locked
Slow Gigabit network RRS feed

  • Question

  • I'm not sure if this is related to WHS or not, but I'm seeing slow speed(for a 1000 GB network).

    My 2 machines are:
    P4 HT 1.8 / 1Gig RAM / Intel Server 1000 Gig card  / XP
    P4 3.0 / 1 Gig RAM / Intel Server 1000 Gig card / WHS

    I have them connected to a Gigabit switch

    When transferring files, Task Manager only shows the Network Utilization at around 12%.
    The speed is about 130 Mb.
    The CPU isn't pegged at 100% so nothing seems to be hogging resources.

    I'm not using Jumbo Frames.. (I tried it, and it didn't make a difference)

    Are there any other test I can run?

    thanks,
    tom


    Wednesday, May 2, 2007 2:39 PM

All replies

  • The file transfer speed will be as fast as the slowest of the following things, the throughput of the source system, the throughput of the network and the throughput of the destination system.

     

    The throughput of the source and destination systems will depend on how fast the disk subsystems are, not just the interface speed but also the rotational speed of the disks (rpm), how fragmented the disks are, the type and speed of the busses that the disk interface and network interfaces are connected to, whether or not, DMA is being used and what else the system is doing at the time.

     

    In short file transfer speed is not a very good way to measure network speed (by the way - 130Mb is a file size not a speed) - for what it's worth I get about 160~170 Mb/s (that's megabits/second)

     

    System #1 - 866 MHz P3, 512 MB RAM, Realtek PCI GbE, Silicon Image SI3112 PCI SATA , Maxtor DiamondMax 9, 250 GB SATA drive - WHS

    System #2 - 2.4 GHz Celeron, 1GB RAM, integrated Intel GbE, integrated Intel SATA, mirrored Maxtor DiamondMax 8, 80 GB SATA drive - Server 2K3

    Network switch is a Netgear FS728TS with the systems connected to two of the four GbE ports - the switch does NOT support jumbo frame

     

    The limit is the WHS system - I know this because I can hit just under 300 Mb/s transferring from System #2 (an IBM xSeries 206 server) to a Dell Inspiron laptop.

     

     

     

    Wednesday, May 2, 2007 8:21 PM
  • Sorry, I meant Mb/s I'm using NetStat Live by AnalogX.. it's an old program but I've used it for years.
    http://www.analogx.com/contents/download/network/nsl.htm

    If file transfer speed is not a good way to measure network speed, then what should I look at?

    Also, what program can I use to see where the bottleneck is?
    I also use Process Explorer and nothing looks pegged so I'm not sure what variable I should be looking at.

    thanks,
    tom




    Wednesday, May 2, 2007 9:06 PM
  • Thats not actually too bad for 2 workstations.  Does it stream at that rate smoothly, or does it start and stop, and restart?

     

    You mentioned server NIC, does this mean that they are on something better than a desktop motherboards single shared 32bit 33Mhz PCI bus?  Is your NIC on a dedicated IRQ, are your HDD controllers?

     

    If you have them on a PCIE or PCI-X or even a dedicated PCI bus 64 bit, 66Mthz

    Try moving a multi gb fine back and forth, watching your network monitor.

    If you are trying to measure just network performance you can use something like ixchariot or iperf.

     

    If you want to do some comparisions, post details on your testing proceedure, and we can try to duplicate it.

     

     

     

    Saturday, May 5, 2007 4:24 AM
  • It streams smoothly but that's the rate of a 100 Mbps NIC and it worked fine when I had the 100 Meg NIC as well.

    It's a server NIC but when it's plugged into a regular slot, it works in 32-bit mode.  I've also performed this same exact test in an actual server that the card fully fits in, and the results are the same.

    I followed up another thread with my specs since the thread seemed to be of similar issues.

    The thread is here

    My post is as follows:

    -----

    I have 2 machines with gig cards connected to a gig switch.
    using iperf with the following parameters:

    iperf -s -u -w2m -i1   (on the server)
    iperf -c <above server ip>  -u -b400M -w2m -i1 -t60    (on the client)

    I get the following results with task manager showing that Network Utilization is only 12%

    [ ID] Interval       Transfer     Bandwidth
    [1880]  0.0- 1.0 sec  18.6 MBytes   156 Mbits/sec
    [1880]  1.0- 2.0 sec  19.2 MBytes   161 Mbits/sec
    ...

    If I swap the cards with 10/100 card and force 100 full on the switch, I get the following results with task manager showing that Network Utilization is 98%

    [ ID] Interval       Transfer     Bandwidth
    [1880]  0.0- 1.0 sec  11.4 MBytes  95.6 Mbits/sec
    [1880]  1.0- 2.0 sec  11.4 MBytes  95.6 Mbits/sec
    ...

    Tuesday, May 8, 2007 6:56 PM
  • Try running perfmon on the destination looking at Physical Disk.% Write Time. If it is at 100+%, that is your bottleneck. Writing 13Mbytes\sec isn't all that bad performance for a lot of older technology single drives.
    Tuesday, May 8, 2007 8:23 PM
  • yeah, running perfmon is probably your best bet here.  A common misconception is that hard drives are faster than the network... they aren't.  The interfaces may look as though they are faster - SATA 3 being 3 gigabits speed, but no task can really be transferred that fast, except possibly to fill the cache on the drive, which would be the first 16 megabytes.  Mechanically, most drives are only able to perform at about 200-400 megabits per second in reading and 100-300 megabits per sec when writing.

    I too was disappointed when i got my gigabit network and didn't see a dramatic speed increase.
    Wednesday, May 9, 2007 8:06 PM
  • Having only two machines is not enough for determining a network throughput.  I have five clients on my WHS and I got roughly the same transfer rate as you did.  I believe you are seeing the result of a hardware limitation, not WHS.

     

    Here's my expeiriment after reading your thread:

     

    I tried have two different sets of PC's, transferring the same size file (197 meg) back and forth between them.  I watched  the Network Utilization as well as timed the transfers. 

     

    Between PC's (WHS not invloved), I could do simultaneous transfers without any difference in time that it took to do a single transfer.  The network utilization value does rise accordingly.

     

    When I transferred between a client and WHS and between two other PC's, the WHS transfer takes slightly longer, but the two other clients were just as fast.

     

    When I do simultaenous transfers from two clients to WHS,  I do see a very noticable drop in trabsfer rate.  Also the netwrok utilization goes up dramatically.  Three, was really pushing it.

     

    Monday, May 14, 2007 4:26 PM
  • I think part of it though is how fast it is relative to other products.  I did a test of this on a machine that I could image to try a few different things.  A retired Compaq Xeon server.. when I'm running NASLite 2.04, I get data transfers over gigabit almost twice as fast as the speeds I'm getting under WHS, especially when doing writes from a connected Windows XP Client.

    I'm still playing around, though, so this may be some configuration issues.        
    Thursday, May 17, 2007 2:17 AM
  • What brand NIC are you using? Some NICs advertise gigabit but can't ever get up to that speed. I've found this to be the case with newer Netgear NICs that use Realtek controllers.
    Thursday, May 24, 2007 10:01 PM
  • What brand NIC is it? Some Netgear NICs use Realtek chipsets that get nowhere near decent gigabit speeds.
    Thursday, May 24, 2007 10:02 PM
  • Quote from  tv on 05-02-2007 9:06 PM UTC

     

    If file transfer speed is not a good way to measure network speed, then what should I look at?

    A good bench mark program to do raw benchmarking is wstcp. Its command line driven and once you wrap your mind around bits to bytes, it will tell you if something is wrong or not.

     

    I have found that workstation os's have a limit of 20-30% of total connection bandwith is consumed per File transfer, so opening 3-5 connections (one per folder) I can use all availible bandwith of the nic. (annoying really)

    Drop a mail to miellsworth gmail com if there is any help with wstcp, I have been using it for the last few years troubleshooting the network at work and have a little bit of the feel for what is a software limit or a hardware limit.

    Basicly 2000/XP SP1 gets 80-99% of 100MBs link depending on nic

    XP SP2 gets 90-95% of 100MBs and 60-95% of 1000MBs

    Server 2000/2003 gets 80-103% (?) of 100MBs and 88-99% of 1000MBs and the one time I was able to test 91.455% of 10Gbs (coolness)

    Good luck!!!

    Friday, May 25, 2007 11:43 PM