Monday-morning network sluggishness

[OLARU Dan]

What kind of coffee should I pour into my network so that these monday-morning symptoms go away:

1. broadcasts for DHCP addresses: all clients request an IP address monday morning

2. ARP broadcasts from clients

3. switches flooding frames out all ports to fill-in their MAC Tables

4. and, ontop of that, all users trying to login and read their e-mail in vain and phoning me about "Hey! I can't see the network #$@!~$%&*^"

[John Collier]

Dan,

If I didn't know any better I would think that you were in an organization that shuts down all of the client PCs at night (or at least every weekend). If this is the case, then I would have to say that no amount of caffeine I can think of will be of much use.

What you describe sounds very much like a network during an initial power-on of all (or a great deal of) network components.

If this is the case, then you really don't need the coffee for the network, you need Valume for the users so they will stay off your back!

Let me know if I have assessed this situation incorrectly.

John

[OLARU Dan]

John, you're right. Around 600 w2k+wxp machines go to sleep every evening, and then wake up each and every morning.

Tuesday to Friday I don't seem to have this problem, tho. I have it only monday morning: and the only thing happening then, compared to other mornings, is the DHCP stuff.

I was thinking of setting the DHCP lease period to 60 days (or maybe more), so this would happen 6 times a year (or maybe less)- which is OK. What would be the downside of such a long lease interval? What are the minimum and maximum values for this lease period? Has anyone experimented with this?

How about fiddling with the ARP Timeout values, on both clients and the inter-VLAN router (a Cisco 3550)? I could easily set the ARP Timeout for all interfaces at 72 hours, to cover for the weekend. But what impact does this have on the network performance?

What about having the MAC Age Time of 72 hours on my ProCurves? How is this impacting the performance of the network when people move their computers around? Put a long time when the network is stable, and put 10 minutes when I know people will move around?

[Ron Kinner · Accepted solution]

Set your DHCP timeout to larger than 72 hours. It doesn't hurt anything unless you have a very limited address pool with a lot of different computers just dropping in for a day or so in which case you might eventually dry up the address pool.

ARP broadcasts are normal and shouldn't be that much worse than during the rest of the week. Routers keep their arp tables entries for 4 hours but most other devices time out in minutes. I wouldn't bother changing them.

MAC tables need only change if your users plug into different ports so unless the company is playing musical chairs you can stretch their timeout without a major problem. You might even set static entries on your servers' ports. (Just don't forget you have done so since that really makes for a strange problem if you move it and don't erase the static entry.)

Is Wake on Lan an option? Perhaps you could have a cron process wake all of the computers up about 15-30 minutes before starting time? Your users would love not having to wait for the POST every morning. You could even put in for an efficiency improvement award. Saving 600 users times 2 minutes times x startups per year.

Ron