Multicast very slow
Multicast of 30 HP 6000 sff machines is only giving 170mib. However, on other vlans with other computers average multicast is around 600-700mib. Network is idle. No other jobs on fog.
Unicast to same machines gives around 800-900mib.
Fog server has 1Gb uplink and is idle. 170 mib per machine is not acceptable.
Sounds like you got this locked down to a problem that only happens to the HP 6000 sff when multicasting right? You’re unicast seems slow also. I average 3.5 GiB/min and you are only seeing 800-900 MiB/min.
Post a spec sheet of the HW and maybe we can compile a custom kernel to try.
Running 0.32 with more hp 6000 we are getting 50mib multicast.
sata operation mode is ahci.
There is something with the kernel and this hardware set that is making this really slow.
We were running fog 0.29. I upgraded to 0.32 and we are now getting in the range of 300-400mib. The problem is with these hp 6000 and their hard disk controller and or bios and or hard disk. The newer kernel included with 0.32 is working great. I will test again with the sata operation mode in “ide” I have seen others claim they get very high speeds with this change.
[quote=“chad-bisd, post: 5248, member: 18”]Are the similar machines at another location using this same FOG server? I’m trying to track down the differences between the two groups and why the speeds would be so different.
Yes, FOG is alive and kicking.[/quote]
BryceZ last edited by
The TTL is 32. It’s set in /opt/fog/service/common/lib/MulticastTask.Class.php.
I’ll clarify my question. What is the TTL that udp-sender sets for the multicast packets in fog?
If you have a support contract with Cisco, you might be able to open a ticket to help resolve this issue, or at least troubleshoot it from a switch/router point of view.
What is the ttl for multicast traffic? Im seeing alot of bad destination errors
sh ip traffic
IP Software Processed Traffic Statistics
Transmission and reception:
Packets received: 995490936, sent: 353032236, consumed: 2062,
Forwarded, unicast: 26241102, multicast: 0, Label: 0
end: 0, nop: 0, basic security: 0, loose source route: 0
timestamp: 0, record route: 0
strict source route: 0, alert: 123715127,
Bad checksum: 2, packet too small: 0, bad version: 0,
Bad header length: 0, bad packet length: 0, bad destination: 9012862,
Bad ttl: 12967115, could not forward: 210684575, no buffer dropped: 0,
Bad encapsulation: 263693866, no route: 63460, non-existent protocol: 0
Stateful Restart Recovery: 0
MBUF pull up fail: 0
Fragments received: 0, fragments sent: 8000, fragments created: 8002,
Fragments dropped: 0, packets with DF: 0, packets reassembled: 0,
Fragments timed out: 0
Are the similar machines at another location using this same FOG server? I’m trying to track down the differences between the two groups and why the speeds would be so different.
Yes, FOG is alive and kicking.
Fog is idle. attempted to image a hp 6000 sff and it is getting 18mib. Something is wrong.
Network is idle.
Nic on clients is intel 82567lm3. Restarting fog services and apache will increase the speed of multicast to about 200mib .
Is fog still alive?
The server has a ton of ram and 2TB raid 5 storage. All machines are brand new. Similar machines at another location get 500/600mib
Can you suggest which kernel is best for the e1000 driver? Intel nic.
multicast will only go as fast as the slowest client. I also think multicast requires much more RAM on the server, so a low RAM server will perform subpar when multicasting vs unicasting.
Using a kernel with best/optimized drivers for the HP 6000 sff may also help, since you do better on another vlan with other machines.