Request for Comments: 2562 IBM Corp.
Category: Standards Track R. Moore
IBM Corp.
April 1999
Definitions of Protocol and Managed Objects for
TN3270E Response Time Collection Using SMIv2
(TN3270E-RT-MIB)
Status of this Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (1999). All Rights Reserved.
Abstract
This memo defines the protocol and the Management Information Base
(MIB) for performing response time data collection on TN3270 and
TN3270E sessions by a TN3270E server. The response time data
collected by a TN3270E server is structured to support both
validation of service level agreements and performance monitoring of
TN3270 and TN3270E Sessions. This MIB has as a prerequisite the
TN3270E-MIB, reference [20].
TN3270E, defined by RFC2355 [19], refers to the enhancements made to
the Telnet 3270 (TN3270) terminal emulation practices. Refer to RFC
1041 [18], STD 8, RFC854 [16], and STD 31, RFC860 [17] for a sample
of what is meant by TN3270 practices.
Table of Contents
1.0 Introduction . . . . . . . . . . . . . . . . . . . . . . . 2
2.0 The SNMP Network Management Framework . . . . . . . . . . 2
3.0 Response Time Collection Methodology . . . . . . . . . . . 3
3.1 General Response Time Collection . . . . . . . . . . . . . 3
3.2 TN3270E Server Response Time Collection . . . . . . . . . 5
3.3 Correlating TN3270E Server and Host Response Times . . . . 10
3.4 Timestamp Calculation . . . . . . . . . . . . . . . . . . 11
3.4.1 DR Usage . . . . . . . . . . . . . . . . . . . . . . . 12
3.4.2 TIMING-MARK Usage . . . . . . . . . . . . . . . . . . 13
3.5 Performance Data Modelling . . . . . . . . . . . . . . . . 15
3.5.1 Averaging Response Times . . . . . . . . . . . . . . . 15
3.5.2 Response Time Buckets . . . . . . . . . . . . . . . . 18
4.0 Structure of the MIB . . . . . . . . . . . . . . . . . . . 19
4.1 tn3270eRtCollCtlTable . . . . . . . . . . . . . . . . . . 19
4.2 tn3270eRtDataTable . . . . . . . . . . . . . . . . . . . . 23
4.3 Notifications . . . . . . . . . . . . . . . . . . . . . . 24
4.4 Advisory Spin Lock Usage . . . . . . . . . . . . . . . . . 26
5.0 Definitions . . . . . . . . . . . . . . . . . . . . . . . 26
6.0 Security Considerations . . . . . . . . . . . . . . . . . 45
7.0 Intellectual Property . . . . . . . . . . . . . . . . . . 45
8.0 Acknowledgments . . . . . . . . . . . . . . . . . . . . . 46
9.0 References . . . . . . . . . . . . . . . . . . . . . . . . 46
10.0 Authors' Addresses . . . . . . . . . . . . . . . . . . . 48
11.0 Full Copyright Statement . . . . . . . . . . . . . . . . 49
1.0 Introduction
This document is a product of the TN3270E Working Group. It defines
a protocol and a MIB module to enable a TN3270E server to collect and
keep track of response time data for both TN3270 and TN3270E clients.
Basis for implementing this MIB:
o TN3270E-MIB, Base Definitions of Managed Objects for TN3270E
Using SMIv2 [20]
o TN3270E RFCs
o Telnet Timing Mark Option RFC[17].
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC2119, reference
[23].
2.0 The SNMP Network Management Framework
The SNMP Management Framework presently consists of five major
components:
o An overall architecture, described in RFC2271 [1].
o Mechanisms for describing and naming objects and events for the
purpose of management. The first version of this Structure of
Management Information (SMI) is called SMIv1 and described in STD
16, RFC1155 [2], STD 16, RFC1212 [3] and RFC1215 [4]. The
second version, called SMIv2, is described in RFC1902 [5], RFC
1903 [6] and RFC1904 [7].
o Message protocols for transferring management information. The
first version of the SNMP message protocol is called SNMPv1 and
described in STD 15, RFC1157 [8]. A second version of the SNMP
message protocol, which is not an Internet standards track
protocol, is called SNMPv2c and described in RFC1901 [9] and RFC
1906 [10]. The third version of the message protocol is called
SNMPv3 and described in RFC1906 [10], RFC2272 [11] and RFC2274
[12].
o Protocol operations for accessing management information. The
first set of protocol operations and associated PDU formats is
described in STD 15, RFC1157 [8]. A second set of protocol
operations and associated PDU formats is described in RFC1905
[13].
o A set of fundamental applications described in RFC2273 [14] and
the view-based access control mechanism described in RFC2275
[15].
Managed objects are accessed via a virtual information store, termed
the Management Information Base or MIB. Objects in the MIB are
defined using the mechanisms defined in the SMI.
This memo specifies a MIB module that is compliant to the SMIv2. A
MIB conforming to the SMIv1 can be produced through the appropriate
translations. The resulting translated MIB must be semantically
equivalent, except where objects or events are omitted because no
translation is possible (use of Counter64). Some machine readable
information in SMIv2 will be converted into textual descriptions in
SMIv1 during the translation process. However, this loss of machine
readable information is not considered to change the semantics of the
MIB.
3.0 Response Time Collection Methodology
This section explains the methodology and approach used by the MIB
defined by this memo for response time data collection by a TN3270E
server.
3.1 General Response Time Collection
Two primary methods exist for measuring response times in SNA
networks:
o The Systems Network Architecture Management Services (SNA/MS)
Response Time Monitoring (RTM) function.
o Timestamping using definite response flows.
This memo defines an approach using definite responses to timestamp
the flows between a client and its TN3270E server, rather than by use
of the RTM method. Extensions to the SNA/MS RTM flow were considered,
but this approach was deemed unsuitable since not all TN3270E server
implementations have access to their underlying SNA stacks. The RTM
concepts of keeping response time buckets for service level
agreements and of interval-based response time collection for
performance monitoring are preserved in the MIB module defined in
this memo.
As mentioned, this memo focuses on using definite responses to
timestamp the flows between a client and its TN3270E server for
generating performance data. Use of a definite response flow
requires that the client supports TN3270E with the RESPONSES function
negotiated. The TN3270 TIMING-MARK option can be used instead of
definite response for supporting TN3270 clients or TN3270E clients
that don't support RESPONSES. This document focuses first on
defining the protocol and methods for generating performance data
using definite responses, and then describes how the TIMING-MARK
option can be used instead of definite response.
In an SNA network, a transaction between a client Logical Unit (LU)
and a target host in general looks as follows:
------------------------------------------------
| |
| Client LU Target SNA Host |
| |
| Timestamps |
| request A |
| -----------------------------------------> |
| reply(DR) B | |
| <---------------------------------------< |
| | +/-RSP C |
| >---------------------------------------> |
| |
| DR: Definite Response requested |
| +/-RSP: Definite Response |
| |
------------------------------------------------
This transaction is a simple one, and is being used only to
illustrate how timestamping at a target SNA host can be used to
generate response times. An IBM redbook [12] provides a more
detailed description of response time collection for a transaction of
this type. Note that for the purpose of calculating an approximation
for network transit time, it doesn't matter if the response is
positive or negative. Two response time values are typically
calculated:
o Host Transit Time: Timestamp B - Timestamp A
o Network Transit Time: Timestamp C - Timestamp B
Network transit time is an approximation for the amount of time that
a transaction requires to flow across a network, since the response
flow is being substituted for the request flow at the start of the
transaction. Network transit time, timestamp C - timestamp B, is the
amount of time that the definite response request and its response
required. Host time, timestamp B - timestamp A, is the actual time
that the host required to process the transaction. Experience has
shown that using the response flow to approximate network transit
times is useful, and does correlate well with actual network transit
times.
A client SHOULD respond to a definite response request when it
completes processing the transaction. This is important since it
increases the accuracy of a total response time. Clients that
immediately respond to a definite response request will be attributed
with lower total response times then those that actually occurred.
The TN3270E-RT-MIB describes a method of collecting performance data
that is not appropriate for printer (LU Type 1 or LU Type 3)
sessions; thus collection of performance data for printer sessions is
excluded from this MIB. This exclusion of printer sessions is not
considered a problem, since these sessions are not the most important
ones for response time monitoring, and since historically they were
excluded from SNA/MS RTM collection. The tn3270eTcpConnResourceType
object in a tn3270eTcpConnEntry (in the TN3270E-MIB) can be examined
to determine if a client session is ineligible for response time data
collection for this reason.
3.2 TN3270E Server Response Time Collection
A TN3270E server connects a Telnet client performing 3270 emulation
to a target SNA host over both a client-side network (client to
TN3270E server) and an SNA Network (TN3270E server to target SNA
host). The client-side network is typically TCP/IP, but it need not
be. For ease of exposition this document uses the term "IP network"
to refer to the client-side network, since IP is by far the most
common protocol for these networks.
A TN3270E server can use SNA definite responses and the TN3270
Enhancement (RFC2355 [19]) RESPONSES function to calculate response
times for a transaction, by timestamping when a client request
arrives at the server, when the reply arrives from the target host,
and when the response acknowledging this reply arrives from the
client.
Section 3.4, Timestamp Calculation, provides specifics on when in the
sequence of flows between a TN3270E client and its target SNA host a
TN3270E server takes the required timestamps. In addition, it
provides information on how a TN3270 TIMING-MARK request/response
flow can be used instead of DR for approximating IP network transit
times.
The following figure adds a TN3270E server between the client, in
this case a TN3270E client and the target SNA host:
------------------------------------------------
| |
| Client TN3270E Target |
| Server SNA Host |
| Timestamps |
| |
| <---IP Network-------><---SNA Network---> |
| |
| request D |
| ------------------------------------------> |
| reply(DR) E | |
| <----------------------------------------< |
| | +/-RSP F |
| >-------------------- - - - - - - - - - > |
| |
------------------------------------------------
A TN3270E server can save timestamp D when it receives a client
request, save timestamp E when the target SNA host replies, and save
timestamp F when the client responds to the definite response request
that flowed with the reply. It doesn't matter whether the target SNA
host requested a definite response on its reply: if it didn't, the
TN3270E server makes the request on its own, to enable it to produce
timestamp F. In this case the TN3270E server does not forward the
response to the target SNA host, as the dotted line in the figure
indicates.
Because it is a special case, a transaction in which a target SNA
host returns an UNBIND in response to a client's request, and the
TN3270E server forwards the UNBIND to the client, is not included in
any response time calculations.
In order to generate timestamp F, a TN3270E server MUST insure that
the transaction specifies DR, and that the TN3270E RESPONSES function
has been negotiated between itself and the client. Negotiation of
the TN3270E RESPONSES function occurs during the client's TN3270E
session initialization. The TN3270E servers that the authors are
aware of do request the RESPONSES function during client session
initialization. TN3270E clients either automatically support the
RESPONSES function, or can be configured during startup to support
it.
Using timestamps D, E, and F the following response times can be
calculated by a TN3270E server:
o Total Response time: Timestamp F - Timestamp D
o IP Network Transit Time: Timestamp F - Timestamp E
Just as in the SNA case presented above, these response times are
also approximations, since the final +/- RSP from the client is being
substituted for the request from the client that began the
transaction.
The MIB provides an object, tn3270eRtCollCtlType, to control several
aspects of response time data collection. One of the available
options in setting up a response time collection policy is to
eliminate the IP-network component altogether. This might be done
because it is determined either that the additional IP network
traffic would not be desirable, or that the IP-network component of
the overall response times is not significant.
Excluding the IP-network component from response times also has an
implication for the way in which response time data is aggregated. A
TN3270E server may find that some of its clients simply don't support
any of the functions necessary for the server to calculate the IP-
network component of response times. For these clients, the most
that the server can calculate is the SNA-network component of their
overall response times; the server records this SNA-network component
as the TOTAL response time each of these clients' transactions. If a
response time collection is aggregating data from a number of
clients, some of which have the support necessary for including the
IP-network component in their total response time calculations, and
some of which do not, then the server aggregates the data differently
depending on whether the collection has been defined to include or
exclude the IP-network component:
o If the IP-network component is included, then transactions for the
clients that don't support calculation of the IP-network component
of their response times are excluded from the aggregation
altogether.
o If the IP-network component is excluded, then total response times
for ALL clients include only the SNA-network component, even
though the server could have included an IP-network component in
the overall response times for some of these clients. The server
does this by setting timestamp F, which marks the end of a
transaction's total response time, equal to timestamp E, the end
of the transaction's SNA-network component.
The principle here is that all the transactions contributing their
response times to an aggregated value MUST make the same
contribution. If the aggregation specifies that an IP-network
component MUST be included in the aggregation's response times, then
transactions for which an IP-network component cannot be calculated
aren't included at all. If the aggregation specifies that an IP-
network component is not to be included, then only the SNA-network
component is used, even for those transactions for which an IP-
network component could have been calculated.
There is one more complication here: the MIB allows a management
application to enable or disable dynamic definite responses for a
response time collection. Once again the purpose of this option is
to give the network operator control over the amount of traffic
introduced into the IP network for response time data collection. A
DYNAMIC definite response is one that the TN3270E server itself adds
to a reply, in a transaction for which the SNA application at the
target SNA host did not specify DR in its reply. When the +/-RSP
comes back from the client, the server uses this response to
calculate timestamp F, but then it does not forward the response on
to the SNA application (since the application is not expecting a
response to its reply).
The dynamic definite responses option is related to the option of
including or excluding the IP-network component of response times
(discussed above) as follows:
o If the IP-network component is excluded, then there is no reason
for enabling dynamic definite responses: the server always sets
timestamp F equal to timestamp E, so the additional IP-network
traffic elicited by a dynamic definite response would serve no
purpose.
o If the IP-network component is included, then enabling dynamic
definite responses causes MORE transactions to be included in the
aggregated response time values:
- For clients that do not support sending of responses, timestamp
F can never be calculated, and so their transactions are never
included in the aggregate.
- For clients that support sending of responses, timestamp F will
always be calculated for transactions in which the host SNA
application specifies DR in its reply, and so these
transactions will always be included in the aggregate.
- For clients that support sending of responses, having dynamic
definite responses enabled for a collection results in the
inclusion of additional transactions in the aggregate:
specifically, those for which the host SNA application did not
specify DR in its reply.
A TN3270E server also has the option of substituting TIMING-MARK
processing for definite responses in calculating the IP-network
component of a transaction's response time. Once again, there is no
reason for the server to do this if the collection has been set up to
exclude the IP-network component altogether in computing response
times.
The MIB is structured to keep counts and averages for total response
times (F - D) and their IP-network components (F - E). A management
application can obviously calculate from these two values an average
SNA-network component (E - D) for the response times. This SNA-
network component includes the SNA node processing time at both the
TN3270E server and at the target application.
A host TN3270E server refers to an implementation where the TN3270E
server is collocated with the Systems Network Architecture (SNA)
System Services Control Point (SSCP) for the dependent Secondary
Logical Units (SLUs) that the server makes available to its clients
for connecting into an SNA network. A gateway TN3270E server resides
on an SNA node other than an SSCP, either an SNA type 2.0 node, a
boundary-function-attached type 2.1 node, or an APPN node acting in
the role of a Dependent LU Requester (DLUR). Host and gateway
TN3270E server implementations typically differ greatly as to their
internal implementation and System Definition (SYSDEF) requirements.
If a host TN3270E server is in the same SNA host as the target
application, then the SNA-network component of a transaction's
response time will approximately equal the host transit time (B - A)
described previously. A host TN3270E server implementation can,
however, typically support the establishment of sessions to target
applications in SNA hosts remote from itself. In this case the SNA-
network component of the response time equals the actual SNA-network
transit time plus two host transit times.
3.3 Correlating TN3270E Server and Host Response Times
It is possible that response time data is collected from TN3270E
servers at the same time as a management application is monitoring
the SNA sessions at a host. For example, a management application
can be monitoring a secondary logical unit (SLU) while retrieving
data from a TN3270E server. Consider the following figure:
------------------------------------------------
| |
| Client TN3270E Target |
| Server SNA Host |
| Timestamps (PLU) |
| (SLU) Timestamps|
| <---IP Network-------><---SNA Network---> |
| |
| request D A |
| ------------------------------------------> |
| reply(DR) E B | |
| <----------------------------------------< |
| | +/-RSP F C |
| >--------------------------------------> |
| |
------------------------------------------------
The following response times are available:
o Target SNA host transit time: Timestamp B - Timestamp A
o Target SNA host network transit time: Timestamp C - Timestamp B
o TN3270E server total response time: Timestamp F - Timestamp D
o TN3270E server IP-network component: Timestamp F - Timestamp E
The value added by the TN3270E server in this situation is its
approximation of the IP-network component of the overall response
time. The IP-network component can be subtracted from the total
network transit time (which can be captured at an SSCP monitoring SNA
traffic from/to the SLU) to see the actual SNA versus IP network
transit times.
The MIB defined by this memo does not specifically address
correlation of the data it contains with response time data collected
by direct monitoring of SNA resources: its focus is exclusively
response time data collection from a TN3270E server perspective. It
has, however, in conjunction with the TN3270E-MIB [10], been
structured to provide the information necessary for correlation
between TN3270E server-provided response time information and that
gathered from directly monitoring SNA resources.
A management application attempting to correlate SNA resource usage
to Telnet clients can monitor either the tn3270eResMapTable or the
tn3270eTcpConnTable to determine resource-to-client address mappings.
Both of these tables are defined by the TN3270E-MIB [10]. Another
helpful table is the tn3270eSnaMapTable, which provides a mapping
between SLU names as they are known at the SSCP (VTAM) and their
local names at the TN3270E server. Neither the
tn3270eClientGroupTable, the tn3270eResPoolTable, nor the
tn3270eClientResMapTable from the TN3270E-MIB can be used for
correlation, since the mappings defined by these tables can overlap,
and may not provide one-to-one mappings.
3.4 Timestamp Calculation
This section goes into more detail concerning when the various
timestamps can be taken as the flows between a TN3270E client and its
target SNA host pass through a TN3270E server. In addition,
information is provided on how the TN3270 TIMING-MARK
request/response flow can be used in place of DR for approximating IP
network transit times.
3.4.1 DR Usage
Consider the following flow:
----------------------------------------------------------
| |
| Client TN3270E Target SNA |
| Server Host |
| Timestamps |
| |
| <---IP Network-------><---SNA Network---> |
| |
| request D (BB,CD,OIC,ER) |
| -------------------------------------------> |
| reply(DR) (FIC,ER,EB) | |
| <-----------------------------------------< |
| reply (MIC,ER) |
| <-----------------------------------------< |
| reply (MIC,ER) |
| <-----------------------------------------< |
| reply E (LIC,DR) |
| <-----------------------------------------< |
| | +/-RSP F |
| >----------------------------------------> |
| |
| BB : Begin Bracket ER : Response by exception |
| EB : End Bracket DR : Definite Response Requested |
| CD : Change Direction FIC : First in chain |
| OIC: Only in chain MIC: Middle in chain |
| LIC: Last in chain |
----------------------------------------------------------
Timestamp D is taken at the TN3270E server when the server has
received data from a client for forwarding to its target SNA host,
and the direction of the SNA session allows the server to forward the
data immediately (either the direction is inbound towards the SNA
host, or the session is between brackets). This is most likely when
the server finds the end of record indicator in the TCP data received
from the client.
The target SNA application returns its reply in one or more SNA
Request Units (RUs); in this example there are four RUs in the reply.
The first RU is marked as first in chain (FIC), the next two are
marked as middle in chain (MIC), and the last is marked as last in
chain (LIC). If the SNA host sends a multiple-RU chain, the server
does not know until the last RU is received whether DR is being
requested. The server's only chance to request DR from the client,
however, comes when it forwards the FIC RU, since this is the only
time that the TN3270E header is included. Since a server may forward
the FIC RU to the client before it receives the LIC RU from the SNA
host, some servers routinely specify DR on all FIC RUs.
If the server has specified DR on the TN3270E request for the FIC RU
in a chain, it takes timestamp E when it forwards the LIC RU to the
client. Since timestamp E is used for calculating the IP-network
time for the transaction, the server SHOULD take timestamp E as close
as possible to its "Telnet edge". The server takes timestamp F when
it receives the RESPONSES response from the client.
A target SNA application doesn't necessarily return data to a client
in a transaction; it may, for example, require more data from the
client before it can formulate a reply. In this case the application
may simply return to the TN3270E server a change of direction
indicator. At this point the server must send something to the
client (typically a Write operation with a WCC) to unlock the
keyboard. If the server specifies DR on the request to the client
triggered by its receipt of the change of direction indicator from
the SNA application, then timestamps E and F can be taken, and the
usual response times can be calculated. When the client sends in the
additional data and gets a textual response from the SNA application,
the server treats this as a separate transaction from the one
involving the change of direction.
3.4.2 TIMING-MARK Usage
It is possible for a TN3270E server to use the TIMING-MARK flow for
approximating IP network transit times. Using TIMING-MARKs would
make it possible for a server to collect performance data for TN3270
clients, as well as for TN3270E clients that do not support the
RESPONSES function. In order for TIMING-MARKs to be used in this
way, a client can't have the NOP option enabled, since responses are
needed to the server's TIMING-MARK requests. An IP network transit
time approximation using a TIMING-MARK is basically the amount of
time it takes for a TN3270 server to receive from a client a response
to a TIMING-MARK request.
To get an estimate for IP network transit time, a TN3270E server
sends a TIMING-MARK request to a client after a LIC RU has been
received, as a means of approximating IP network transit time:
---------------------------------------------------
| |
| Client TN3270E Target |
| Server Host |
| Timestamps |
| |
| <---IP Network-------><---SNA Network---> |
| |
| request D (BB,CD,OIC,ER) |
| -------------------------------------------> |
| reply (FIC,ER,EB) | |
| <-----------------------------------------< |
| reply (MIC,ER) |
| <-----------------------------------------< |
| reply (MIC,ER) |
| <-----------------------------------------< |
| reply E (LIC,ER) |
| <-----------------------------------------< |
| TIMING-MARK Rqst E' |
| <--------------------- |
| | TIMING-MARK Rsp F' |
| >-------------------> |
| |
---------------------------------------------------
The response times can then be calculated as follows:
o TN3270E server total response time:
(Timestamp E - Timestamp D) + (Timestamp F' - Timestamp E')
o TN3270E server IP network time: Timestamp F' - Timestamp E'
If a TN3270E server is performing the TIMING-MARK function
(independent of the response time monitoring use of the function
discussed here), then it most likely has a TIMING-MARK interval for
determining when to examine client sessions for sending the TIMING-
MARK request. This interval, which is ordinarily a global value for
an entire TN3270E server, is represented in the TN3270E-MIB by the
tn3270eSrvrConfTmNopInterval object. A TIMING-MARK request is sent
only if, when it is examined, a client session is found to have had
no activity for a different fixed length of time, represented in the
TN3270E-MIB by the tn3270eSrvrConfTmNopInactTime object.
Servers that support a large number of client sessions should spread
out the TIMING-MARK requests they send to these clients over the
activity interval, rather than sending them all in a single burst,
since otherwise the network may be flooded with TIMING-MARK requests.
When a server uses TIMING-MARKs for approximating response times,
this tends to introduce a natural spreading into its TIMING-MARK
requests, since the requests are triggered by the arrival of traffic
from an SNA host.
A TN3270E server MUST integrate its normal TIMING-MARK processing
with its use of TIMING-MARKs for computing response times. In
particular, it MUST NOT send a second TIMING-MARK request to a client
while waiting for the first to return, since this is ruled out by the
TIMING-MARK protocol itself. If a TIMING-MARK flow has just been
performed for a client shortly before the LIC RU arrives, the server
MAY use the interval from this flow as its approximation for IP
network transit time, (in other words, as its (F' - E') value) when
calculating its approximation for the transaction's total response
time, rather than sending a second TIMING-MARK request so soon after
the preceding one.
Regardless of when the server sends its TIMING-MARK request, the
accuracy of its total response time calculation depends on exactly
when the client responds to the TIMING-MARK request.
3.5 Performance Data Modelling
The following two subsections detail how the TN3270E-RT-MIB models
and controls capture of two types of response time data: average
response times and response time buckets.
3.5.1 Averaging Response Times
Average response times play two different roles in the MIB:
o They are made available for management applications to retrieve.
o They serve as triggers for emitting notifications.
Sliding-window averages are used rather than straight interval-based
averages, because they are often more meaningful, and because they
cause less notification thrashing. Sliding-window average
calculation can, if necessary, be disabled, by setting the sample
period multiplier, tn3270eRtCollCtlSPMult, to 1, and setting the
sample period, tn3270eRtCollCtlSPeriod, to the required collection
interval.
In order to calculate sliding-window averages, a TN3270E server MUST:
o Select a fixed, relatively short, sample period SPeriod; the
default value for SPeriod in the MIB is 20 seconds.
o Select an averaging period multiplier SPMult. The actual
collection interval will then be SPMult times SPeriod. The
default value for SPMult in the MIB is 30, yielding a default
collection interval of 10 minutes. Note that the collection
interval (SPMult*SPeriod) is always a multiple of the sample
period.
Clearlly, SPMult*SPeriod should not be thought of as literally
the averaging period. The average calculated will include
contributions older than that time, and does not weight equally
all contributions since that time. In fact, it gives a smoother
result than a traditional sliding average, as used in finance.
More subtly, it is best to think of the effective averaging
period as being 2*SPMult*SPeriod. To see this, consider how long
the contribution to the result made by a particular transaction
lasts. With a traditional sliding average, it lasts exactly the
averaging period. With the aging mechanism described here, it
has a half-life of SPMult*SPeriod.
o Maintain the following counters to keep track of activity within
the current sample period; these are internal counters, not made
visible to a management application via the MIB.
- T (number of transactions in the period)
- TotalRts (sum of the total response times for all
transactions in the period)
- TotalIpRts (sum of the IP network transit times for all
transactions in the period; note that if IP network transit
times are being excluded from the response time collection,
this value will always be 0).
o Also maintain sliding counters, initialized to zero, for each of
the quantities being counted:
- AvgCountTrans (sliding count of transactions)
- TotalRtsSliding (sliding count of total response times)
- TotalIpRtsSliding (sliding count of IP network transit times)
o At the end of each sample period, update the sliding interval
counters, using the following floating-point calculations:
AvgCountTrans = AvgCountTrans + T
- (AvgCountTrans / SPMult)
TotalRtsSliding = TotalRtsSliding + TotalRts
- (TotalRtsSliding / SPMult)
TotalIpRtsSliding = TotalIpRtsSliding + TotalIpRts
- (TotalIpRtsSliding / SPMult)
Then reset T, TotalRts, and TotalIpRts to zero for use during the
next sample period.
o At the end of a collection interval, update the following MIB
objects as indicated; the floating-point numbers are rounded
rather than truncated.
tn3270eRtDataAvgCountTrans = AvgCountTrans
tn3270eRtDataAvgRt = TotalRtsSliding / AvgCountTrans
tn3270eRtDataAvgIpRt = TotalIpRtsSliding / AvgCountTrans
As expected, if IP network transit times are being excluded from
response time collection, then tn3270eRtDataAvgIpRt will always
return 0.
The sliding transaction counter AvgCountTrans is not used for
updating the MIB object tn3270eRtDataCountTrans: this object is an
ordinary SMI Counter32, which maintains a total count of transactions
since its last discontinuity event. The sliding counters are used
only for calculating averages.
Two mechanisms are present in the MIB to inhibit the generation of an
excessive number of notifications related to average response times.
First, there are high and low thresholds for average response times.
A tn3270eRtExceeded notification is generated the first time a
statistically significant average response time is found to have
exceeded the high threshold. (The test for statistical significance
is described below.) After this, no other tn3270eRtExceeded
notifications are generated until an average response time is found
to have fallen below the low threshold.
The other mechanism to limit notifications is the significance test
for a high average response time. Intuitively, the significance of
an average is directly related to the number of samples that go into
it; so we might be inclined to use a rule such as "for the purpose of
generating tn3270eRtExceeded notifications, ignore average response
times based on fewer than 20 transactions in the sample period."
In the case of response times, however, the number of transactions
sampled in a fixed sampling period is tied to these transactions'
response times. A few transactions with long response times can
guarantee that there will not be many transactions in a sample,
because these transactions "use up" the sampling time. Yet this case
of a few transactions with very poor response times should obviously
be classified as a problem, not as a statistical anomaly based on too
small a sample.
The solution is to make the significance level for a sample a
function of the average response time. A value IdleCount is
specified, which is used to qualify an sample as statistically
significant. In order to determine at a collection interval whether
to generate a tn3270eRtExceeded notification, a TN3270E server uses
the following algorithm:
if AvgCountTrans * ((AvgRt/ThreshHigh - 1) ** 2) >= IdleCount
then generate the notification,
where AvgRt is the value that would be returned by the object
tn3270eRtDataAvgRt at the end of the interval, and the "**" notation
indicates exponientiation.
Two examples illustrate how this algorithm works. Suppose that
IdleCount has been set to 20 transactions, and the high threshold to
200 msecs per transaction. If the average observed response time is
300 msecs, then a notification will be generated only if
AvgCountTrans >= 80. If, however, the observed response time is 500
msecs, then a notification is generated if AvgCountTrans >= 9.
There is no corresponding significance test for the tn3270eRtOkay
notification: this notification is generated based on an average
response time that falls below the low threshold, regardless of the
sample size behind that average.
3.5.2 Response Time Buckets
The MIB also supports collection of response time data into a set of
five buckets. This data is suitable either for verification of
service level agreements, or for monitoring by a management
application to identify performance problems. The buckets provide
counts of transactions whose total response times fall into a set of
specified ranges.
Like everything for a collection, the "total" response times
collected in the buckets are governed by the specification of whether
IP network transit times are to be included in the totals. Depending
on how this option is specified, the response times being counted in
the buckets will either be total response times (F - D), or only SNA
network transit times (effectively E - D, because when it is
excluding the IP-network component of transactions, a server makes
timestamp F identical to timestamp E).
Four bucket boundaries are specified for a response time collection,
resulting in five buckets. The first response time bucket counts
those transactions whose total response times were less than or equal
to Boundary 1, the second bucket counts those whose response times
were greater than Boundary 1 but less than or equal to Boundary 2,
and so on. The fifth bucket is unbounded on the top, counting all
transactions whose response times were greater than Boundary 4.
The four bucket boundaries have default values of: 1 second, 2
seconds, 5 seconds, and 10 seconds, respectively. These values are
the defaults in the 3174 controller's implementation of the SNA/MS
RTM function, and are thought to be appropriate for this MIB as well.
In SNA/MS the counter buckets were (by today's standards) relatively
small, with a maximum value of 65,535. The bucket objects in the MIB
are all Counter32's.
The following figure represents the buckets pictorially:
----------------------------------------------
| |
| Response Time Boundaries |
| | | | | | | |
| | | | | | | |
| | | | | | no |
| 0 B-1 B-2 B-3 B-4 bound|
| | | | | | | |
| |Bucket1|Bucket2|Bucket3|Bucket4|Bucket5| |
| ----------------------------------------- |
| |
----------------------------------------------
4.0 Structure of the MIB
The TN3270E-RT-MIB has the following components:
o tn3270eRtCollCtlTable
o tn3270eRtDataTable
o Notifications
o Advisory Spin Lock Usage
4.1 tn3270eRtCollCtlTable
The tn3270eRtCollCtlTable is indexed by tn3270eSrvrConfIndex and
tn3270eClientGroupName imported from the TN3270E-MIB.
tn3270eSrvrConfIndex identifies within a host a particular TN3270E
server. tn3270eClientGroupName identifies a collection of IP clients
for which response time data is to be collected. The set of clients
is defined using the tn3270eClientGroupTable from the TN3270E-MIB.
A tn3270eRtCollCtlEntry contains the following objects:
--------------------------------------------------
1st Index | tn3270eSrvrConfIndex Unsigned32 |
2nd Index | tn3270eClientGroupName Utf8String |
| tn3270eRtCollCtlType BITS |
| tn3270eRtCollCtlSPeriod Unsigned32 |
| tn3270eRtCollCtlSPMult Unsigned32 |
| tn3270eRtCollCtlThreshHigh Unsigned32 |
| tn3270eRtCollCtlThreshLow Unsigned32 |
| tn3270eRtCollCtlIdleCount Unsigned32 |
| tn3270eRtCollCtlBucketBndry1 Unsigned32 |
| tn3270eRtCollCtlBucketBndry2 Unsigned32 |
| tn3270eRtCollCtlBucketBndry3 Unsigned32 |
| tn3270eRtCollCtlBucketBndry4 Unsigned32 |
| tn3270eRtCollCtlRowStatus RowStatus |
--------------------------------------------------
The tn3270eRtCollCtlType object controls the type(s) of response time
collection that occur, the granularity of the collection, whether
dynamic definite responses SHOULD be initiated, and whether
notifications SHOULD be generated. This object is of BITS SYNTAX,
and thus allows selection of multiple options.
The BITS in the tn3270eRtCollCtlType object have the following
meanings:
o aggregate(0) - If this bit is set to 1, then data SHOULD be
aggregated for the whole client group. In this case there will
be only one row created for the collection in the
tn3270eRtDataTable. The first two indexes for this row,
tn3270eSrvrConfIndex and tn3270eClientGroupName, will have the
same values as the indexes for the corresponding
tn3270eRtCollCtlEntry. The third and fourth indexes of an
aggregated tn3270eRtDataEntry have the values unknown(0)
(tn3270eRtDataClientAddrType) and a zero-length octet string
(tn3270eRtDataClientAddress). The fifth index,
tn3270eRtDataClientPort, has the value 0.
If this bit is set to 0, then a separate entry is created in the
tn3270eRtDataTable from each member of the client group. In this
case tn3270eRtDataClientAddress contains the client's actual IP
Address, tn3270eRtDataClientAddrType indicates the address type,
and tn3270eRtDataClientPort contains the number of the port the
client is using for its TN3270/TN3270E session.
o excludeIpComponent(1) - If this bit is set to 1, then the server
SHOULD exclude the IP-network component from all the response
times for this collection. If the target SNA application
specifies DR in any of its replies, this DR will still be passed
down to the client, and the client's response will still be
forwarded to the application. But this response will play no
role in the server's response time calculations.
If this bit is set to 0, then the server includes in the
collection only those transactions for which it can include an
(approximate) IP-network component in the total response time for
the transaction. This component MAY be derived from a "natural"
DR (if the client supports the RESPONSES function), from a
dynamic DR introduced by the server (if the client supports the
RESPONSES function and the ddr(2) bit has been set to 1), or from
TIMING-MARK processing (if the client supports TIMING-MARKs).
If this bit is set to 1, then the ddr(2) bit is ignored, since
there is no reason for the server to request additional responses
from the client(s) in the group.
o ddr(2) - If this bit is set to 1, then the server SHOULD, for
those clients in the group that support the RESPONSES function,
add a DR request to the FIC reply in each transaction, and use
the client's subsequent response for calculating an (approximate)
IP-network component to include in the transaction's total
response times.
If this bit is set to 0, then the server does not add a DR
request that it was not otherwise going to add to any replies
from the target SNA application.
If the excludeIpComponent(1) bit is set to 1, then this bit is
ignored by the server.
o average(3) - If this bit is set to 1, then the server SHOULD
calculate a sliding-window average for the collection, based on
the parameters specified for the group.
If this bit is set to 0, then an average is not calculated. In
this case the tn3270eRtExceeded and tn3270eRtOkay notifications
are not generated, even if the traps(5) bit is set to 1.
o buckets(4) - If this bit is set to 1, then the server SHOULD
create and increment response time buckets for the collection,
based on the parameters specified for the group.
If this bit is set to 0, then response time buckets are not
created.
o traps(5) - If this bit is set to 1, then a TN3270E Server is
enabled to generate notifications pertaining to an
tn3270eCollCtlEntry. tn3270CollStart and tn3270CollEnd
generation is enabled simply by traps(5) being set to 1.
tn3270eRtExceeded and tn3270eRtOkay generation enablement
requires that average(3) be set to 1 in addition to the traps(5)
requirement.
If traps(5) is set to 0, then none of the notifications defined
in this MIB are generated for a particular tn3270eRtCollCtlEntry.
Either the average(3) or the buckets(4) bit MUST be set to 1 in order
for response time data collection to occur; both bits MAY be set to
1. If the average(3) bit is set to 1, then the following objects
have meaning, and are used to control the calculation of the
averages, as well as the generation of the two notifications related
to them:
o tn3270eRtCollCtlSPeriod
o tn3270eRtCollCtlSPMult
o tn3270eRtCollCtlThreshHigh
o tn3270eRtCollCtlThreshLow
o tn3270eRtCollCtlIdleCount
The previous objects' values are meaningless if the associated
average(3) bit is not set to 1.
If the buckets(4) bit is set to 1, then the following objects have
meaning, and specify the bucket boundaries:
o tn3270eRtCollCtlBucketBndry1
o tn3270eRtCollCtlBucketBndry2
o tn3270eRtCollCtlBucketBndry3
o tn3270eRtCollCtlBucketBndry4
The previous objects' values are meaningless if the associated
buckets(4) bit is not set to 1.
If an entry in the tn3270RtCollCtlTable has the value active(1) for
its RowStatus, then an implementation SHALL NOT allow Set operations
for any objects in the entry except:
o tn3270eRtCollCtlThreshHigh
o tn3270eRtCollCtlThreshLow
o tn3270eRtCollCtlRowStatus
4.2 tn3270eRtDataTable
Either a single entry or multiple entries are created in the
tn3270eRtDataTable for each tn3270eRtCollCtlEntry, depending on
whether tn3270eRtCollCtlType in the control entry has aggregate(0)
selected. The contents of an entry in the tn3270eRtDataTable depend
on the contents of the corresponding entry in the