONFIG_NETFILTER turned off, tune to non-existent channels to stop the live buffering, etc.

The only thing that is old news because it is now the default in the drivers is the log2_irq_thresh=4 tweak.

Same tweaks apply to SA tivos and dtivos. I believe that standby mode turns off live buffering on the dtivos, but not on the SA, but other than that the issues and tweaks are all the same.

[blickdot]

The most I've been able to get out of my SA T800 is ~1.5MB/sec. Using all the above mentioned tweaks. That is download out of TiVo, going through my D-Link 100MB switch. See my sig. for Tivo Specs.

[Jamie]

The most I've been able to get out of my SA T800 is ~1.5MB/sec. Using all the above mentioned tweaks. That is download out of TiVo, going through my D-Link 100MB switch. See my sig. for Tivo Specs.

Yes. Series 2.5's (rather than Series 2's) have inferior usb implementations. Sounds like you are about par for the course.

[buechel]

Using MFS_FTP on my hacked S3 i get

transfer to PC - 3.88 MB/s on a 2.4 GB analog recording

PC -> Tivo - 2.20 MB/s on a 2.1 GB mpg

I did nothing to optimize the speed so dual tuner live buffering was running at this time.

This is using a 100 speed wired Ethernet cable with tivo software 8.0.1c

Of course everything is encrypted so besides measuring the speed, there isn't much point.

[rbreding]

Can someone comment as to the download speed that have had with the new drivers on 6.3c ?

Mine:

2.1M extraction
4.0M insertion

That is the best I have ever had for insertion, but my extraction speed has been better with 3.1.5f (3.5M).

Just curious what others have had...thanks

[michaeljc70]

I have a Dtivo with 6.2 on it and an Airlink 101. I am getting 1.2MB/sec with mfs_ftp which is double what I was getting before upgrading the drivers and USB adapter from 1.1. I am using:

* 2.4.20 backport drivers

* Unified MFS software

I am wondering if I can squeeze any more out of this. I noticed that after I installed the backport drivers and did a test I got 1.2MB/Sec. I then removed ax8817x.o, created the link from usbnet.o to ax8817x.o and rebooted and got the same speed. Is that normal?

[dburckh]

If you have the AGIGAUSB adapter, you can squeeze a little more out by enabling jumbo frames (MTU=9000 vs 1500) if you haven't done that already.

The only way beyond that is a kernel monte. It's a little hairy. Search for posts by "drez". He helped me out.

[michaeljc70]

If you have the AGIGAUSB adapter, you can squeeze a little more out by enabling jumbo frames (MTU=9000 vs 1500) if you haven't done that already.

The only way beyond that is a kernel monte. It's a little hairy. Search for posts by "drez". He helped me out.

No, I have the ASOHOUSB airlink. I take it that adjustment won't help me?

[opticalcarrier]

Is it yet possible to do USB 2.0 host-to-host from HR10-250 to WinXP? I saw some pretty old postings talking about it not yet being possible back then, so I'm suspecting that its still not possible, but I'm still holding out hope.

Isn't USB 2.0 host-to-host still faster than GIGE?

[Jamie]

Is it yet possible to do USB 2.0 host-to-host from HR10-250 to WinXP? I saw some pretty old postings talking about it not yet being possible back then, so I'm suspecting that its still not possible, but I'm still holding out hope.

Isn't USB 2.0 host-to-host still faster than GIGE?

As far as I know, nothing has changed. The windows drivers for the host-to-host cables use a different, Microsoft proprietary, framing protocol, that isn't supported in the backport driver.

[dburckh]

I found this out the hard way. Most of the gigabit switches with wireless (including my Linksys 350N) do NOT support jumbo frames. Further a number of wired only switches don't support jumbo frames. Gigabit does not automatically mean jumbo frame support.

I was wondering how people were getting these fantastic numbers with jumbo frames enabled and I wasn't with the same hardware. My cards were falling back to a standard frame due to my router not supporting jumbo frames.

[JustDan]

It is still a sad state of affairs with regards to JF support in the SoHo market.

Some of the newer Linksys and Netgear switches claim to have JF support,
but I am still quite pleased with my SMC 8508T units.

[jkozee]

Here are some raw numbers for the TiVoHD

TivoHD, 100mbps, stock kernel (except initrd), stock driver (built in NIC)

TiVoHD:/# netperf -H 192.168.168.4 -- -S 65536 -s 65536
TCP STREAM TEST to 192.168.168.4
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 65536 131070 131070    10.01      32.77

TivoHD, 1000mbps, custom kernel, agigausb

TiVoHD:/# netperf -H 192.168.168.4 -- -S 65536 -s 65536
TCP STREAM TEST to 192.168.168.4
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 65536 131070 131070    10.01      38.93

TivoHD, 100mbps, custom kernel, bcmemac driver (built in NIC)

TiVoHD:/var/hack# netperf -H 192.168.168.4 -- -S 65536 -s 65536
TCP STREAM TEST to 192.168.168.4
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 65536 131070 131070    10.01      44.08

TivoHD, 1000mbps, jumbo frames, custom kernel, agigausb

TiVoHD:/# netperf -H 192.168.168.4 -- -S 65536 -s 65536
TCP STREAM TEST to 192.168.168.4
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 65536 131070 131070    10.01      77.83

[bcc]

Those tivohd numbers make me wonder if the bcmenet driver could be hacked to support larger MTUs. I see it's hard coded at 0x600 max right now. If the broadcom chip is actually capable of a larger mtu, I'd think that'd provide the best performance.