Slow, Slow Super Slow

[itelecom]

Why is The Remote Administration on these Windows Servers so slow? I have servers at other two locations and just got a server with ServerBeach 'cause I need to concentrate my servers at a single location, but I can't understand why my server is so slow. I have a full T1 and the performance is bad. The CPU usage looks normal and so does memory. Any ideas? Also what's the idea with using 10 mbps cards? is full duplex enabled on this cards?

[knightfoo]

Are other services on your server slow? You should try downloading a fair-sized binary file from your server via HTTP and see how long it takes. As for the network cards, they are 10/100Mbit cards .. however, the switch port is 10Mbit half-duplex. Full-duplex is not part of the 10Mbit specification and autonegotiation does not work. In order to use 10Mbit, every switch port and NIC would have to be hard set at 10Mbit full-duplex. It is quite a hassle to do this and can make network troubleshooting a nightmare. In my experience, 10Mbit half-duplex is more than enough to allow most people to get themselves in trouble.

-knightfoo

[itelecom]

I think something wrong with the actual server. It takes 32 seconds to load IIS Manager, etc. Internet Explorer also takes a long time to open. I have opened a ticked with support. Hopefully they can shed some light on this. On the NIC speed I don't share your thoughts. Running NICs on half duplex causes colission and there is really no reason why we should run NICs on half duplex nowadays. It seems to me upgrading the switches to support 100mbps should be the right move. While using more than 10mbps at a time does not really happen often, I have found during internal test on my switched network that using 10mbps causes more network problems than their 100mbps counterparts. If you have a switch with 24 ports and all ports are 10mbps that means you would have 24 servers or at least 20 sharing a 10mbps bus and that's simply a limiting factor. Just my two cents worth!

[knightfoo]

Well, the switch backplane and uplink can handle the bandwidth of all 24 ports doing a full 10Mbps, so the switch is not going to be a bottleneck. These are quality Cisco switches we are talking about, not $100 BestBuy special. ;) You will get collisions on any half-duplex link, but as long as the number of collisions is less than 10% of the total packets transmitted and received, you are OK. The limiting factor to network design is the cost of the switch .. a 100Mbps switch costs about 5 times more than a 10Mbps switch. At 10Mbps half-duplex, your server can still push 1.6TB to 2.5TB per month.

-knightfoo

[Forti]

Is this ammount of collisions good or bad ?
how can i get rid of the collisions?
Thank you




collisions:248600 txqueuelen:100


lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:2739 errors:0 dropped:0 overruns:0 frame:0
TX packets:2739 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:815109 (796.0 Kb) TX bytes:815109 (796.0 Kb)

[root@cs root]# dmesg | grep eth
divert: allocating divert_blk for eth0
eth0: RealTek RTL8139 Fast Ethernet at 0xf8936000, 00:40:ca:38:fc:b4, IRQ 11
eth0: Identified 8139 chip type 'RTL-8139C'
eth0: Setting half-duplex based on auto-negotiated partner ability 0000.
eth0: Promiscuous mode enabled.
device eth0 entered promiscuous mode
device eth0 left promiscuous mode
eth0: Promiscuous mode enabled.
device eth0 entered promiscuous mode
device eth0 left promiscuous mode
[root@cs root]# eth0 Link encap:Ethernet HWaddr 00:40:CA:38:FC:B4
-bash: eth0: command not found

[knightfoo]

Looking for collisions on the "lo" interface won't get you very far. Since you are on a half-duplex switch port, you will always have collisions. This is a good way to determine if you have an acceptable amount of collisions:

collisions / ( RX packets + TX packets ) * 100 = %

(collisions divided by RX + TX times 100)

As long as your collision rate is below 10% then you are OK.

-knightfoo

[jaltuve]

No way this is normal...

eth0 Link encap:Ethernet HWaddr 00:20:ED:6C:60:2D
inet addr:66.135.32.97 Bcast:66.135.32.127 Mask:255.255.255.192
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:10083498 errors:0 dropped:0 overruns:0 frame:0
TX packets:18707819 errors:0 dropped:0 overruns:0 carrier:0
collisions:5284889 txqueuelen:100
RX bytes:797306293 (760.3 Mb) TX bytes:979017062 (933.6 Mb)
Interrupt:12 Base address:0x8f00

collisions:5284889 NO Way... I have two servers with you guys the other shows a similiar problem look:

eth0 Link encap:Ethernet HWaddr 00:20:ED:71:46:FA
inet addr:66.135.34.75 Bcast:66.135.34.127 Mask:255.255.255.192
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:6275552 errors:0 dropped:0 overruns:0 frame:0
TX packets:3808011 errors:0 dropped:0 overruns:0 carrier:0
collisions:1571364 txqueuelen:100
RX bytes:4139967978 (3948.1 Mb) TX bytes:1444986245 (1378.0 Mb)
Interrupt:12 Base address:0xbf00

And to make matters worse EVERY TIME this collisions happens my Apache web server process grows and grows until it completely kills my machine, can someody please tell me what's going on?

[Einewton]

Think it could be a DoS Attack?

[knightfoo]

Originally posted by jaltuve
No way this is normal...

eth0 Link encap:Ethernet HWaddr 00:20:ED:6C:60:2D
inet addr:66.135.32.97 Bcast:66.135.32.127 Mask:255.255.255.192
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:10083498 errors:0 dropped:0 overruns:0 frame:0
TX packets:18707819 errors:0 dropped:0 overruns:0 carrier:0
collisions:5284889 txqueuelen:100
RX bytes:797306293 (760.3 Mb) TX bytes:979017062 (933.6 Mb)
Interrupt:12 Base address:0x8f00

collisions:5284889 NO Way...

This looks like a hardware problem, considering the amount of traffic this server is pushing (almost none) and the number of collisions. You should submit a ticket for hardware troubleshooting.

I have two servers with you guys the other shows a similiar problem look:

eth0 Link encap:Ethernet HWaddr 00:20:ED:71:46:FA
inet addr:66.135.34.75 Bcast:66.135.34.127 Mask:255.255.255.192
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:6275552 errors:0 dropped:0 overruns:0 frame:0
TX packets:3808011 errors:0 dropped:0 overruns:0 carrier:0
collisions:1571364 txqueuelen:100
RX bytes:4139967978 (3948.1 Mb) TX bytes:1444986245 (1378.0 Mb)
Interrupt:12 Base address:0xbf00

And to make matters worse EVERY TIME this collisions happens my Apache web server process grows and grows until it completely kills my machine, can someody please tell me what's going on?

This second server has been spiking to 8Mbit of traffic for the past 2 days. The number of collisions increases exponentially with throughput, so it looks like this server is just very busy. Collisions will not cause your web server to use more resources .. more concurrent connections and more traffic will cause collisions and increased resource usage.

-knightfoo

[jaltuve]

In either case I think the problem will continue as long as there's a faulty NIC. What i'm thinking now is that the server with the faulty NIC is the one causing the collisions and problems in the good server also, so this is a critical problem since it's completely affecting my two servers in serverebeach, can someone help me and schedule an urgent NIC replacement ASAP?

Originally posted by knightfoo
This looks like a hardware problem, considering the amount of traffic this server is pushing (almost none) and the number of collisions. You should submit a ticket for hardware troubleshooting.

[b]

This second server has been spiking to 8Mbit of traffic for the past 2 days. The number of collisions increases exponentially with throughput, so it looks like this server is just very busy. Collisions will not cause your web server to use more resources .. more concurrent connections and more traffic will cause collisions and increased resource usage.

-knightfoo

[knightfoo]

Originally posted by jaltuve
In either case I think the problem will continue as long as there's a faulty NIC. What i'm thinking now is that the server with the faulty NIC is the one causing the collisions and problems in the good server also, so this is a critical problem since it's completely affecting my two servers in serverebeach, can someone help me and schedule an urgent NIC replacement ASAP?

You need to communicate this to support, not the forums.

-knightfoo

[jaltuve]

Ok, I think I know what's going on.

I checked my rddtool graphs for bandwidth consumption at the times I experienced the problems and they all peak at 1Mb. There is a bandwidth cap set at 1Mb causing me the trouble, what I don't understand is why you did not mention the bandwidth cap in the first place.


Originally posted by knightfoo
You need to communicate this to support, not the forums.

-knightfoo

[QT]

Originally posted by jaltuve
There is a bandwidth cap set at 1Mb causing me the trouble, what I don't understand is why you did not mention the bandwidth cap in the first place.

Stated from another thread here:

Originally posted by knightfoo
To clarify the bandwidth issue, all of the servers are on a 10Mbps switch port. The bandwidth is not capped but it is physically impossible to use more than 10Mbps.

[jaltuve]

I've been closely monitoring my server and to make troubleshooting easier I'm now using a single server. I;ve polled statistics at regular intervals and I've noticed that 'the ammount of collisions is NOT proportional to the ammount of traffic on my server, so the collisions are the results of heavy traffic at the specific hub where i'm connected, do you monitor traffic in specific clusters within your network? if you were using full duplex this would not be a problem, but since half duplex is being used in the hubs then traffic for your other clients if the specific hub where i'm connected is affecting me.

Is there a solution to this problem or is time to start looking for another provider? I know ServerBeach is a discount provider, so I don't expect to be connected to a 100Mb/s switched enviroment.


Originally posted by knightfoo
You need to communicate this to support, not the forums.

-knightfoo

[jaltuve]

This is not correct, based on my observations, measurements and statistics you are not using 10Mb switched ports but rather 10Mb/s hubs using half duplex, so ALL ports are sharing a 10Mb connection. I've nagios and rddtools graphs including my own perl scripts doing regular stat collection and the number of collisions are not proportional to the traffic i'm getting, so it is other port/client in the hub affecting me, this is a fact and no, I don't expect a miracle solution since I know this is a discount provider, I just want to know if there's alternatives, like have my servers in a network cluster where switched ports are being used. (it doesn't matter if I have to pay more, I just need to know if you can provide this) otherwise I'll need to look somewhere else, and please don't get me wrong.. ServerBeach IS a wonderful provider, problem is I don't have a low traffic site anymore and the high traffic is causing me trouble.


Originally posted by QT
Stated from another thread here:

[jaltuve]

knightfoo,

Did you or anyone at serverbeach change something? no collisions, no problems. no peaks, no warnings or alarms in a full 48 hour period in my monitoring software, and ping times to the default router that were normally taking 1-2 ms are now taking 0.02 ms, please if you did something let me know.

Thanks,

Originally posted by knightfoo
You need to communicate this to support, not the forums.

-knightfoo