Why does AEX Verbose report only "3376 Bytes of memory used" to support "2 Transmit buffers" with a "Network buffer size of 16384".
Transmit buffers are set on the aex command line, network buffers are set in server.
The transmit buffers are buffers that are actually in the memory on the NIC that hold packets before they are transmitted.,not the computers memory so the change should not impact the amount of computers memory used by the Aex command. This information is in the AEX manual.
Aren't network buffers set in SERVER? Yes
Server network buffers are set by server using values from either the command line or from the net_mgr configuration file. The network buffer size will effect the size reported by server/verbose.
How do AILANBIO Buffers function? This is tempory storage for information moving between redir or server and the adapter cards, this buffer is not normally used except of very large busy systems getting many resource exhuasts. This is also used by applications that bypass the
network operating system and talk directly to the netbios. The buffers specified are actually in addition to the listen and send buffers in Ailanbio that it is not possible to change.
Are two of them always twice as good as one? If needed by system load yes if not needed then no.
Are three of them 50% better than two? See above.
The Upgrade Manual says that if AILANBIO Size is not equal to
Initial_Send_Size, the system will not "function properly"?
The initial send size is the size of the packet sent, the size is the size of the buffer used to build the packet. If the buffer is larger than the packet no problem, if the buffer is smaller you have to use more buffers or the same buffer twice which will increase overhead.
What does that mean? This means that the desired effect of speeding the system will be negated by the additional overhead.
What are the symptoms of not functioning properly? See above.
The Upgrade Manual says that allocating more REDIR Buffers "will not improvethe performance of LANtastic." It also says that "setting larger buffers [Size=] ... should improve performance." Is this an obtuse recommendation for just one very large REDIR buffer?
The manual is in error here, this statement is not true. This depends on the applications you are running, if you are running an application that
uses mulptile small sequential access simultaneously they it will increase network performance.These type of applications include Word processors, spread sheets, etc. As well as some databases (ie dBase scrolling).
How can the WSPERF.SWI example say "REDIR doesn't benefit from buffers" when the SRVPERF.SWI example implies that there should be one buffer for every file"in use at a time by local Workstation"?
If you read the manual these files were designed for a very specific performance where every single call was known. If you plan on running nothing but the PCMAG performance test then you should use these files. If you first analyse the applications you are running, you can use this as an expample to configure the system to a well known apllication.
How does AILANBIO Force_Ack_Mode work? How much overhead does it impose? We are hoping it might help us get past an apparent timeout problem with SDI's PrintQ when running PKZIP on a workstation that is being accessed by a PrintQ writer machine.
Prior to LANtastic 4.0 all netbios packets required a seperate acknowledgement, before the next packet is sent. We have added a new method where the server does NOT wait for the ack before sending the next packet and the acknowledgement is chained to the reply packet being sent (this new method is only used when massive two way transmissions are occuring). Force Ack mode was added for those connections where comunication are very slow and connection is not reliable. This should only be used on a remote station using LantasticZ or LANtasitc Dial Up stationware. DO NOT use it opn any other system. It will cut network speed by over half.
What are the performance implications and buffer requirements of chained sends?
Chained sends are fast, there no buffer requirements, and chained sends should always be used if the netbios will support them. This switch was added to support Foriegn netbios that DO not support chained sends. An exapmle is the PCoffice netbios. This is included in the general manual but only applies to LANtatic/AI.
How many are chained at one time?.
2, this is most effective in writes, rather than a packet saying write this many bytes and then another packet with the data, the request is chained
into one packet. For more informaiton check out any netbios programming manual.
How can we tune SERVER Request_Size and AILANBIO Initial_Send_Size? Is there any diagnostic software that would monitor these events and talk to us about optimium sizes?
Optimum settings unless you are running a netbios application are
14, 60, or 500 bytes in each. 14 handles the very basic requests lock, unlock close a file etc, 60 will handle most requests ie find first , open a file, etc. 500 will handle small writes. If you are using a netbios application that you know exactly how large the requests are you may consider going higher than these values.
What constitutes a SERVER Resource for the Resource_Cache?
A resource is those resources you create in net_mgr, shared RESOURCES.
Is it each file in use by each user? No see above.
Is it each file in use by any user? No see above.
How can we determine the high-water mark of resources used in order to set this parameter?
By anylysing system use. When server is starter there are no resources cached. As each workstation uses a resource it is placed into the cache. If you have 15 resources on a system and only 10 caches set up, when the 11th resource is used, the least frequently used resource is released. If you have a simpel system with 4 resources c-drive, d-drive, @printer and @mail and c ,d, and the printer are used constantly and the mail system used only occationally you can set it to 3. If you have dozens of resources ie wp =>c:\wp51, lotus=>c:\123, account=>c:\sbt, you should determin how ofter each resource is accessed during the day and set the resource accordingly.
How much memory does this parameter consume? 700-800 bytes subject to change with each new release.
Should we set it safely high? it should be no larger thant he total resources.
How exactly does it enhance performance?
Each tiime a user access a resource his privleges have to be checked, this means that if you want to copy a file to c-drive before you start teh copy the network check to see if you have file create permission and write permision. With this information cached the check is very fast if the resource is not cached then the hard drive has to be accessed to check your access permission.
How does the SERVER Seek_Cache work? How big is each chunk of random access location information in it? Is 2K twice as good as 1K?
Seek cache not only stores the location of the file on the disk as fastiopen does (you should not use fastopen with NOS 4.1 or above) but also each users last location in the file.
You can compute the optimum size for this cache with the following formula:
Add the size of any large frequently used database files (in bytes) and devide this by your hard drives cluster size (normally 4096) This gives the number of clusters used. Each cluster used requires 32 bytes , each open file requires 32 bytes. Maximun performance is acheived when the cache size is >=32*(number of clusters used + number of files.)
