Will 2 NICs solve this problem? - SUN

This is a discussion on Will 2 NICs solve this problem? - SUN ; I have an Ultra 10 running Solaris 8 controlling a mass spectrometer via an ethernet connection. We can also connect to the Ultra via ssh to download files and do whatever. Lately it seems, when people are downloading files the ...

+ Reply to Thread
Results 1 to 18 of 18

Thread: Will 2 NICs solve this problem?

  1. Will 2 NICs solve this problem?

    I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    via an ethernet connection.
    We can also connect to the Ultra via ssh to download files and do
    whatever.

    Lately it seems, when people are downloading files the connection to
    the mass spec gets stomped
    and the program that handles communication between the Ultra and the
    mass spec dies or stalls.
    I'm told it doesn't happen all the time, but fairly often.

    The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    (an Inmac Microhub 4
    4x10BaseT + AUI hub). Further, there's a connection between the hub
    and a Linksys home
    router which goes to the cloud.

    It's been suggested that throwing a second NIC on the Ultra 10 and
    having that as a dedicated
    connection to the mass spec might/will solve our problem. Would a
    second NIC have a reasonable
    chance of fixing my problem? Why or why not?

    My thought is to first, replace the existing hub with a little 5-port
    switch (I have a Netgear FS-105
    sitting around as a spare). If that doesn't work I'm thinking about
    futzing with the nice level of
    either sshd or the mass spec's server program to give the server
    program priority over ssh, but
    here I'm in deep water. Would either of these steps be as or more
    likely to fix my problem (I'd
    prefer not to install and configure a second NIC)?

    Also if anyone has any thoughts as to what more I could be doing to
    get a better understanding of
    this problem I'd appreciate hearing it. I don't much network
    troubleshooting and this problem seems
    to be really far under the hood for me.

    Thanks in advance,
    eric

  2. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    > via an ethernet connection.
    > We can also connect to the Ultra via ssh to download files and do
    > whatever.
    >
    > Lately it seems, when people are downloading files the connection to
    > the mass spec gets stomped
    > and the program that handles communication between the Ultra and the
    > mass spec dies or stalls.
    > I'm told it doesn't happen all the time, but fairly often.
    >
    > The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    > (an Inmac Microhub 4
    > 4x10BaseT + AUI hub). Further, there's a connection between the hub
    > and a Linksys home
    > router which goes to the cloud.
    >
    > It's been suggested that throwing a second NIC on the Ultra 10 and
    > having that as a dedicated
    > connection to the mass spec might/will solve our problem. Would a
    > second NIC have a reasonable
    > chance of fixing my problem? Why or why not?
    >
    > My thought is to first, replace the existing hub with a little 5-port
    > switch (I have a Netgear FS-105



    It would have been my first thought also. A hub just begs for
    collisions; the switch would be a cleaner way to go. If the mass spec
    isn't too bright, it might not be handling collisions properly. It has
    been a while since I've seen an unswitched network; my own hub has been
    gathering dust for several years now. If the people who designed the
    mass spec were thinking in terms of switched, the mass spec may not be
    able to handle a shared net.


  3. Re: Will 2 NICs solve this problem?

    Richard B. Gilbert wrote:
    > It would have been my first thought also. A hub just begs for
    > collisions;


    I don't think that collisions in and of themselves would be the issue
    as much as shared bandwidth. Unless... the downloads to the system
    controlling the mass spectrometer are triggering capture effect. But
    that would IIRC require that the link into the cloud from the Linksys
    had a bandwidth close to that of 10BT.

    > the switch would be a cleaner way to go. If the mass
    > spec isn't too bright, it might not be handling collisions properly.
    > It has been a while since I've seen an unswitched network; my own
    > hub has been gathering dust for several years now. If the people
    > who designed the mass spec were thinking in terms of switched, the
    > mass spec may not be able to handle a shared net.


    Is this a 100BT switch? If it is a "plain" 10BT switch, then since
    the initial problem statment suggested most of the downloads are to
    the system controlling the mass spectrometer it would need to be a
    100BT switch to make much of a difference. Just going to a 10BT
    switch would only help if there was a lot of traffic involving systems
    other than the one controlling the mass spectrometer.

    It would also be interesting to know a little bit more about the CPU
    utilization on the system controlling the mass spectrometer.
    Particularly while some of these downloads are taking place.

    rick jones
    from the peanut gallery
    --
    oxymoron n, commuter in a gas-guzzling luxury SUV with an American flag
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

  4. Re: Will 2 NICs solve this problem?

    Richard:

    The vendor supplied the hub with the mass spec so I'm fairly sure
    they weren't thinking switched network as a requirement. The
    computer on the mass spec itself is, I believe, a Motorola VME
    based single board computer. No idea what kind of network
    interface it has. The vendor refers to it as a "real time"
    controller.
    If they mean "real time" there may be some hairy timing issues
    that are only now, after seven years, manifesting themselves.
    I doubt that they tested the network under load.


    Rick:

    What's a "plain" 10BT switch? Everything I've seen over the last
    several years have been 10/100 switches. The switch I'm thinking
    of using is a 10/100BT switch with (they say) auto config of
    speed, xover, and duplex.

    Let me clarify, the downloading is going _from_ the Sun into the
    cloud, mostly.

    Typically, the Sun is talking to the mass spec, let's say, all the
    time.
    Sometimes to frequently someone(s) from a remote pc will connect
    to the Sun to run an interactive shell session or transfer files.
    That's
    a day in the life of the mass spec's network.

    CPU utilization and maybe a snoop/tcpdump study (gotta figure out
    how to run snoop/tcpdump first, though).


    regards,
    eric

  5. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    > via an ethernet connection.
    > We can also connect to the Ultra via ssh to download files and do
    > whatever.
    >
    > Lately it seems, when people are downloading files the connection to
    > the mass spec gets stomped
    > and the program that handles communication between the Ultra and the
    > mass spec dies or stalls.
    > I'm told it doesn't happen all the time, but fairly often.
    >
    > The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    > (an Inmac Microhub 4
    > 4x10BaseT + AUI hub). Further, there's a connection between the hub


    ha, wow. an inmac. How old is this thing?

    I'd suggest replacing hubs or checking cables before tossing in new nics,
    especially considering that 10Mb networks aren't going to tax an ultra10
    in the first place.



  6. Re: Will 2 NICs solve this problem?

    On Apr 2, 8:39 am, Cydrome Leader wrote:
    > Eric wrote:
    > > I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    > > via an ethernet connection.
    > > We can also connect to the Ultra via ssh to download files and do
    > > whatever.

    >
    > > Lately it seems, when people are downloading files the connection to
    > > the mass spec gets stomped
    > > and the program that handles communication between the Ultra and the
    > > mass spec dies or stalls.
    > > I'm told it doesn't happen all the time, but fairly often.

    >
    > > The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    > > (an Inmac Microhub 4
    > > 4x10BaseT + AUI hub). Further, there's a connection between the hub

    >
    > ha, wow. an inmac. How old is this thing?
    >
    > I'd suggest replacing hubs or checking cables before tossing in new nics,
    > especially considering that 10Mb networks aren't going to tax an ultra10
    > in the first place.


    We've had the mass spec for about seven years. It was also the mfg's
    proto
    so the instrument itself is probably around ten or more years old. If
    that hub
    was with the instrument the whole time I wouldn't be a bit surprised.

    And tell me, when was the last time you saw an AUI connector on
    anything?

    thanks,
    eric

  7. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > On Apr 2, 8:39 am, Cydrome Leader wrote:
    >
    >>Eric wrote:
    >>
    >>>I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    >>>via an ethernet connection.
    >>>We can also connect to the Ultra via ssh to download files and do
    >>>whatever.

    >>
    >>>Lately it seems, when people are downloading files the connection to
    >>>the mass spec gets stomped
    >>>and the program that handles communication between the Ultra and the
    >>>mass spec dies or stalls.
    >>>I'm told it doesn't happen all the time, but fairly often.

    >>
    >>>The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    >>>(an Inmac Microhub 4
    >>>4x10BaseT + AUI hub). Further, there's a connection between the hub

    >>
    >>ha, wow. an inmac. How old is this thing?
    >>
    >>I'd suggest replacing hubs or checking cables before tossing in new nics,
    >>especially considering that 10Mb networks aren't going to tax an ultra10
    >>in the first place.

    >
    >
    > We've had the mass spec for about seven years. It was also the mfg's
    > proto
    > so the instrument itself is probably around ten or more years old. If
    > that hub
    > was with the instrument the whole time I wouldn't be a bit surprised.
    >
    > And tell me, when was the last time you saw an AUI connector on
    > anything?
    >
    > thanks,
    > eric


    I still have a box with an AUI connector. I even have a 10BT
    transceiver to go with it. It's been years since it was powered on but
    it's still there. . . .



  8. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > On Apr 2, 8:39 am, Cydrome Leader wrote:
    >> Eric wrote:
    >> > I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    >> > via an ethernet connection.
    >> > We can also connect to the Ultra via ssh to download files and do
    >> > whatever.

    >>
    >> > Lately it seems, when people are downloading files the connection to
    >> > the mass spec gets stomped
    >> > and the program that handles communication between the Ultra and the
    >> > mass spec dies or stalls.
    >> > I'm told it doesn't happen all the time, but fairly often.

    >>
    >> > The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    >> > (an Inmac Microhub 4
    >> > 4x10BaseT + AUI hub). Further, there's a connection between the hub

    >>
    >> ha, wow. an inmac. How old is this thing?
    >>
    >> I'd suggest replacing hubs or checking cables before tossing in new nics,
    >> especially considering that 10Mb networks aren't going to tax an ultra10
    >> in the first place.

    >
    > We've had the mass spec for about seven years. It was also the mfg's
    > proto
    > so the instrument itself is probably around ten or more years old. If
    > that hub
    > was with the instrument the whole time I wouldn't be a bit surprised.
    >
    > And tell me, when was the last time you saw an AUI connector on
    > anything?


    With a few days actually, on an Ascent ISDN terminal adapter, It was still
    "new" enough to have a RJ45 jack as well.

    In my box of legacy nonsense, I have an IBM branded AUI to 10base-2
    transciever.

    It's almost as silly as those older NICs that actually had 10base2,
    twister pair AND an AUI port.

  9. Re: Will 2 NICs solve this problem?

    Cydrome Leader wrote:
    > Eric wrote:
    >
    >>On Apr 2, 8:39 am, Cydrome Leader wrote:
    >>
    >>>Eric wrote:
    >>>
    >>>>I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    >>>>via an ethernet connection.
    >>>>We can also connect to the Ultra via ssh to download files and do
    >>>>whatever.
    >>>
    >>>>Lately it seems, when people are downloading files the connection to
    >>>>the mass spec gets stomped
    >>>>and the program that handles communication between the Ultra and the
    >>>>mass spec dies or stalls.
    >>>>I'm told it doesn't happen all the time, but fairly often.
    >>>
    >>>>The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    >>>>(an Inmac Microhub 4
    >>>>4x10BaseT + AUI hub). Further, there's a connection between the hub
    >>>
    >>>ha, wow. an inmac. How old is this thing?
    >>>
    >>>I'd suggest replacing hubs or checking cables before tossing in new nics,
    >>>especially considering that 10Mb networks aren't going to tax an ultra10
    >>>in the first place.

    >>
    >>We've had the mass spec for about seven years. It was also the mfg's
    >>proto
    >>so the instrument itself is probably around ten or more years old. If
    >>that hub
    >>was with the instrument the whole time I wouldn't be a bit surprised.
    >>
    >>And tell me, when was the last time you saw an AUI connector on
    >>anything?

    >
    >
    > With a few days actually, on an Ascent ISDN terminal adapter, It was still
    > "new" enough to have a RJ45 jack as well.
    >
    > In my box of legacy nonsense, I have an IBM branded AUI to 10base-2
    > transciever.
    >
    > It's almost as silly as those older NICs that actually had 10base2,
    > twister pair AND an AUI port.


    I think I might have one those too. DE400-???


  10. Re: Will 2 NICs solve this problem?

    On Apr 3, 9:48 am, "Richard B. Gilbert"
    wrote:
    > Cydrome Leader wrote:
    > > Eric wrote:

    >
    > >>On Apr 2, 8:39 am, Cydrome Leader wrote:

    >
    > >>>Eric wrote:

    >
    > >>>>I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    > >>>>via an ethernet connection.
    > >>>>We can also connect to the Ultra via ssh to download files and do
    > >>>>whatever.

    >
    > >>>>Lately it seems, when people are downloading files the connection to
    > >>>>the mass spec gets stomped
    > >>>>and the program that handles communication between the Ultra and the
    > >>>>mass spec dies or stalls.
    > >>>>I'm told it doesn't happen all the time, but fairly often.

    >
    > >>>>The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    > >>>>(an Inmac Microhub 4
    > >>>>4x10BaseT + AUI hub). Further, there's a connection between the hub

    >
    > >>>ha, wow. an inmac. How old is this thing?

    >
    > >>>I'd suggest replacing hubs or checking cables before tossing in new nics,
    > >>>especially considering that 10Mb networks aren't going to tax an ultra10
    > >>>in the first place.

    >
    > >>We've had the mass spec for about seven years. It was also the mfg's
    > >>proto
    > >>so the instrument itself is probably around ten or more years old. If
    > >>that hub
    > >>was with the instrument the whole time I wouldn't be a bit surprised.

    >
    > >>And tell me, when was the last time you saw an AUI connector on
    > >>anything?

    >
    > > With a few days actually, on an Ascent ISDN terminal adapter, It was still
    > > "new" enough to have a RJ45 jack as well.

    >
    > > In my box of legacy nonsense, I have an IBM branded AUI to 10base-2
    > > transciever.

    >
    > > It's almost as silly as those older NICs that actually had 10base2,
    > > twister pair AND an AUI port.

    >
    > I think I might have one those too. DE400-???


    You know those things really are all over the place if you stop to
    look.
    I have '90s vintage Personal Iris (in another lab, still working) w/
    an
    AUI (it goes through an AUI to UTP transceiver). Also that single
    board computer in the mass spec I come to find out uses an AUI
    (it too gets turned into UTP). Until a few years ago we had a DEC
    DELNI that was gathering dust.

  11. Re: Will 2 NICs solve this problem?

    Cydrome Leader writes:
    >It's almost as silly as those older NICs that actually had 10base2,
    >twister pair AND an AUI port.


    Why is that silly?

    At the time they were made, AUI drops off the 10-base-5 were a valid
    form of ethernet still in use in many enterprise shops?

    10-base-5 was a touch more stable than many 10-base-2 installations.

    Granted, once 1X came o , most places couldn't rip it out
    quick enough, but there were still some holdouts for quite some time..



  12. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > Rick:


    > What's a "plain" 10BT switch?


    My poor wording for a switch that only speaks 10BT.

    > Everything I've seen over the last several years have been 10/100
    > switches. The switch I'm thinking of using is a 10/100BT switch
    > with (they say) auto config of speed, xover, and duplex.


    OK, so you could indeed see a 10X increase in potential bandwidth out
    of a single system.

    > Let me clarify, the downloading is going _from_ the Sun into the
    > cloud, mostly.


    OK. I guess there could be short bursts then of "capture effect" with
    the Sun grabbing the ether and not letting go.

    You mentioned ssh - are the downloads tunneling through that or
    otherwise encrypted?

    > Typically, the Sun is talking to the mass spec, let's say, all the
    > time. Sometimes to frequently someone(s) from a remote pc will
    > connect to the Sun to run an interactive shell session or transfer
    > files. That's a day in the life of the mass spec's network.


    > CPU utilization and maybe a snoop/tcpdump study (gotta figure out
    > how to run snoop/tcpdump first, though).


    tcpdump -i -w

    will run tcpdump over the specified interface, storing the packets it
    captures into for later analysis using tcpdump -r or some
    other utility able to read tcpdumps dumpfiles.

    rick jones
    --
    denial, anger, bargaining, depression, acceptance, rebirth...
    where do you want to be today?
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

  13. Re: Will 2 NICs solve this problem?

    On Apr 3, 3:01 pm, Rick Jones wrote:

    >
    > You mentioned ssh - are the downloads tunneling through that or
    > otherwise encrypted?
    >


    Well, all our remote access to the Sun is via either sftp or the
    ssh client. So the downloads are encrypted but only to the
    extent that sftp is encrypted. And we're not doing stuff like
    running ftp or X sessions through an ssh tunnel if that's what
    you mean by tunneling.

    Seeing as how the mass spec's computer is only running 10Mb/s
    (I think), I've got a pretty good feeling about this switch business.

    thanks,
    eric

  14. Re: Will 2 NICs solve this problem?

    Richard B. Gilbert wrote:
    > Cydrome Leader wrote:
    >> Eric wrote:
    >>
    >>>On Apr 2, 8:39 am, Cydrome Leader wrote:
    >>>
    >>>>Eric wrote:
    >>>>
    >>>>>I have an Ultra 10 running Solaris 8 controlling a mass spectrometer
    >>>>>via an ethernet connection.
    >>>>>We can also connect to the Ultra via ssh to download files and do
    >>>>>whatever.
    >>>>
    >>>>>Lately it seems, when people are downloading files the connection to
    >>>>>the mass spec gets stomped
    >>>>>and the program that handles communication between the Ultra and the
    >>>>>mass spec dies or stalls.
    >>>>>I'm told it doesn't happen all the time, but fairly often.
    >>>>
    >>>>>The Ultra 10, the mass spec, a printer and a PC all connect to a hub
    >>>>>(an Inmac Microhub 4
    >>>>>4x10BaseT + AUI hub). Further, there's a connection between the hub
    >>>>
    >>>>ha, wow. an inmac. How old is this thing?
    >>>>
    >>>>I'd suggest replacing hubs or checking cables before tossing in new nics,
    >>>>especially considering that 10Mb networks aren't going to tax an ultra10
    >>>>in the first place.
    >>>
    >>>We've had the mass spec for about seven years. It was also the mfg's
    >>>proto
    >>>so the instrument itself is probably around ten or more years old. If
    >>>that hub
    >>>was with the instrument the whole time I wouldn't be a bit surprised.
    >>>
    >>>And tell me, when was the last time you saw an AUI connector on
    >>>anything?

    >>
    >>
    >> With a few days actually, on an Ascent ISDN terminal adapter, It was still
    >> "new" enough to have a RJ45 jack as well.
    >>
    >> In my box of legacy nonsense, I have an IBM branded AUI to 10base-2
    >> transciever.
    >>
    >> It's almost as silly as those older NICs that actually had 10base2,
    >> twister pair AND an AUI port.

    >
    > I think I might have one those too. DE400-???


    not sure of the model, but it had amber lights and is just big enough that
    it needs to attach to the AUI port with some 1980s IBM-ish overly thick
    cable. It's not nearly as large as the Synoptics ash tray sized
    transcievers though.

  15. Re: Will 2 NICs solve this problem?

    Doug McIntyre wrote:
    > Cydrome Leader writes:
    >>It's almost as silly as those older NICs that actually had 10base2,
    >>twister pair AND an AUI port.

    >
    > Why is that silly?


    They were such a hassle to use. You had to keep a stack of softset config
    diskettes to see how each card was configured.

    As I mentioned with SCSI cards somewhere else, there were many more
    choices of cards years ago than these days.

    The first 10/100 cards were retarded like that as well. I recall setting
    speed on some 3com cards with a boot floppy because they didn't have
    autonegotiation yet.

    Dip switches on the rear of the card would have been most convenient for
    stuff like that.

    media converter always seem to have dip switches for the SQE stuff.


  16. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > Well, all our remote access to the Sun is via either sftp or the ssh
    > client. So the downloads are encrypted but only to the extent that
    > sftp is encrypted.


    That thought was behind my asking about CPU utilization during these
    downloads

    I suppose the mass spectrometer isn't kind enough to have a way to
    give you its link-level statistics If it were, to check on the
    capture effect question, you would check them for "excessive retries"
    (or something like that) which would be sufficient but not necessary
    to show capture effect. Capture effect is when one host on the
    CSMA/CD network is able to transmit fast enough to get back-to-back
    frames on the wire, and cause other stations to have their backoff
    timers keep growing. That could put some serious delays into the
    communications. When one is able to run full-duplex (ie with a switch
    rather than a hub) then there is no CSMA/CD any longer, so no capture
    effect.

    rick jones
    --
    denial, anger, bargaining, depression, acceptance, rebirth...
    where do you want to be today?
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

  17. Re: Will 2 NICs solve this problem?

    On Apr 3, 7:22 pm, Rick Jones wrote:
    > Eric wrote:
    > > Well, all our remote access to the Sun is via either sftp or the ssh
    > > client. So the downloads are encrypted but only to the extent that
    > > sftp is encrypted.

    >
    > That thought was behind my asking about CPU utilization during these
    > downloads
    >
    > I suppose the mass spectrometer isn't kind enough to have a way to
    > give you its link-level statistics If it were, to check on the
    > capture effect question, you would check them for "excessive retries"
    > (or something like that) which would be sufficient but not necessary
    > to show capture effect. Capture effect is when one host on the
    > CSMA/CD network is able to transmit fast enough to get back-to-back
    > frames on the wire, and cause other stations to have their backoff
    > timers keep growing. That could put some serious delays into the
    > communications. When one is able to run full-duplex (ie with a switch
    > rather than a hub) then there is no CSMA/CD any longer, so no capture
    > effect.
    >
    > rick jones
    > --
    > denial, anger, bargaining, depression, acceptance, rebirth...
    > where do you want to be today?
    > these opinions are mine, all mine; HP might not want them anyway...
    > feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...


    You're thinking that ssh could be saturating the cpu and having a
    trickle
    down effect on the network interface?

    As far as I know, this is a recent problem. The network
    infrastructure
    hasn't changed in some time (it's just as slow as it's ever been). I
    don't see anything that would account for increased traffic, not
    greatly increased, anyway. It may be that we got some new software
    that's chewing up more cpu time.

    I'm not expecting any help from the mass spec's SBC.

    So if network capture is the problem, I wouldn't necessarily see it
    from the Sun, the machine doing the capturing?

    I'll probably find out sooner or later but, does snoop (on Solaris 8)
    have a large cpu footprint? If it's set up to not do hostname lookup
    does snoop introduce any network latency?

    thanks,
    eric

  18. Re: Will 2 NICs solve this problem?

    Eric wrote:
    > You're thinking that ssh could be saturating the cpu and having a
    > trickle down effect on the network interface?


    Yep. Probability might be small.

    > So if network capture is the problem, I wouldn't necessarily see it
    > from the Sun, the machine doing the capturing?


    Nope. That is the "fast" system able to grab and hold the ether while
    the other stations' backoff timers grow. Actually I suppose it could
    appear - you could see larger than normal gaps between requests sent
    to the mass spectrometer and the replies back.

    rick jones
    --
    web2.0 n, the dot.com reunion tour...
    these opinions are mine, all mine; HP might not want them anyway...
    feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

+ Reply to Thread