speeding up LAVC with switch instead of hub? - DEC

This is a discussion on speeding up LAVC with switch instead of hub? - DEC ; I'm looking to speed up communication in my LAN. I have a cluster with all communication on the LAN (SCS, TCPIP, LAT etc; shadow copies/merges go across the network as well). Now, everything is connected via a hub. If I ...

+ Reply to Thread
Page 1 of 2 1 2 LastLast
Results 1 to 20 of 39

Thread: speeding up LAVC with switch instead of hub?

  1. speeding up LAVC with switch instead of hub?

    I'm looking to speed up communication in my LAN. I have a cluster with
    all communication on the LAN (SCS, TCPIP, LAT etc; shadow copies/merges
    go across the network as well). Now, everything is connected via a hub.

    If I understand things correctly, a switch gives me two advantages with
    regard to speed: ethernet controllers can operate in full-duplex mode,
    and collisions are eliminated.

    Full duplex would double the bandwidth, whereas avoiding collisions
    would only increase bandwidth if collisions are reducing it, right?

    Normally, collisions aren't a problem and there are many of them only
    during shadow copies and merges.

    To benefit from full duplex, the ethernet controllers would have to be
    capable of full duplex. Would that be the case for any of the following
    machines (which have the "stock" ethernet cards)?

    o VAXstation 4000/90

    o VAX 4000/105A

    o DEC 3000/600

    o DEC 3000/300LX

    o ALPHAserver 1200 (aka Digital Server 5305)


  2. Re: speeding up LAVC with switch instead of hub?

    In article , helbig@astro.multiCLOTHESvax.de (Phillip Helbig---remove CLOTHES to reply) writes:
    >To benefit from full duplex, the ethernet controllers would have to be
    >capable of full duplex. Would that be the case for any of the following
    >machines (which have the "stock" ethernet cards)?
    >
    > o VAXstation 4000/90
    >
    > o VAX 4000/105A
    >
    > o DEC 3000/600
    >
    > o DEC 3000/300LX


    No. 10MBit/Half Duplex only

    > o ALPHAserver 1200 (aka Digital Server 5305)


    Probably, Yes.
    If it is 100MBit, then FD is likely, if 10MB (which I doubt) then unlikely.

    --
    Peter "EPLAN" LANGSTOEGER
    Network and OpenVMS system specialist
    E-mail peter@langstoeger.at
    A-1030 VIENNA AUSTRIA I'm not a pessimist, I'm a realist

  3. Re: speeding up LAVC with switch instead of hub?


    "Phillip Helbig---remove CLOTHES to reply" wrote in
    message news:e4f2js$m11$1@online.de...

    > If I understand things correctly, a switch gives me two advantages with
    > regard to speed: ethernet controllers can operate in full-duplex mode,
    > and collisions are eliminated.


    OK, so far - although they are pretty much the same thing. Two big
    things you omitted: the backplane of a switch will usually have enough
    capacity to drive all ports simultaneously at full speed (assuming you
    can match data sources and sinks). Secondly, you can run systems at
    different speeds.

    > Full duplex would double the bandwidth, whereas avoiding collisions
    > would only increase bandwidth if collisions are reducing it, right?


    You have to be slightly careful about 'double'. With a small number of
    systems (unlike a big network trunk), you are likely to find traffic
    predominately in one direction at any given moment. You win a bit
    in that your bulk data and acknowledgements don't contend for the
    network.

    > Normally, collisions aren't a problem and there are many of them only
    > during shadow copies and merges.


    Best just to forget about them, their overhead in a five node network is
    going to be minute.

    > To benefit from full duplex, the ethernet controllers would have to be
    > capable of full duplex. Would that be the case for any of the following
    > machines (which have the "stock" ethernet cards)?


    Yes. An Alphaserver 1200 would likely have a 100Mb/s full duplex card.
    No to the VAXes.
    Maybe - Turbochannel Alphas. I can't recall.

    You don't give a lot of information about your disk layout or how
    your load is balanced:

    If most of your traffic were to/from the 1200 then letting it rip at
    100Mb/s looks like a big win. Although having two faster servers
    to run the shadow sets off would be better.




  4. Re: speeding up LAVC with switch instead of hub?

    In article <446b359f@news.langstoeger.at>, peter@langstoeger.at (Peter
    'EPLAN' LANGSTOEGER) writes:

    > > o VAXstation 4000/90
    > >
    > > o VAX 4000/105A
    > >
    > > o DEC 3000/600
    > >
    > > o DEC 3000/300LX

    >
    > No. 10MBit/Half Duplex only
    >
    > > o ALPHAserver 1200 (aka Digital Server 5305)

    >
    > Probably, Yes.
    > If it is 100MBit, then FD is likely, if 10MB (which I doubt) then unlikely.


    Since collisions are usually not a problem (and when they are, such as
    when a shadow copy is taking place, "normal use" is not recommended
    anyway), then without full-duplex capability, I suppose a switch isn't
    worth the investment. (I only use the 5305 when I need to use Mozilla,
    and that tends to generate a lot of traffic. Since I have a spare port
    in the switch in the DSL router (to which the cluster hub is connected
    by an uplink), it would make more sense to plug the 5305 directly into
    a switch port in the DSL router.)


  5. Re: speeding up LAVC with switch instead of hub?

    In article , "Richard Brodie"
    writes:

    > OK, so far - although they are pretty much the same thing. Two big
    > things you omitted: the backplane of a switch will usually have enough
    > capacity to drive all ports simultaneously at full speed (assuming you
    > can match data sources and sinks). Secondly, you can run systems at
    > different speeds.


    The first isn't an issue; the second could be important (see below).

    > > Normally, collisions aren't a problem and there are many of them only
    > > during shadow copies and merges.

    >
    > Best just to forget about them, their overhead in a five node network is
    > going to be minute.


    I do see the collision LED light up during a shadow copy.

    > Yes. An Alphaserver 1200 would likely have a 100Mb/s full duplex card.
    >
    > You don't give a lot of information about your disk layout or how
    > your load is balanced:
    >
    > If most of your traffic were to/from the 1200 then letting it rip at
    > 100Mb/s looks like a big win. Although having two faster servers
    > to run the shadow sets off would be better.


    As I mentioned in another post, the 5305 is a satellite which I only
    switch on when I need Mozilla/CSWB (no other system is powerful enough
    to run it, but the 5305 uses too much power to keep it on all the time).
    In some cases, most of the bandwidth on the LAN would be taken up by the
    5305, so plugging that into the full-duplex--capable 100 Mb/s
    ethernet-switch port on the DSL router would be the thing to do.

    Except for system disks, each shadow set has members on two different
    nodes. The biggest (9 GB) is on the 2 VAXes. Originally, this was to
    avoid shadow copies when rebooting for patches (which happens a lot more
    on ALPHA). However, with MINICOPY I can even reboot a VAX (as long as
    an ALPHA stays in the cluster). Were a member on the one permanent
    ALPHA in the cluster, then taking it down for maintenance would always
    mean a full copy. As long as no copies or merges are going on, the two
    VAXes are fast enough for the big shadow set. (When I am doing I/O
    intensive work on the 5305 and on the big shadow set, I notice it is a
    lot faster than from one of the VAXes, so the bottleneck (when using,
    say, SEARCH in MAIL) is probably the CPU speed of the VAXes, not the
    access to the disks.

    I would like to use some disks on the 5305, for example big disks (9 or
    36 GB) for disk-to-disk backup. (No disadvantage if the 5305 is
    switched off most of the time.) However, I need a SCSI controller which
    will work with VMS. (As it stands, only one of the SCSI controllers
    works with VMS; the CD-ROM drive and an internal page/swap disk (it is a
    satellite, so boots off another ALPHA) are on it. There are a lot of
    internal SBB slots. I think it is split-bus, so for one controller I
    could use some of those and additionally put something on externally.)
    Anyone have one they don't need?


  6. Re: speeding up LAVC with switch instead of hub?

    Phillip Helbig---remove CLOTHES to reply wrote:
    > I'm looking to speed up communication in my LAN. I have a cluster with
    > all communication on the LAN (SCS, TCPIP, LAT etc; shadow copies/merges
    > go across the network as well). Now, everything is connected via a hub.


    I would not expect the LAN infrastructure to be the central limit
    here. A switch will certainly help isolate the traffic, but it won't
    help make a 10Mb VAX NIC any faster.

    I'd look to see if it is the VAX boxes, or the 10Mb connections, or
    both. With a shadow copy, you're jamming a whole pile of data over a
    comparatively small pipe, so you will see collisions and back-offs.

    (Run some math, and see what the theoretical best case for a full
    copy is, given your disk sizes and your NIC speeds. If that calculation
    is not hugely different from what you are experiencing, then the limit
    likely isn't the LAN infrastructure.)

    With current releases of OpenVMS, you can use some of the throttling
    capabilities within shadowing (see the documentation for the shadowing
    ECO kits, or the OpenVMS documentation for shadowing on the current
    releases) and dial back the network usage (trading off network bandwidth
    for time), but I'd expect the core configuration limit is comparatively
    big data and comparatively small pipes.

  7. Re: speeding up LAVC with switch instead of hub?


    "Phillip Helbig---remove CLOTHES to reply" wrote in
    message news:e4f8ub$gvk$1@online.de...

    >> Best just to forget about them, their overhead in a five node network is
    >> going to be minute.

    >
    > I do see the collision LED light up during a shadow copy.


    Yes, I expect you would. The LED indicators are invariably hugely
    pulse stretched, though.



  8. Re: speeding up LAVC with switch instead of hub?

    I've got my MACs and DSL stuff on a Switch. The cluster is on a hub with
    one port going to the switch. This way, the cluster doesn't see traffic
    between the macs and the internet, or appletalk traffic etc.

    Within the cluster, I can still use ethermon to look at all the traffic
    between nodes. Unless you get a managed switch where you can set a port
    onto promiscuous mode, one node can no longer look at all the traffic
    since the switch filters any traffic that node isn't supposed to see.

  9. Re: speeding up LAVC with switch instead of hub?

    In article , Hoff Hoffman
    writes:

    > Phillip Helbig---remove CLOTHES to reply wrote:
    > > I'm looking to speed up communication in my LAN. I have a cluster with
    > > all communication on the LAN (SCS, TCPIP, LAT etc; shadow copies/merges
    > > go across the network as well). Now, everything is connected via a hub.

    >
    > I would not expect the LAN infrastructure to be the central limit
    > here. A switch will certainly help isolate the traffic, but it won't
    > help make a 10Mb VAX NIC any faster.
    >
    > I'd look to see if it is the VAX boxes, or the 10Mb connections, or
    > both. With a shadow copy, you're jamming a whole pile of data over a
    > comparatively small pipe, so you will see collisions and back-offs.


    Right. That's normally the only time the collision LED on the hub
    lights uup a lot.

    > (Run some math, and see what the theoretical best case for a full
    > copy is, given your disk sizes and your NIC speeds. If that calculation
    > is not hugely different from what you are experiencing, then the limit
    > likely isn't the LAN infrastructure.)


    I get 2 hours for a full copy; it takes about 6 or 7.

    > With current releases of OpenVMS, you can use some of the throttling
    > capabilities within shadowing (see the documentation for the shadowing
    > ECO kits, or the OpenVMS documentation for shadowing on the current
    > releases) and dial back the network usage (trading off network bandwidth
    > for time), but I'd expect the core configuration limit is comparatively
    > big data and comparatively small pipes.


    I do have SHADOW_MAX_COPY set to 1 for that reason.


  10. Re: speeding up LAVC with switch instead of hub?

    In article <446B8AFB.98487D3C@teksavvy.com>, JF Mezei
    writes:

    > I've got my MACs and DSL stuff on a Switch. The cluster is on a hub with
    > one port going to the switch. This way, the cluster doesn't see traffic
    > between the macs and the internet, or appletalk traffic etc.


    Yes, I have a similar setup. I don't have a MAC, but have a VOIP
    thingy.


  11. Re: speeding up LAVC with switch instead of hub?

    In article <446B8AFB.98487D3C@teksavvy.com>, JF Mezei writes:
    > I've got my MACs and DSL stuff on a Switch. The cluster is on a hub with
    > one port going to the switch. This way, the cluster doesn't see traffic
    > between the macs and the internet, or appletalk traffic etc.


    I've got all my home stuff on a switch. You can tell what's what
    because the old DEC stuff is doing 10MB which makes the green lights
    come on and the rest are newer boxes doing 100MB which makes the yellow
    lights come on. You can tell when there's a lot of SCS traffic, it
    makes the green lights flash a lot.


  12. Re: speeding up LAVC with switch instead of hub?

    On Thu, 18 May 2006 05:56:11 -0700, Bob Koehler
    wrote:

    > In article <446B8AFB.98487D3C@teksavvy.com>, JF Mezei
    > writes:
    >> I've got my MACs and DSL stuff on a Switch. The cluster is on a hub with
    >> one port going to the switch. This way, the cluster doesn't see traffic
    >> between the macs and the internet, or appletalk traffic etc.

    >
    > I've got all my home stuff on a switch. You can tell what's what
    > because the old DEC stuff is doing 10MB which makes the green lights
    > come on and the rest are newer boxes doing 100MB which makes the
    > yellow
    > lights come on. You can tell when there's a lot of SCS traffic, it
    > makes the green lights flash a lot.
    >

    I don't think yellow is good, at least not on a Cisco switch

  13. Re: speeding up LAVC with switch instead of hub?

    Tom Linden wrote:

    > I don't think yellow is good, at least not on a Cisco switch


    Cisco apparently understands the basic concept of traffic light color
    coding, then.

    On desktop switches, tri-state LEDs tend to show 10 Mb as green, 100
    Mb as yellow, and faults or collisions as red. I've certainly worked
    with several of these switches.

    --

    A theoretical two hour transfer of a full volume copy over a 10 Mb
    network implies circa seven GB disks, give or take, and particularly
    assuming I've done the math correctly. Given the time to read the data
    off and the time to write the data out and the glacial speeds and feeds
    of most VAX systems and add a dollop of contention and given you that
    won't ever hit the theoretical speed of an IEEE 802.3/Ethernet network,
    and your six-to-seven hour environment doesn't look all that far out of
    line.

    Old hardware -- old VAX systems, old I/O, old NICs -- is slow.

  14. Re: speeding up LAVC with switch instead of hub?

    In article , "Tom Linden" writes:
    > On Thu, 18 May 2006 05:56:11 -0700, Bob Koehler
    > wrote:
    >
    >> In article <446B8AFB.98487D3C@teksavvy.com>, JF Mezei
    >> writes:
    >>> I've got my MACs and DSL stuff on a Switch. The cluster is on a hub with
    >>> one port going to the switch. This way, the cluster doesn't see traffic
    >>> between the macs and the internet, or appletalk traffic etc.

    >>
    >> I've got all my home stuff on a switch. You can tell what's what
    >> because the old DEC stuff is doing 10MB which makes the green lights
    >> come on and the rest are newer boxes doing 100MB which makes the
    >> yellow
    >> lights come on. You can tell when there's a lot of SCS traffic, it
    >> makes the green lights flash a lot.
    >>

    > I don't think yellow is good, at least not on a Cisco switch


    I think it's a Linksys switch. The documented meaning of the
    lights is simply 10MB vs. 100MB vs. no-link (dark).


  15. Re: speeding up LAVC with switch instead of hub?

    In article , helbig@astro.multiCLOTHESvax.de
    (Phillip Helbig---remove CLOTHES to reply) writes:

    > > (Run some math, and see what the theoretical best case for a full
    > > copy is, given your disk sizes and your NIC speeds. If that calculation
    > > is not hugely different from what you are experiencing, then the limit
    > > likely isn't the LAN infrastructure.)

    >
    > I get 2 hours for a full copy; it takes about 6 or 7.


    Considering that the VAX SCSI is, I believe, 5 Mb/s, then the ideal
    expected time would be 4 hours, so I suppose I have about 50% too much.


  16. Re: speeding up LAVC with switch instead of hub?

    In article , Hoff Hoffman
    writes:

    > A theoretical two hour transfer of a full volume copy over a 10 Mb
    > network implies circa seven GB disks, give or take, and particularly
    > assuming I've done the math correctly.


    Right; they're 9-GB disks.

    > Given the time to read the data
    > off and the time to write the data out and the glacial speeds and feeds
    > of most VAX systems and add a dollop of contention and given you that
    > won't ever hit the theoretical speed of an IEEE 802.3/Ethernet network,
    > and your six-to-seven hour environment doesn't look all that far out of
    > line.
    >
    > Old hardware -- old VAX systems, old I/O, old NICs -- is slow.


    Yes, but cheap (free), dependable (been running for 15 years
    continuously with no faults), small, quiet, consume little power, spare
    parts on hand etc.

    When I'm rich, I'll buy a faster computer. :-)


  17. Re: speeding up LAVC with switch instead of hub?

    "Phillip Helbig---remove CLOTHES to reply"
    schreef in bericht news:e4f2js$m11$1@online.de...
    > I'm looking to speed up communication in my LAN. I have a cluster with
    > all communication on the LAN (SCS, TCPIP, LAT etc; shadow copies/merges
    > go across the network as well). Now, everything is connected via a hub.
    >
    > If I understand things correctly, a switch gives me two advantages with
    > regard to speed: ethernet controllers can operate in full-duplex mode,
    > and collisions are eliminated.
    >
    > Full duplex would double the bandwidth, whereas avoiding collisions
    > would only increase bandwidth if collisions are reducing it, right?
    >
    > Normally, collisions aren't a problem and there are many of them only
    > during shadow copies and merges.
    >
    > To benefit from full duplex, the ethernet controllers would have to be
    > capable of full duplex. Would that be the case for any of the following
    > machines (which have the "stock" ethernet cards)?
    >
    > o VAXstation 4000/90
    >
    > o VAX 4000/105A
    >
    > o DEC 3000/600
    >
    > o DEC 3000/300LX
    >
    > o ALPHAserver 1200 (aka Digital Server 5305)
    >


    Hi Phillip

    The VAX systems and the DEC 3000 systems all use 10 Mb/s ethernet which is
    half duplex, even if you put a UTP transceiver on the AUI ports. The 5305
    came with a 100 Mb/s FD interface IIRC.
    A bridge is better than a repeater, err, a switch is preferable to a hub ;-)
    The best solution would be to buy a managed 100/10 switch and hard code the
    desired configuration for each port.
    BTW my 5305 runs at 100 Mb/s in HD mode. That way performance, read
    throughput, (especially noticable with DECnet) is much better between
    Alpha's.

    Hans



  18. Re: speeding up LAVC with switch instead of hub?

    In article <81b90$44735188$513b818b$5518@news.versatel.net>, "H Vlems"
    writes:

    > The best solution would be to buy a managed 100/10 switch and hard code the
    > desired configuration for each port.
    > BTW my 5305 runs at 100 Mb/s in HD mode. That way performance, read
    > throughput, (especially noticable with DECnet) is much better between
    > Alpha's.


    The network is full when shadow copies take place or complex web pages
    are viewed, otherwise there is not much activity. However, with 5 Mb/s
    SCSI on the VAXes, that will be the bottleneck for shadow copies.

    I only fire up the 5305 for Mozilla. I COULD use more bandwidth then,
    between it and the internet (I have 6 Mb/s DSL, so 10 Mb/s would be
    enough, but if it is split, then the hub might be decreasing
    performance). However, I have a free port on the switch in my DSL
    router, so I could plug the 5305 into that instead of the hub which is
    connected to the switch via an uplink.


  19. Re: speeding up LAVC with switch instead of hub?

    Phillip,
    when you connect the 5305 directly to the DSL router then you might
    want to measure performance between the FD and HD settings. I'd be
    interested to know if you measure differences. My 5305 lives in the
    attic and it's too b*y heavy to take it downstairs!
    Hans


  20. Re: speeding up LAVC with switch instead of hub?

    In article <1148451629.383312.318900@g10g2000cwb.googlegroups. com>, "H
    Vlems" writes:

    > when you connect the 5305 directly to the DSL router then you might
    > want to measure performance between the FD and HD settings. I'd be
    > interested to know if you measure differences. My 5305 lives in the
    > attic and it's too b*y heavy to take it downstairs!


    No difference! :-(

    I had been asking about whether it was worth it to replace my hub with a
    switch. For the VAXes, probably not, since when I need speed is during
    a shadow copy, but the bottleneck there is the 5 Mb/s SCSI of the VAXes.
    However, I have a 6 Mb/s DSL connection and it would be nice to be able
    to use that speed, especially when running CSWB on the 5305.

    Normally, I have all the VMS machines on a hub (which itself is probably
    capable of only 10 Mb/s) and the hub is connected via an uplink to the
    switch on a DSL router. This switch can do 100 Mb/s full duplex. I
    have a VOIP box connected to it, and the "full" and "100" lights light
    up. Also, a PC can make use of the "full" and "100" lights.

    My ISP has a web page where one can test the current speed. From the
    PC, I get the advertised 6 Mb/s. From one of the machines in my
    cluster, it varies between a few hundred kb/s and 2 Mb/s.

    I connected the 5305 directly to the switch and booted it. It still
    functioned OK in the cluster. However, the speed indicated by my ISP
    was still the slow one, and also the "full" and "100" lights did not
    light up on the switch. I was hoping that, even if it has to
    communicate with my cluster at 10 Mb/s half duplex, perhaps it could
    communicate with the internet at 100 Mb/s full duplex. That doesn't
    seem to be the case, though.

    Of course, it still has to communicate with the cluster at the slow
    speed. Maybe that slows down ALL traffic on its ethernet card.

    Replacing my 10 Mb/s hub with a 10 Mb/s switch would offer only a
    minimal improvement. Would it make sense to replace it with a 100 Mb/s
    hub or switch? (If I buy something new, it might as well be a switch.)
    (Of course, it would have to be capable of negotiating at 10 Mb/s as
    well.)

    Of course, this assumes that the 5305 is capable of 100 Mb/s and/or full
    duplex.

    I would like to have the 5305 communicate with the internet at 6 Mb/s
    (or more, if I have a faster connection in the future) even if it has to
    communicate with the cluster at 10 Mb/s. One way around this might be
    to have two ethernet cards in the 5305 and use one for cluster
    communication and one for ethernet communication, but that would be a
    lot of work, I have no experience with that and it would provide no
    performance improvement at all for the rest of the cluster (except for
    reducing the number of collisions somewhat).

    I had hoped that plugging the 5305 directly into the switch, with the
    rest of the nodes in the cluster connected to a hub uplinked to the
    switch, would allow the 5305 to communicate faster with the internet,
    but apparently that's not the case.

    Of course, if bandwidth-splitting is the main problem, then a faster
    internet connection would present a larger pie to be split up among the
    various connections, but naturally it would be better for one node to
    make full use of the high speed if there is no other traffic.


    Cheers,

    Phillip



+ Reply to Thread
Page 1 of 2 1 2 LastLast