[Okta Webinar] Learn how to a build a cloud-first strategyRegister Now

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 453
  • Last Modified:

informations about network configuration in a client server environment

Hello all,
I need some documentation (better) or book references to unterstand in detail the network configuration (see advanced configuration) to improve performances in a client server with multiple vlans environment (i.e. jumbo frames, flow control, different types of spanning tree, ....)
please let me know.
thank you very much
Nick
0
nschwend
Asked:
nschwend
  • 9
  • 4
  • 4
  • +2
4 Solutions
 
giltjrCommented:
Well first thing is to prove that the problem is the network. Has that been done?  If so, how?
0
 
bbaoIT ConsultantCommented:
is it just for academic research or for a real enterprise environment?

nowadays most modern applications may be not running in typical or traditional C/S model hence the network topology, configuration and protection may be significantly different.

we better know more details for giving a more proper suggestion.
0
 
nschwendAuthor Commented:
hello all,
it's for real enterprise environment... we have some issues (slowness, dhcp delay, ...) and I want to know deeper if some protocols are raising this strange behaviours..
thank you very much

Nick
0
Concerto Cloud for Software Providers & ISVs

Can Concerto Cloud Services help you focus on evolving your application offerings, while delivering the best cloud experience to your customers? From DevOps to revenue models and customer support, the answer is yes!

Learn how Concerto can help you.

 
bbaoIT ConsultantCommented:
> slowness

do you mean high latency for PING or file sharing between local computers?

> dhcp delay

do you mean local computers are failed or having difficulty to obtain an IP from the DHCP server?
0
 
nschwendAuthor Commented:
do you mean high latency for PING or file sharing between local computers?
the ping seems ok, the file transfer (and all normal activity of an end user is quite slow (we tried with brand new computer and with old ones...)

do you mean local computers are failed or having difficulty to obtain an IP from the DHCP server?
yes exactly, we have issues in both PXE request and normal IP request (often the clint pc wont get the IP, but not always) .
This since we change the switches on the floors (HP 2920). On these switches I've enabled spanning tree option (default option) and configured static vlans, only


regards
Nick
0
 
giltjrCommented:
Can your define "slow" and describe the network connectivity between the hosts?

Examples:

1) It take 60 seconds to copy a 1 KB file, both computers are connected to the same LAN/IP subnet via 1 Gbps Ethernet.

2) It take 5 minutes to copy a 10GB file, computers have 1 Gbps Ethernet connection, but are connect via VPN tunnel over the Internet at two different sites.  Site#1 is located in the Eastern USA has cable 15Mbps/5Mpbs and Site #2 has 100 Mpbs Ethernet and is located in Eastern Europe.
0
 
bbaoIT ConsultantCommented:
how many VLANs does your network have? is there a central L3 switch doing routing?
0
 
nschwendAuthor Commented:
the switches are all 1Gb, the client workstation same.
Normally when you browse a folder on the DFS or open a simple excel or word file (less than 1MB) it takes a long time (20 seconds to 1 minute)...
there is no VPN or any well-known bottleneck... we have 40 servers connected to a couple of HP 3800 switches (in stack) and all the fibers are connected to 2x HP 3800. Before we had a network 100Mb by Nortel and was worst (at least for the file transfer).
but the DHCP delay has been introduced by replacing the switches with the HP.

hope I've been clearer
regards
Nick
0
 
nschwendAuthor Commented:
we have 34 VLANs, (5 VLANs for client PC, 1 for the servers and the rest for other segments).
the HP 3800 are doing the routing between VLANs
0
 
bbaoIT ConsultantCommented:
if all servers, especially file servers, are sitting in the same VLAN or subnet while most client computers on various VLANs ALWAYS access the centrally shared files on the servers via routing, the trunck connection could be a bottleneck especially if the connection has a single link.
0
 
nschwendAuthor Commented:
from the floor switches to the core switches we have 2 HP3800 that are connected to the HP 3800 servers switches via 2 fibers of 10Gb. I could (if this could improve the performances, also stack these 2 switches with the other ones.
thank you again for your interest
Nick
0
 
giltjrCommented:
I would start looking to make sure all interfaces are the same duplex.

Do you have any network management/monitoring tools?

If not I would suggest you get something to monitor link utilziation, drop packet/frames, ect.
0
 
nociSoftware EngineerCommented:
That is the first angle try to eliminate all things "not normal".

Like duplex errors, or mis negotiated speeds.. check the logs for anomalies, unexpected cross links, loops etc.
Also check how big your VLAN's are, better have several smaller networks then one huge one.
A tool for testing effective bandwidth (across many platforms) is netio
http://www.ars.de/ars/ars.nsf/docs/netio

Try to get effective statistics on port level, first on trunks later on all ports.
and try to identify bottlenecks.  cacti is a nice tool for this
   http://www.cacti.net/

Also last but certainly not least.
Make a drawing of your network for layer 2 and one layer 3.
And map the network topology on the real world topology.
Not as you want it to be, but based on info you find on the equipment. MAC tables, comments in config, routing tables, then compare it to the desired network and identify hotspots...
0
 
nschwendAuthor Commented:
I'm looking at the netio and cacti (actually I was using Observer).
but someone has some whitepapers on best practice on how to configure a network?

thank you again
Nick
0
 
nociSoftware EngineerCommented:
Most boil down to these:

Choose a valid design for your network (logical) so that all functionality required can be provided. Then translate that to a physical setup.
In general choose the simple design (KISS).  
Dynamic / auto configurations also leaves a possibility with misconfiguration. esp. is equipment from various vendors is used. So static infra cables can be static setup, whereas your endpoints can also be dynamic if needed [ think wired laptops f.e. ].

Hunt down all errors on links (errors cause retransmissions ==> delays). [ collisions are not errors on Half Duplex lines ]  
lines should be balanced, [ both sides of a cable the same, speed, duplex, etc. ].

if new equipment is needed, choose manageable equipment over "auto" configuring devices [ hubs/switches].
0
 
giltjrCommented:
Network Observer is not really a network management tool, not sure I would call it a network monitoring tool either.  It is more of a traffic analysis tool.  It's used to do packet captures to see what traffic is flowing through your network and normally to trouble shoot specific problems where you need to see the network traffic.

I've never heard of NetIO, so I will have to look at that.

Cacti is a good network tool.  It can be used to ping network devices/hosts to make sure they are up and using SNMP query information about the device, such as port utilization and network level errors.

I agree with noci's points.  Especially about getting managed devices.  Although strides have been made in auto speed/auto duplex negotiations it is really better to control it.  There are no question about what the speed is or what the duplex is.  In the long run you will save money by knowing vs. guessing.  

Also, most (maybe all) non-managed devices are fairly dumb.  You can't ping them and they don't support SNMP.  So tools like Cacti (PRTG, MRTG, Solarwinds, ect.) are useless because the device has nothing to allow you to monitor it.
0
 
nociSoftware EngineerCommented:
Not only duplex, but also LACP (link aggregation, teaming)  may fail horribly when the load rises. I have seen auto-LACP connections that collapsed under heavy load, only to add insult to the injury.

Netio is a bandwidth checking tool. It will sature a link with packets and tell you the real effective bandwidth, it does it in two directions and if the link is symetrical the speed should be too. Also the UDP mode can be used to get an estimate for packet loss.
Obviously not intended to use on a heavy loaded server, unless you want to check if it still has something left in the attick.
0
 
nschwendAuthor Commented:
thank you very much... I was guessing me too if LACP on the core switches introduced a delay... (at least it seems on the 4 nics of the dfs server)...
Right now I've installed cacti and I'm studying how to shows the correct informations I want...
0
 
nociSoftware EngineerCommented:
well lacp isn't introducing a delay, but under heavy traffic the negotiation frames for keeping the LACP live might get dropped, if a one to many gets dropped the LACP will dissolve and needs to be re negotiated again.

Help yourself by making this a fixed etherchannel (cisco) / Team / Bond / Trunk(hp)
even in a fixed setup one failing line won't be used. no need to use the auto configuring stuff for this.
0
 
nschwendAuthor Commented:
Ok, so I'll dissolve the LACP and I will create a Trunk...
thank you very much
0
 
vivigattCommented:
If you have PXE client, STP (Spanning Tree Protocol) must be set to "port-fast" or disabled on all routers and switches.
PXE prom should be updated (usually in the BIOS of the PXE client, so BIOS should be updated).
Switches and routers firmware must be updated too.
You may have to force link speed, some PXE drivers are not good at using auto-negotiation.

Flow control may have to be enabled on all devices, including clients (PXE and OS), switches, routers, servers.

For details about PXE/DHCP config and options, check my article and its comments:
http://www.experts-exchange.com/Networking/Misc/A_2978-PXEClient-what-is-it-for-Can-I-use-PXE-without-it.html
0
 
nschwendAuthor Commented:
thank you all for your replies
0

Featured Post

New feature and membership benefit!

New feature! Upgrade and increase expert visibility of your issues with Priority Questions.

  • 9
  • 4
  • 4
  • +2
Tackle projects and never again get stuck behind a technical roadblock.
Join Now