List queue tree in Winbox is not always complete. In the lower right corner of the window, for example, “800 out of 2,672 items,” even though it is not included in any display filter.
It helps to close the Winbox several times and log back in, sometimes the complete list of shows up until the third or fourth time
Does not always work in the CLI to add queues - often pops timeout
The same problems were in the RC3.
I have about 2600 queues and up to 300Mbps of traffic.
Shaper is working on MSI server with 3.0GHz PIV D processor but I tested it on SuperMicro with 2x Xeon 3GHz.
Multi-cpu is set to yes.
3.30 works fine but CPU utilization is up to 50% on MSI and 25% in SuperMicro for even 2-3 hour a day and during this time there are a small packet losses. It seems that this is the end capabilities of the hardware, hence the hope of SMP works fine in 5.0.
Why madness?
I have 2600 queues on single shaper, two (download / upload) per customer.
Since when did mangle hashing with jumps to subchains it works without problems.
PCQ is not especially suited because I have dozens of different fares - from 512kbps through 1, 2, 5, 10Mbps to 30, 50 and 100Mbps.
Part is symmetrical, rest has a smaller part of the upload. In addition, manage marks and the queues by using the API directly from the billing system.
Of course, you can choose. Spend thousands of dollars on high tech hardware that can support such configuration, or offer PCQ with less choices for download speeds. Anyway, you are not getting what you think you are getting, the config with 2600 queues is very inefficient. Watch this: http://www.tiktube.com/index.php?video=247
Does this mean that MT can not handle this configuration?
I was hoping that 5.0 with Intel Core 2 Quad and Intel NIC with Receive Side Scaling can handle twice the number of queues and traffic at 900Mbps
normis - you dont understand, question not about 2600 queue (i have shaper with 5120 queue), question about problem with showing a lot of stuff in winbox…
Have the same problem with interfaces or dhcp-leases, winbox not show all items, but in bottom i can see - “173 of 962 items” - it show only 173 item from 962 !!! If i close winbox and open again it might show all items or show 300 or 400 items )) and again if i reopen winbox it show all item.
Reconfigured the setup of my shaper for PCQ queues.
Now I have 20 PCQ queues (10 download and upload) and 20 of the address lists, in which added the IP addresses of customers according to their tariffs.
I’m curious how this configuration will behave in traffic peak.
I have a similar problem.
Very often Tools/Graphing/Interface Graph not displayed Interface (for which statistics are collected).
So , I need to to watch it thru www.
This happens for all versions - 3.x, 4.x, 5.x …
Maybe I just dnot know right way to “refrash” winbox window …
After I swapped the SFQ at PCQ performance of the hardware did not increase.
I have two servers connected at the same time, and I can flip freely move between them. One is a 2x SuperMicro Xeon 3GHz, the other is a Pentium D 3GHz MSI.
In the case of the first large packet loss begins at 70% CPU. Second calmly reaches 100% CPU and still works without noticeable problems.
CPU load looks very similar, both using SFQ and PCQ.
When configured with the SFQ had about 2600 queues. After moving to the PCQ queues have 20, 10 for download and upload, and a little more than 3,000 addresses in address lists (more due to the fact that some customers have 2 IP addresses - for SFQ queues I marked both the same packet mark).
Maybe am I doing something wrong?
So my setup looks like:
In the configuration of the PCQ queues have only two main queues and twenty child queues exactly as pasted in the configuration above.
Previously, the configuration of the queue SFQ main idea lies in the fact that each customer may have a different tariff. For each client, the script generates two rules and two mangle queue. It looks like this:
In total I have about 50 subnets so have about 100 mangle rules: two for each subnet (one in postrouting and one in prerouting).
Then for each IP address have two mark rules and finally two queues for each customer.
BTW: even with the new configuration 5.0RC4 has a problem with displaying long lists. My address-list is not always complete.
Why would I jump all the subnets to the same chain?
The idea is that I have 50 rules in prerouting and 50 in postrouting to jump to the chain in which they are usually mark rules for 32 IP addresess. In this way, the average number of tests is 50 / 2 * 32 / 2 = 400 for each packet traversing the firewall.
If all subnets jumps to the same chain average number of tests would be 50 / 2 * 1600 / 2 = 40000.
That was the configuration of the SFQ.
Now, using the PCQ average number of tests is 10 / 2 = 5 because I have defined 10 tariffs and thus 10 rules in the mangle.
Despite this performance has not increased or even declined.
Besides, from the viewpoint of customers, the situation got worse because, as everyone had their own SFQ was not dependent on other users with the same tariff as it is now. As the user does not generate traffic on the sea for their maximum tariffs do not have any loss or delay packets. If he generate it it was his responsibility.
In conclusion I am not convinced by the superiority of the PCQ in relation to the SFQ and still do not know why my configuration was wrong and PCQ is ok, since my work better.
Maybe it turns out that something’s wrong in my PCQ configuration.
a) in your mangle you have 2 rules per packet mark - i suggest to go for 3
mark connection if src-address-list=x
mark-connection if dst-address-list=x
mark-packet for this connection mark
b) think carefully about situations like this - 100D client are downloading from 05U client, what should be speed of this connection - 0,5 or 100M??? Choose one and in mangle place corresponding rules above others
c) PCQ finally works with subqueue rates that are higher than 10Mbps.
What’s new in 5.0rc4 (2010-Nov-22 16:58):
*) fixed pcq queue type;
So without any hesitation i would go for an upgrade
d) if it is necessary it is possible to distribute traffic between customer classes based on limit-at and max-limit of queue
My shaper is not router, so there is not traffic between customers. There is only traffic form/to internet.
Shaper works in bridge mode and is positioned at link between border router (Cisco with BPG) and access router (Layer 3 switch).
c) PCQ finally works with subqueue rates that are higher than 10Mbps.
What’s new in 5.0rc4 (2010-Nov-22 16:58):
*) fixed pcq queue type;
So without any hesitation i would go for an upgrade
