Files
pva2pva/documentation/doc.lyx
Michael Davidsaver 09a6583e80 more writeup
2015-12-07 22:25:46 -05:00

868 lines
17 KiB
Plaintext

#LyX 2.1 created this file. For more info see http://www.lyx.org/
\lyxformat 474
\begin_document
\begin_header
\textclass article
\use_default_options true
\maintain_unincluded_children false
\language english
\language_package default
\inputencoding auto
\fontencoding global
\font_roman lmodern
\font_sans lmss
\font_typewriter lmtt
\font_math auto
\font_default_family default
\use_non_tex_fonts false
\font_sc false
\font_osf false
\font_sf_scale 100
\font_tt_scale 100
\graphics default
\default_output_format pdf2
\output_sync 0
\bibtex_command default
\index_command default
\paperfontsize default
\spacing single
\use_hyperref true
\pdf_bookmarks true
\pdf_bookmarksnumbered false
\pdf_bookmarksopen false
\pdf_bookmarksopenlevel 1
\pdf_breaklinks false
\pdf_pdfborder true
\pdf_colorlinks true
\pdf_backref false
\pdf_pdfusetitle true
\pdf_quoted_options "linkcolor=blue,citecolor=blue"
\papersize letterpaper
\use_geometry false
\use_package amsmath 1
\use_package amssymb 1
\use_package cancel 1
\use_package esint 1
\use_package mathdots 1
\use_package mathtools 1
\use_package mhchem 1
\use_package stackrel 1
\use_package stmaryrd 1
\use_package undertilde 1
\cite_engine basic
\cite_engine_type default
\biblio_style plain
\use_bibtopic false
\use_indices false
\paperorientation portrait
\suppress_date false
\justification true
\use_refstyle 1
\index Index
\shortcut idx
\color #008000
\end_index
\secnumdepth 2
\tocdepth 2
\paragraph_separation indent
\paragraph_indentation default
\quotes_language english
\papercolumns 1
\papersides 1
\paperpagestyle default
\tracking_changes false
\output_changes false
\html_math_output 0
\html_css_as_file 0
\html_be_strict false
\end_header
\begin_body
\begin_layout Title
PVAccess Gateway Design
\end_layout
\begin_layout Author
Michael Davidsaver
\end_layout
\begin_layout Date
\begin_inset ERT
status collapsed
\begin_layout Plain Layout
\backslash
today
\end_layout
\end_inset
\end_layout
\begin_layout Section
Scope
\end_layout
\begin_layout Standard
A PVAccess gateway will act as a
\begin_inset Quotes eld
\end_inset
switch
\begin_inset Quotes erd
\end_inset
to route PVA traffic between IP subnets.
The primary use cases for a gateway are: 1) de-duplicate data when many
PVA clients make a similar request to limit the load on the PVA server
handling this request.
And 2) to provide the mechanisms to implement site specific access control
policies specifying which requests are permitted.
\end_layout
\begin_layout Standard
\begin_inset Float figure
wide false
sideways false
status collapsed
\begin_layout Plain Layout
\align center
\begin_inset Graphics
filename parts.dia
height 1in
\end_inset
\end_layout
\begin_layout Plain Layout
\begin_inset Caption Standard
\begin_layout Plain Layout
Gateway Components
\end_layout
\end_inset
\begin_inset CommandInset label
LatexCommand label
name "fig:parts"
\end_inset
\end_layout
\begin_layout Plain Layout
\end_layout
\end_inset
\end_layout
\begin_layout Section
Terms
\end_layout
\begin_layout Description
IP Internet Protocol (v4 and/or v6)
\end_layout
\begin_layout Description
PV Process Variable.
Addressable unit in PVA.
A character string.
\end_layout
\begin_layout Description
PVA PVAccess network protocol
\end_layout
\begin_layout Description
GW Gateway
\end_layout
\begin_layout Description
CLI An arbitrary PVAccess client (end user or another gateway)
\end_layout
\begin_layout Description
GWS Gateway server side (CLI communicates with this)
\end_layout
\begin_layout Description
GWC Gateway client side (SRV communicates with this)
\end_layout
\begin_layout Description
SRV An arbitrary PVAccess server (may be another gateway)
\end_layout
\begin_layout Standard
Figure
\begin_inset CommandInset ref
LatexCommand ref
reference "fig:parts"
\end_inset
shows the relationships between the CLI, GWS, GWC, and SRV actors.
The
\begin_inset Quotes eld
\end_inset
Gateway
\begin_inset Quotes erd
\end_inset
consists of the GWS, GWC, and the logic which sits between them.
\end_layout
\begin_layout Section
Goals/Features
\end_layout
\begin_layout Standard
Identified features and design goals.
\end_layout
\begin_layout Paragraph
De-duplication
\end_layout
\begin_layout Standard
of sockets and data to reduce overall resource use.
\end_layout
\begin_layout Paragraph
Policies
\end_layout
\begin_layout Standard
to permit or deny individual requests.
This include access control policy (ie.
client
\begin_inset Quotes eld
\end_inset
X
\begin_inset Quotes erd
\end_inset
may not access resource
\begin_inset Quotes eld
\end_inset
Y
\begin_inset Quotes erd
\end_inset
) and administrative limits to bound per-client resource usage (ie.
a client may have only 15 concurrent
\begin_inset Quotes eld
\end_inset
gets
\begin_inset Quotes erd
\end_inset
in progress).
\end_layout
\begin_layout Section
Theory of Operation
\end_layout
\begin_layout Subsection
Name resolution/socket setup
\end_layout
\begin_layout Standard
Use of periodically re-sent UDP messages by PVA for name search permits
an asynchronous mode of operation for name resolution and socket (circuit)
setup.
\end_layout
\begin_layout Standard
\begin_inset listings
lstparams "language={C++}"
inline false
status open
\begin_layout Plain Layout
struct ChannelCacheEntry {
\end_layout
\begin_layout Plain Layout
string name; // key
\end_layout
\begin_layout Plain Layout
int priority; // key
\end_layout
\begin_layout Plain Layout
unsigned refcount; // implicily via.
shared_ptr
\end_layout
\begin_layout Plain Layout
bool searched;
\end_layout
\begin_layout Plain Layout
shared_ptr<Channel> chanGWC;
\end_layout
\begin_layout Plain Layout
};
\end_layout
\end_inset
\end_layout
\begin_layout Paragraph
A cache of GWC Channels
\end_layout
\begin_layout Standard
index by name and priority (the parameters of the createChannel message)
is maintained.
Entries in this ChannelCache have an associated GWC Channel, and a garbage
cleanup flag.
\end_layout
\begin_layout Standard
When GWS receives a new CLI search request message a lookup is made to the
ChannelCache which has three possible outcomes (see figure
\begin_inset CommandInset ref
LatexCommand ref
reference "fig:name-search"
\end_inset
).
\end_layout
\begin_layout Standard
\begin_inset Float figure
wide false
sideways false
status collapsed
\begin_layout Plain Layout
\align center
\begin_inset Graphics
filename name-search.msc
width 3in
height 2in
keepAspectRatio
\end_inset
\end_layout
\begin_layout Plain Layout
\begin_inset Caption Standard
\begin_layout Plain Layout
CLI name search outcomes
\end_layout
\end_inset
\begin_inset CommandInset label
LatexCommand label
name "fig:name-search"
\end_inset
\end_layout
\begin_layout Plain Layout
\end_layout
\end_inset
\end_layout
\begin_layout Description
Miss There is no entry in the cache.
A new entry is created.
The new GWC Channel begins the search/connect process.
No reply to CLI is made.
\end_layout
\begin_layout Description
Not
\begin_inset space ~
\end_inset
Conn An entry exists, but the associated GWC Channel is not yet connected.
No further action is taken.
No reply to CLI is made.
\end_layout
\begin_layout Description
Hit An entry exists with a connected GWC Channel.
A positive search reply is sent to CLI.
\end_layout
\begin_layout Standard
Handling of TCP channelCreate messages by GWS is similar except that a negative
reply is sent for Miss and Not Conn outcomes.
A negative reply to a channelCreate message by CLI will cause it to restart
the name search phase.
\end_layout
\begin_layout Paragraph
A reference count
\end_layout
\begin_layout Standard
is maintained by each ChannelCache entry for the GWS channels using the
associated GWC channel.
However, an entry should not be immediately removed when its ref.
count drops to zero.
Instead it should be kept for some time so that it may be found by future
CLI search requests.
\end_layout
\begin_layout Standard
Each ChannelCache entry will have a boolean flag which is set on creation,
and re-set whenever it is found by a name lookup (Miss or Not Conn).
A periodic cleanup task should run which removes all entries with this
flag cleared and a zero ref.
count.
Each time it run, the cleanup task clears the flag of any entry not removed.
\end_layout
\begin_layout Standard
This should ensure that unused entries are eventually removed when no client
is searching for them.
\end_layout
\begin_layout Paragraph
Ownership of an active GWC Channel
\end_layout
\begin_layout Standard
is shared by several active GWS channels, and a ChannelCache entry as shown
in figure
\begin_inset CommandInset ref
LatexCommand ref
reference "fig:clichanown"
\end_inset
.
Red arrows represent strong ownership and blue weak ownership.
\end_layout
\begin_layout Standard
\begin_inset Float figure
wide false
sideways false
status collapsed
\begin_layout Plain Layout
\align center
in general
\begin_inset Graphics
filename structs.dot
width 4in
height 3in
keepAspectRatio
\end_inset
\end_layout
\begin_layout Plain Layout
\begin_inset Caption Standard
\begin_layout Plain Layout
GW Channel ownership
\end_layout
\end_inset
\begin_inset CommandInset label
LatexCommand label
name "fig:clichanown"
\end_inset
\end_layout
\begin_layout Plain Layout
\end_layout
\end_inset
\end_layout
\begin_layout Paragraph
Notification of loss of a GWC Channel
\end_layout
\begin_layout Standard
should result in the disconnection of any associated GWS Channels, and the
immediate removal of the associated ChannelCache entry.
This allows the ref.
count for an entry to fall to zero once all GWS Channels close.
\end_layout
\begin_layout Standard
All pending operations associated with this channel will also be canceled.
\end_layout
\begin_layout Subsection
Get/Put/RPC/...
operations
\end_layout
\begin_layout Standard
For operations other than Monitor, the timing of the client request can
effect the results.
No caching or de-duplication can be done without special knowledge about
the intended behavior of CLI and SRV.
Therefore, by default these operations pass through the GW without de-duplicati
on.
\end_layout
\begin_layout Subsection
Monitor operations
\end_layout
\begin_layout Standard
Each ChannelCache entry will also include a MonitorCache.
This cache is indexed by the pvrequest given with the corresponding monitorCrea
te operation.
As a pvrequest may contain arbitrary data, two pvrequests may not be compared
for anything other than exact equality without special knowledge.
Therefore, by default MonitorCache hits are only generated when the CLI
provides a pvrequest which exactly matches the MonitorCache entry.
\end_layout
\begin_layout Standard
Each MonitorCache entry should also keep the most recent value received
by the GWC so that this may be returned immediately for new GWS subscriptions.
\end_layout
\begin_layout Standard
A list of GWS consumers (interested in event data) will also be maintained.
Each new event data value is passed into the MonitorRequester of each consumer.
\end_layout
\begin_layout Standard
\begin_inset listings
lstparams "language={C++}"
inline false
status open
\begin_layout Plain Layout
struct MonitorCacheEntry {
\end_layout
\begin_layout Plain Layout
PVField request; // key
\end_layout
\begin_layout Plain Layout
weak_ptr<ChannelCacheEntry> chan;
\end_layout
\begin_layout Plain Layout
unsigned refcount; // implicily via.
shared_ptr
\end_layout
\begin_layout Plain Layout
shared_ptr<Monitor> mon;
\end_layout
\begin_layout Plain Layout
PVField lastval;
\end_layout
\begin_layout Plain Layout
list<MonitorConsumer*> consumers;
\end_layout
\begin_layout Plain Layout
};
\end_layout
\begin_layout Plain Layout
struct MonitorConsumer {
\end_layout
\begin_layout Plain Layout
shared_ptr<MonitorRequester> queueGWS;
\end_layout
\begin_layout Plain Layout
shared_ptr<MonitorCacheEntry> entry;
\end_layout
\begin_layout Plain Layout
bool GC;
\end_layout
\begin_layout Plain Layout
};
\end_layout
\end_inset
\end_layout
\begin_layout Paragraph
A reference count
\end_layout
\begin_layout Standard
is maintained for each MonitorCacheEntry, which will be closed as soon as
its ref.
count falls to zero.
\end_layout
\begin_layout Section
Channel Transmit Queuing
\end_layout
\begin_layout Standard
A major potential pitfall of connection/socket sharing for distinct sub-channels
(PVA Channels) is the handling of prioritization (or lack thereof) of traffic
for different sub-channels.
For example, monitoring a single high data rate PV can cause other PVA
Channels to experience higher latency.
One way to mitigate this, in part, is to introduce some
\begin_inset Quotes eld
\end_inset
fairness
\begin_inset Quotes erd
\end_inset
to the circuit/Transport transmit message queue.
\end_layout
\begin_layout Standard
Instead of a simple FIFO fed by all Channels, each channel given a FIFO.
The task which dequeues would do so by taking from each FIFO in turn in
a round robin.
This should prevent the overall latency through the queue from being dominated
by one fast PV.
\end_layout
\begin_layout Standard
However, this is only a partial solution as PVs with a large data size can
still introduce latency in proportion to this size.
\end_layout
\begin_layout Standard
For example, data updates from three source (A, B, and C) are added in one
order (line
\begin_inset CommandInset ref
LatexCommand ref
reference "line:queue1"
\end_inset
).
The
\begin_inset Quotes eld
\end_inset
fair
\begin_inset Quotes erd
\end_inset
queuing algorithm internally maintains a list of lists (line
\begin_inset CommandInset ref
LatexCommand ref
reference "line:queue2"
\end_inset
).
Round robin dequeuing from this list of lists gives the a different ordering
as seen on line
\begin_inset CommandInset ref
LatexCommand ref
reference "line:queue3"
\end_inset
.
\end_layout
\begin_layout Standard
\begin_inset listings
lstparams "language=Python,numbers=left,escapechar={|}"
inline false
status open
\begin_layout Plain Layout
queue = [A, A, B, A, B, C, C, A, A, B]|
\backslash
label{line:queue1}|
\end_layout
\begin_layout Plain Layout
sort_fair(queue) # [[A, A, A, A, A], [B, B, B], [C, C]]|
\backslash
label{line:queue2}|
\end_layout
\begin_layout Plain Layout
queue == [A, B, C, A, B, C, A, B, A, A]|
\backslash
label{line:queue3}|
\end_layout
\end_inset
\end_layout
\begin_layout Section
Loop avoidance
\end_layout
\begin_layout Standard
Another potential pitfall inherent in using UDP broadcasts for name resolution
is the possibility of loops should GWS receive search requests from GWC.
This can be avoided provided that GWS is aware of the set of endpoints
that GWC uses to send requests, and ignores an requests from them.
\end_layout
\begin_layout Section
Policies
\end_layout
\begin_layout Standard
Along with de-duplication, enforcement of administrative policies is a major
function of a GW.
Areas of policy include: access control, resource limits, and queuing behavior.
\end_layout
\begin_layout Subsection
Access Control
\end_layout
\begin_layout Standard
In order to share GWC channels, the GW will make all access control decisions.
Authentication information provided by CLI is
\series bold
never
\series default
forwarded to SRV.
Instead, the GW's own authentication information is sent to SRV.
\end_layout
\begin_layout Standard
Access control needs to be configurable on a per-PV and per-operation basis.
In similar contexts this typically takes the form of an Access Control
List, where rules are traversed in some order.
Each rule makes a decision to Allow, Deny, or Pass to the next rule.
\end_layout
\begin_layout Standard
To allow dynamic policies, a rule may itself subscribe to several PVs and
use the values obtained, in addition to client provided information and
static configuration, to make a decision.
\end_layout
\begin_layout Subsection
Administrative Limits
\end_layout
\begin_layout Standard
A number of configurable administrative limits should also be provided to
limit the resource usage of potentially misbehaving clients including:
\end_layout
\begin_layout Itemize
Max # of clients for GWS
\end_layout
\begin_layout Itemize
Max # of channels per client
\end_layout
\begin_layout Itemize
Max # of concurrent operations, for each operation type
\end_layout
\begin_layout Itemize
Max # monitor queue depth
\end_layout
\begin_layout Standard
Such limits could be made hard (fail further requests) or soft (log and
allow).
\end_layout
\begin_layout Subsection
Queuing
\end_layout
\begin_layout Standard
At a minimum, the default and max queue sizes should be settable by policy.
Additionally, a choice of algorithms could be provided to decide which
entries to drop when a queue overflows.
\end_layout
\end_body
\end_document