Sign in with
Sign up | Sign in
Your question

Switch Questions

Last response: in Networking
Share
Anonymous
August 22, 2005 1:02:32 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

I manage a small network (75 desktops, 4 servers) with a single HP
Procurve 4000m switch which has one 1-port gigabit module and the other
7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
only hook one of the NICS on one of the servers to it. What I'm
thinking of doing is purchasing a separate smaller Linksys 16 port
gigabit switch (under $400) so I can team the dual NICS on both of the
proliant servers to get 2GB on each. FWIW, the two servers are a
terminal server which hosts about 15-20 users (currently dual NIC'd at
200 Mbps), and our main database/file/print server (only one NIC at
1GBps) which sees some pretty heavy use. Both are win2k.

My questions, therefore, are:

1) I'm guessing I can plug the Linksys into the Procurve through the
gigabit port, correct? Should I be using a CAT 6 cable for this or
will CAT 5e do? If this is not how I would hook this up, what are the
options? Should I take the cable coming from the router/firewall and
plug it into the Linksys, then plug the Procurve into the Linksys?

2) Since the Linksys is a managed switch I could configure Port
trunking. Is this something that should always be done when setting up
dual NICS (primary goal is load balancing with lesser emphasis on fault
tolerance)? I tried port trunking when configuring dual 100 NICS on
the terminal server, but saw no readily apparent benefit. HP Procurve
has told me different stories each time I've called them.

3) Will this scenario actually give me noticeable improvements? I
don't want to invest even a little bit of money if improvements are
minimal.

Thanks for any insight into this.
Brian

More about : switch questions

Anonymous
August 22, 2005 1:55:08 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

The Linksys switch I'm looking at is the SRW2016 which indicates
managed rackmount with webview. Price after rebate is around $375 at
CDW online.

I configured port trunking on the Procurve by telnetting into the
switch, and selecting the two ports being used by the NIC team, then I
configured the group as "Trk1" saved and exited.

I've heard the cisco products have better throughput, but am not sure I
can afford the Cisco price. I quickly looked for the Cat2960, but
haven't come up with anything. Other Recommendations?

Thanks for the help-
Brian
Anonymous
August 22, 2005 4:12:33 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <1124726552.628160.58700@g47g2000cwa.googlegroups.com>,
<bpanders71@hotmail.com> wrote:
>I manage a small network (75 desktops, 4 servers) with a single HP
>Procurve 4000m switch which has one 1-port gigabit module and the other
>7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
>with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
>only hook one of the NICS on one of the servers to it. What I'm
>thinking of doing is purchasing a separate smaller Linksys 16 port
>gigabit switch (under $400) so I can team the dual NICS on both of the
>proliant servers to get 2GB on each. FWIW, the two servers are a
>terminal server which hosts about 15-20 users (currently dual NIC'd at
>200 Mbps), and our main database/file/print server (only one NIC at
>1GBps) which sees some pretty heavy use. Both are win2k.
>


I think with the right 100MB cards in a server it can be bandwidth
aggregated to use two or three cards. Never done it, though.

I'd avoid an unmanaged switch in any production environment,
especially one with a non-trivial topology.

With "only" 75 desktops I'd be suprised if you are bandwidth-limited
on more than one server. Learn to use the management screens in your
4000m (a nice box) to see what your traffic is. When you find your
busy server use perfmon on it to see what it's bottlenecked on.


--
a d y k e s @ p a n i x . c o m

Don't blame me. I voted for Gore.
Related resources
Anonymous
August 22, 2005 5:57:23 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

bpanders71@hotmail.com wrote:

> I manage a small network (75 desktops, 4 servers) with a single HP
> Procurve 4000m switch which has one 1-port gigabit module and the other
> 7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
> with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
> only hook one of the NICS on one of the servers to it. What I'm
> thinking of doing is purchasing a separate smaller Linksys 16 port
> gigabit switch (under $400) so I can team the dual NICS on both of the
> proliant servers to get 2GB on each.

Read the specs carefully. If you're talking about the SRW2016, there's no
mention of either 802.3ad or Fast Etherchannel, and I can't imagine a Cisco
product supporting any other variant of aggregation if it doesn't support
those.

> FWIW, the two servers are a
> terminal server which hosts about 15-20 users (currently dual NIC'd at
> 200 Mbps), and our main database/file/print server (only one NIC at
> 1GBps) which sees some pretty heavy use. Both are win2k.

Are you looking for performance or fault tolerance?

If you're looking for performance, the first thing to consider is that
you're still only going to have a 1 Gb pipe from the second switch into the
Procurve, so unless you're encountering high traffic between servers you're
not going to gain anything. You'd at the minimum need to put a second
gigabit port in the Procurve.

Next, Terminal Services can run fine over a dialup line. 20 sessions under
worst case conditions are still getting 5 MB/sec out of 100TX--you're not
going to see much gain giving that server more bandwidth unless there's
something you're not telling us.

Next, what transfer rate do you get out of your file/print/database server
now? Filling even a 1 GB/sec pipe more than momentarily takes an
exceedingly heavy duty server and if it's to be done steady state on disk
access you need rather massive storage system.

If you're going for fault tolerance, then you're back to the single link
between the second switch and the Procurve, so you don't really gain
anything.

If you really want aggregation, you'd do better to put three more gigabit
ports in the Procurve, which will cost you about $1000.

> My questions, therefore, are:
>
> 1) I'm guessing I can plug the Linksys into the Procurve through the
> gigabit port, correct? Should I be using a CAT 6 cable for this or
> will CAT 5e do?

Yes, generally speaking any 1000TX device can be connected to any other in
the manner you describe.

Generally speaking CAT5E is adequate for 1000TX--it was _designed_ for CAT5,
5E nails down some numbers that most existing CAT5 meets but was never
tested for, and one of the gigabit vendors demonstrated at a trade show
that it can run error-free over 8 strands of barbed wire.

If it's not running properly on CAT5E and there's no unusual source of EMI
then something is broken.

> If this is not how I would hook this up, what are the
> options? Should I take the cable coming from the router/firewall and
> plug it into the Linksys, then plug the Procurve into the Linksys?

No reason to use a gigabit port for the router unless you've got an unusual
situation--most small businesses have at most 3 Mb/sec or so access.

> 2) Since the Linksys is a managed switch I could configure Port
> trunking. Is this something that should always be done when setting up
> dual NICS (primary goal is load balancing with lesser emphasis on fault
> tolerance)? I tried port trunking when configuring dual 100 NICS on
> the terminal server, but saw no readily apparent benefit. HP Procurve
> has told me different stories each time I've called them.

No. What you seem to be trying to do is called "aggregation", not
"trunking" (although sometimes aggregation is referred to as trunking for
some reason). If you are looking for load balancing and fault tolerance
then aggregation is what you need to use.

> 3) Will this scenario actually give me noticeable improvements? I
> don't want to invest even a little bit of money if improvements are
> minimal.

Based on what you've said so far it seems unlikely.

> Thanks for any insight into this.
> Brian

--
--John
to email, dial "usenet" and validate
(was jclarke at eye bee em dot net)
Anonymous
August 22, 2005 8:30:50 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <1124726552.628160.58700@g47g2000cwa.googlegroups.com>,
<bpanders71@hotmail.com> wrote:
:I manage a small network (75 desktops, 4 servers) with a single HP
:p rocurve 4000m switch which has one 1-port gigabit module and the other
:7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
:with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
:o nly hook one of the NICS on one of the servers to it. What I'm
:thinking of doing is purchasing a separate smaller Linksys 16 port
:gigabit switch (under $400) so I can team the dual NICS on both of the
:p roliant servers to get 2GB on each. FWIW, the two servers are a
:terminal server which hosts about 15-20 users (currently dual NIC'd at
:200 Mbps), and our main database/file/print server (only one NIC at
:1GBps) which sees some pretty heavy use. Both are win2k.

I'd think it likely that that plan is headed for trouble.


:My questions, therefore, are:

:1) I'm guessing I can plug the Linksys into the Procurve through the
:gigabit port, correct? Should I be using a CAT 6 cable for this or
:will CAT 5e do?

Cat5e will do, at least for moderate distances.


:2) Since the Linksys is a managed switch

Could you check that again? Last time I checked about 3 months ago,
the least expensive *managed* gigabit switch I could find was about $1000.
The $400 Linksys switches were NOT managed at the time.

:I could configure Port
:trunking. Is this something that should always be done when setting up
:D ual NICS (primary goal is load balancing with lesser emphasis on fault
:tolerance)?

Port trunking? As in putting in multiple 802.1Q VLANs? That's not going
to help fault tolerance: if your fault tolerance is MAC based then
the MAC in the other VLAN would not be visible; if your fault tolerance
is IP based then the two ports would have to be in different subnets
in order to be able to select the proper port during normal operations.


: I tried port trunking when configuring dual 100 NICS on
:the terminal server, but saw no readily apparent benefit.

Do you perhaps mean port aggregation?

Port aggregation depends on the algorithm being used to distribute
the packets between the available links. I don't recall which algorithms
the Procurves have available. Often the algorithms are based upon
taking the last N bits (2^N >= number of aggregated ports) of the source
or destination and using that as the link index; sometimes the source
and destination bits are xor'd. Unless you have configured per-packet
load balancing on your aggregation ports (I -seem- to recall seeing
that the Procurve could do that), the result is that any one
source + destination MAC pair will always use the same link. Aggregating
N ports does not get you N times the bandwidth (unless per-packet):
instead it gives you opportunities to parallelize up to N different
streams. Like a multilane highway: your car can't drive at 180 miles
per hour just because there are three lanes, but 3 cars can independantly
drive at 60 miles per hour on the three lanes.


:3) Will this scenario actually give me noticeable improvements? I
:D on't want to invest even a little bit of money if improvements are
:minimal.

If you are expecting heavy traffic over the gigabit ports, then
chances are that you will be disappointed with the $400 Linksys.
On the consumer gigabit devices, the aggregate throughput supported
is usually fairly low. Many of the consumer gigabit devices use the
same chipset, so you pretty much have to take a jump in price class
in order to get serious simultaneous bandwidth. Something like a
Cisco Cat2960 perhaps; I imagine others will pitch in with other
reliable high-throughput devices. [Note: there have been some reports
of bad experiences with Netgear's line of managed gigabit switches,
which look nice on paper...]
--
Look out, there are llamas!
Anonymous
August 22, 2005 8:30:51 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <decujq$rvu$1@canopus.cc.umanitoba.ca>,
Walter Roberson <roberson@ibd.nrc-cnrc.gc.ca> wrote:
>In article <1124726552.628160.58700@g47g2000cwa.googlegroups.com>,
> <bpanders71@hotmail.com> wrote:
>:I manage a small network (75 desktops, 4 servers) with a single HP
>:p rocurve 4000m switch which has one 1-port gigabit module and the other
>:7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
>:with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
>:o nly hook one of the NICS on one of the servers to it. What I'm
>:thinking of doing is purchasing a separate smaller Linksys 16 port
>:gigabit switch (under $400) so I can team the dual NICS on both of the
>:p roliant servers to get 2GB on each. FWIW, the two servers are a
>:terminal server which hosts about 15-20 users (currently dual NIC'd at
>:200 Mbps), and our main database/file/print server (only one NIC at
>:1GBps) which sees some pretty heavy use. Both are win2k.
>
>I'd think it likely that that plan is headed for trouble.
>
>
>:My questions, therefore, are:
>
>:1) I'm guessing I can plug the Linksys into the Procurve through the
>:gigabit port, correct? Should I be using a CAT 6 cable for this or
>:will CAT 5e do?
>
>Cat5e will do, at least for moderate distances.
>
>
>:2) Since the Linksys is a managed switch
>
>Could you check that again? Last time I checked about 3 months ago,
>the least expensive *managed* gigabit switch I could find was about $1000.
>The $400 Linksys switches were NOT managed at the time.
>


I looked at the specs for this switch and they seem to avoid the word
"managed". Since Linsys is owned by Cicso I assume that want to
distinguish the Cicso products from the Linksys line I wonder what the
difference is.

I'd d/l and read the manual before I bought it.


--
a d y k e s @ p a n i x . c o m

Don't blame me. I voted for Gore.
Anonymous
August 22, 2005 8:30:51 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <decujq$rvu$1@canopus.cc.umanitoba.ca>,
Walter Roberson <roberson@ibd.nrc-cnrc.gc.ca> wrote:
>In article <1124726552.628160.58700@g47g2000cwa.googlegroups.com>,
> <bpanders71@hotmail.com> wrote:
>:I manage a small network (75 desktops, 4 servers) with a single HP
>:p rocurve 4000m switch which has one 1-port gigabit module and the other
>:7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
>:with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
>:o nly hook one of the NICS on one of the servers to it. What I'm
>:thinking of doing is purchasing a separate smaller Linksys 16 port
>:gigabit switch (under $400) so I can team the dual NICS on both of the
>:p roliant servers to get 2GB on each. FWIW, the two servers are a
>:terminal server which hosts about 15-20 users (currently dual NIC'd at
>:200 Mbps), and our main database/file/print server (only one NIC at
>:1GBps) which sees some pretty heavy use. Both are win2k.
>
>I'd think it likely that that plan is headed for trouble.
>
>
>:My questions, therefore, are:
>
>:1) I'm guessing I can plug the Linksys into the Procurve through the
>:gigabit port, correct? Should I be using a CAT 6 cable for this or
>:will CAT 5e do?
>
>Cat5e will do, at least for moderate distances.
>
>
>:2) Since the Linksys is a managed switch
>
>Could you check that again? Last time I checked about 3 months ago,
>the least expensive *managed* gigabit switch I could find was about $1000.
>The $400 Linksys switches were NOT managed at the time.
>
>:I could configure Port
>:trunking. Is this something that should always be done when setting up
>:D ual NICS (primary goal is load balancing with lesser emphasis on fault
>:tolerance)?
>
>Port trunking? As in putting in multiple 802.1Q VLANs? That's not going
>to help fault tolerance: if your fault tolerance is MAC based then
>the MAC in the other VLAN would not be visible; if your fault tolerance
>is IP based then the two ports would have to be in different subnets
>in order to be able to select the proper port during normal operations.
>
>
>: I tried port trunking when configuring dual 100 NICS on
>:the terminal server, but saw no readily apparent benefit.
>
>Do you perhaps mean port aggregation?
>
>Port aggregation depends on the algorithm being used to distribute
>the packets between the available links. I don't recall which algorithms
>the Procurves have available. Often the algorithms are based upon
>taking the last N bits (2^N >= number of aggregated ports) of the source
>or destination and using that as the link index; sometimes the source
>and destination bits are xor'd. Unless you have configured per-packet
>load balancing on your aggregation ports (I -seem- to recall seeing
>that the Procurve could do that), the result is that any one
>source + destination MAC pair will always use the same link. Aggregating
>N ports does not get you N times the bandwidth (unless per-packet):
>instead it gives you opportunities to parallelize up to N different
>streams. Like a multilane highway: your car can't drive at 180 miles
>per hour just because there are three lanes, but 3 cars can independantly
>drive at 60 miles per hour on the three lanes.
>
>
>:3) Will this scenario actually give me noticeable improvements? I
>:D on't want to invest even a little bit of money if improvements are
>:minimal.
>
>If you are expecting heavy traffic over the gigabit ports, then
>chances are that you will be disappointed with the $400 Linksys.
>On the consumer gigabit devices, the aggregate throughput supported
>is usually fairly low. Many of the consumer gigabit devices use the
>same chipset, so you pretty much have to take a jump in price class
>in order to get serious simultaneous bandwidth. Something like a
>Cisco Cat2960 perhaps; I imagine others will pitch in with other
>reliable high-throughput devices. [Note: there have been some reports
>of bad experiences with Netgear's line of managed gigabit switches,
>which look nice on paper...]
>--
> Look out, there are llamas!


have you done some measurements and determined traffic to each of your
servers? Do you know that any of your servers are
bandwidth-bottlenecked?




--
a d y k e s @ p a n i x . c o m

Don't blame me. I voted for Gore.
Anonymous
August 22, 2005 8:30:51 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Walter Roberson wrote:

> In article <1124726552.628160.58700@g47g2000cwa.googlegroups.com>,
> <bpanders71@hotmail.com> wrote:
> :I manage a small network (75 desktops, 4 servers) with a single HP
> :p rocurve 4000m switch which has one 1-port gigabit module and the other
> :7 modules having ten 10/100 ports. I've got two HP Proliant DL servers
> :with dual gigabit NICS. Obvioulsy, with only one gigabit port, I can
> :o nly hook one of the NICS on one of the servers to it. What I'm
> :thinking of doing is purchasing a separate smaller Linksys 16 port
> :gigabit switch (under $400) so I can team the dual NICS on both of the
> :p roliant servers to get 2GB on each. FWIW, the two servers are a
> :terminal server which hosts about 15-20 users (currently dual NIC'd at
> :200 Mbps), and our main database/file/print server (only one NIC at
> :1GBps) which sees some pretty heavy use. Both are win2k.
>
> I'd think it likely that that plan is headed for trouble.
>
>
> :My questions, therefore, are:
>
> :1) I'm guessing I can plug the Linksys into the Procurve through the
> :gigabit port, correct? Should I be using a CAT 6 cable for this or
> :will CAT 5e do?
>
> Cat5e will do, at least for moderate distances.
>
>
> :2) Since the Linksys is a managed switch
>
> Could you check that again? Last time I checked about 3 months ago,
> the least expensive *managed* gigabit switch I could find was about $1000.
> The $400 Linksys switches were NOT managed at the time.

There seem to have been some price reductions. Linksys does indeed seem to
have some gigabit switches offering some degree of manageability (or at
least configurability) for under $400, see for example the SRW2016--we may
be on the way to seeing managed switches become consumer devices--at that
price I'm almost tempted to get one for my home network.

On the other hand, it does not seem to support any kind of aggregation,
which defeats the OP's purpose.

<snip>

--
--John
to email, dial "usenet" and validate
(was jclarke at eye bee em dot net)
Anonymous
August 22, 2005 8:30:52 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Al Dykes wrote:

> I looked at the specs for this switch and they seem to avoid the word
> "managed". Since Linsys is owned by Cicso I assume that want to
> distinguish the Cicso products from the Linksys line I wonder what the
> difference is.

Price? ;-)
Anonymous
August 22, 2005 10:37:16 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <1124729708.785441.37570@o13g2000cwo.googlegroups.com>,
<bpanders71@hotmail.com> wrote:
:The Linksys switch I'm looking at is the SRW2016 which indicates
:managed rackmount with webview. Price after rebate is around $375 at
:CDW online.

Indeed, looking at the documentation it does claim to be non-blocking
and managed. (Non-blocking on all ports, though?)


:I configured port trunking on the Procurve by telnetting into the
:switch, and selecting the two ports being used by the NIC team, then I
:configured the group as "Trk1" saved and exited.

Sounds like port aggregation.


:I've heard the cisco products have better throughput, but am not sure I
:can afford the Cisco price. I quickly looked for the Cat2960, but
:haven't come up with anything. Other Recommendations?

Sorry, I misremembered the part number: it is Cat2970 . Unfortunately
the datasheets are mostly missing.

http://www.cisco.com/warp/public/cc/pd/si/casi/ps5206/i...
http://www.cisco.com/warp/public/cc/pd/si/casi/ps5206/p...

If you were going for that line, you might find the Cat3560G or
Cat3750G more to your liking; both are "multilayer switches" with
various degrees of routing and QoS and IPv6 capabilities. [Even if you
don't need the features now, if you are looking at the Cat2970,
price-compare against those other models, as the Cat3750G can come out
-less- than some of the others (volume sales, I guess.)]


Anyhow, this presumes that the SRW2016 cannot do the job. I checked
around and there are no public reviews of it that I could find.
Generally speaking, my experience in the past has been that devices
that are several months on the market already and have no public reviews,
tend to have limitations that aren't obvious from reading the spec sheet.

If you know a vendor well, perhaps you could get the SRW2016 on trial ?
And if you just want to play around with gigabit topologies and
aren't too worried about high throughput on multiple gigabit ports,
you could get one of those cheap (<$50) unmanaged gigabit switches
and give it a try.
--
Daylight is a trademark of OSRAM SYLVANIA INC.
Anonymous
August 23, 2005 3:03:52 PM

Archived from groups: comp.dcom.lans.ethernet (More info?)

Thanks for all the help and suggestions. Based on what everyone is
saying it sounds like I'd probably be better off to purchase three more
gigabit modules for the Procurve, which would displace 20+ other ports,
but then perhaps I could just get a standard 10/100 switch to handle
those.

I really don't know how much bandwidth these servers are eating up, but
in my wierd way of thinking thought that if I opened up the pipe a bit
it might allow for better traffic to the server. We have a touchy DB
app that frequently crashes and I figured if there was a bottleneck
causing the issue, a gigabit switch might help things out.

Thanks-
Brian
!