Transcript INF5060: Multimedia data communication using network processors

INF5060:
Multimedia data communication using network processors
ACEs
24/9 - 2004
ACE
Active Computing Element
Defined by Intel’s SDK
Not part of hardware
Also sometimes called Action/Classification Engine
Idea behind ACEs
Element in a graph
Part of a classification engine
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
Demultiplexing
 Packet demultiplexing
 Used with layered
protocols
 Packet proceeds through
one layer at a time
 On input, software in
each layer choose
module at next higher
layer
 On output, type field in
each header specifies
encapsulation
Applications
TCP
UDP
IP
Payload
TCP Header
Ethernet Header
MAC
IPPayload
Header
TCP Header
Payload
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
 Packet classification
 Alternative to
demultiplexing
 Designed for higher speed
 Considers all layers at the
same time
TCP
 Linear in number of fields
 Two possible
implementation options
Applications
UDP
IP
 Software
 Hardware
MAC
Ethernet Header
IP Header
TCP Header
…
Payload
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
Idea
Classifiers perform a logical AND between fields
Hardware classifiers can perform AND operation in parallel
Reality
Need detailed distinctions involving many fields
Fields are not always at the same offset
Requirements for classification change over time
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
Need detailed distinctions involving many fields
Multiple categories
Web traffic
Telnet traffic
Mail traffic
ICMP traffic
other traffic
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
0
8
16
24
32
40
48
56
dest
type
identification
header checksum
dest. address (2-3)
seq. number (2-3)
window
possible
IP Options !!
flgs
frag. offsett
src. address
src. port
.local etype
immed[etype, ETH_IP]
alu_shf[ --, etype, -,
$$hdr3, >>16]
br!=0[NotWeb#]
.endlocal
vers hlen
service
ttl
type
dest. address (0-1)
dest. port
acknowledgement
checksum
…
src (0-1)
src (2-4)
total length
64
seq. number (0-1)
vers reserv.
urgent pointer
.local base boff dpoff dport
ld_field_w_clr[dpoff, 0001, $$hdr3,
.local mask
immed[mask, 0x3c]
alu [ dpoff, dpoff, AND, mask
.endlocal
alu[dpoff, dpoff, +, 16]
Buf_GetData[base, dl_buffer_handle]
DL_GetBufferOffset[boff]
alu[boff, boff, +, dpoff]
alu_shf[boff, --, B, boff, >>3]
sdram[read, $$hdr0, base, boff, 1],
alu[ dpoff, dpoff, AND, 0x7 ]
alu[ --, dpoff, -, 4]
br>=0[SecondWord#]
ld_field_w_clr[dport, 0011, $$hdr0,
br[GotDstPort#]
SecondWord#:
ld_field_w_clr[dport, 0011, $$hdr1,
GotDstPort#:
code
…
br!=byte[$$hdr5, 0,
IPT_TCP,
NotWeb#]
data
>>6]
…
.local wprt
immed[wprt, TCP_WWW]
alu[--, dport, -, wprt]
br!=0[NotWeb#]
.endlocal
]
…
ctx_swap
Logical AND …
>>16]
>>16]
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
Need detailed distinctions involving many fields
Multiple categories
Web traffic
Telnet traffic
Mail traffic
ICMP traffic
other traffic
Fields are not always at the same offset
Variable-sized headers
Fields not at fixed offsets
Easily handled with software
Finite cases can be specified in rules
Hybrid classification can be considered (hardware first, software next)
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
 Desirable
1. Fixed headers
 Protocols without any optional fields
 XTP
 Options in dedicated locations
 IPv6 – options follow header
2. Multiple classification stages
Hardware
classifier
Software
classifier
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Classification
Static Classification
Flow specified in rule sets
Header fields and values know a priori
Dynamic classification
“Flows” created by observing the packet stream
Values taken from headers
Allows fine-grained flows
Requires state information
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Example static classification
 Allocate one flow per
service type
 DiffServ
Version
IHL
DS
Total length
Identification
DM
Fragment offset
Time to live
Protocol
Header checksum
Source address
Destination Address
Options (0 or more)
 One header field used to
identify flow
Data
 IP ToS field / IP priority and
DS fields
DiffServ-enabled
Router
Padding
0 0

Class selector codepoints
 If of the form xxx000
DiffServ
classifiers

Routing table
INF5060 – multimedia communication using network processors
Differentiated Services Codepoint
 xxxxx0 reserved for standardization
 xxxx11 reserved for local use
 xxxx01 open for local use, may be
standardized later
2004 Carsten Griwodz & Pål Halvorsen
Dynamic classification
 Allocate flow per TCP connection
 Header fields used to identify flow
 IP source address
 IP destination address
 TCP source port number
 TCP destination port number
 Applications
 Clustered web servers
 Load leveller dispatching TCP connection to several web servers
 Firewalls
 Allow incoming packets only when outgoing connection was made
 More complicated firewall cases
 FTP data connection
 Voice over IP
 RTP video streaming
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Dynamic classification
 Flow identification
 By observing the packet stream
 Packets belonging to the flow
 FTP, SSH, …
 Packets outside the flow
 RSVP, IGMP, …
 Connection-oriented network
 Per-flow SVC can be created on demand
 Flow ID equals connection ID
 State
 Connection
 Allocated resources
 Connectionless network
 Flow ID used internally
 Each flow ID mapped to (next hop, interface)
 State
 Next hop, interface mapping
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Dynamic classification
Dynamic classification
Usually performed in software
State kept in memory
State information created/updated at wire speed
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
ACEs
ACE Features
 Active Computing Element
 Defined by Intel’s SDK
 Not part of hardware
 Also sometimes called Action/Classification Engine
 One ACE: one step in dynamic classification
 ACEs in Intel’s SDK
 Fundamental software building block
 Runs on StrongARM, microengine, or host
 Handles control plane and fast or slow data path processing
 Coordinates and synchronizes with other ACEs
 ACEs form a graph
 Used to construct packet processing systems
 Can have multiple inputs or outputs
 Can serve as part of a pipeline
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
ACE Terminology
 Microengine
 One of the specialized processors of the IXP card
 Microblock
 Code that runs
 in one thread
 on one microengine
 MicroACE
 Core component runs on StrongARM
 Microblock component runs on microengines
 Source microblock
 Handles ingress from I/O device
 Sink microblock
 Handles egress to I/O device
 Transform microblock
 Intermediate position in a pipeline
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Four conceptual parts of an ACE
Initialization
Classification
Actions associated with each classification
Message and event management
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Four conceptual parts of an ACE
 Initialization
 Classification
 Actions associated with each
classification
 Message and event
management
source/init.c
…
ix_error ix_init(int argc, char **argv, ix_ace ** ap)
{
struct wwbump *wwb;
ix_error e;
(void)argc;
*ap = 0;
wwb = malloc(sizeof(struct wwbump));
if ( wwb == NULL ) …
wwb->ue = atoi(argv[2]);
strcpy(wwb->bname, "WWBUMP");
wwb->name[sizeof(wwb->name) - 1] = '\0';
strncpy(wwb->name, argv[1], sizeof(wwb->name) - 1);
e = ix_ace_init(&wwb->ace, wwb->name);
if (e) …
e = RmInit();
if (e) …
e = RmRegister(&wwb->tag, wwb->bname,&wwb->ace,
exception, wwb,
wwb->ue);
if (e) …
e = cc_init(wwb);
if ( e )
…
*ap = &wwb->ace;
return 0;
}
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Four conceptual parts of an ACE
 Initialization
 Classification
 Actions associated with each
classification
 Message and event
management
ucbuild/WWBump.uc
…
.local etype
immed[etype, ETH_IP]
alu_shf[ --, etype, -,
$$hdr3, >>16]
br!=0[NotWeb#]
.endlocal
…
br!=byte[$$hdr5, 0,
IPT_TCP,
NotWeb#]
…
.local wprt
immed[wprt, TCP_WWW]
alu[--, dport, -, wprt]
br!=0[NotWeb#]
.endlocal
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Four conceptual parts of an ACE
 Initialization
 Classification
 Actions associated with
each classification
 Message and event
management
source/action.c
…
ix_error exception(void *ctx,
ix_ring r, ix_buffer b)
{
struct wwbump *wwb = (struct
wwbump *) ctx;
ix_error e;
unsigned char c;
(void) r;
e = RmGetExceptionCode(wwb->tag,
&c);
if ( e ) …
Webcnt++;
e = RmSendPacket(wwb->tag, b);
if ( e )
…
return 0;
}
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Four conceptual parts of an ACE
 Initialization
 Classification
 Actions associated with each
classification
 Message and event
management
INF5060 – multimedia communication using network processors
source/wwbcc.c
ix_error getcnt(ix_base_t* bp,
long* rv)
{
(void)bp;
*rv = Webcnt;
return 0;
}
2004 Carsten Griwodz & Pål Halvorsen
Output targets and late binding
 ACE has set of outputs
 Not in wwbump – only default
 Each output given target
name
 Outputs bound dynamically
at run time
 Unbound target corresponds
to packet discard
ixsys.config-wwbump
…
microace ifaceInput
./ingressAce none 0 1
microace ifaceOutput
./egressAce none 1 2
microace wwbump ./wwbump
none 0 0
…
bind static
ifaceInput/default wwbump
bind static wwbump/default
ifaceOutput
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Output targets and late binding
 ACE has set of outputs
 Create other targets in
source/init.c
 Each output given target
name
 Outputs bound dynamically
at run time
 Unbound target corresponds
to packet discard
INF5060 – multimedia communication using network processors
SDK Reference
ix_error ix_target_init(
ix_target* targp, ix_ace*
acep, const char* name);
 Targp – where to store the new
target handle
 Acep – ACE to which the new
target belongs
 Name – symbolic name of the new
target
2004 Carsten Griwodz & Pål Halvorsen
Conceptual ACE interconnection




Ingress ACE acts as source
Process ACE acts as transform
Egress ACE acts as sink
Connections created at run time
ixsys.config-wwbump
…
bind static ifaceInput/default wwbump
bind static wwbump/default ifaceOutput
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Components of MicroACE
Single ACE with two
components
StrongARM (core
component)
Microengines (microblock
component)
arm-be/wwbump
SlowIngressWWBump.uof
ucbuild/WWBump.uc
…
immed[dl_next_block, IX_EXCEPTION]
br[Finish#]
…
ucbuild/WWB_dl.uc
Communication possible
between components
…
alu[ --, dl_next_block, -,
IX_EXCEPTION]
br=0[Send_SA#]
…
Send_SA#:
DL_SASink[ ]
.continue
…
source/action.c
…
ix_error exception(void *ctx,
ix_ring r, ix_buffer b)
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Communication: Exceptions
Packets passed from microblock to core component
Mechanism
Microcode set dl_next_block to IX_EXCEPTION
Dispatch loop forwards packet to core
ACE tag used to identify corresponding component
Exception handler is involed in core component
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Actions applied to exceptions
 Consume the packet and free
the buffer
SDK Reference
ix_error ix_buffer_del(ix_buffer*
buf)
source/action.c
 Modify the packet before
sending it on
 Send the packet back to the
microblock for further
processing
ix_error exception(
void *ctx, ix_ring r, ix_buffer
b)
{ …
e = RmSendPacket(
wwb->tag, b);
… }
SDK Reference
 Forward the packet to another
ACE on the StrongARM
INF5060 – multimedia communication using network processors
ix_error ix_target_take(ix_target*
targp, ix_buffer* buf);
targp – bound target to pass to
buf – buffer to pass to target
2004 Carsten Griwodz & Pål Halvorsen
Division of ACE into components
 Microblocks form fast path
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Illustration of microblock groups




Set of one or more microblocks
Treated as single unit
Loaded onto one microengine for execution
Can be replicated on microengines for higher speed
 Entire microblock group assigned to same microengine
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Illustration of microblock groups




Set of one or more microblocks
Treated as single unit
Loaded onto one microengine for execution
Can be replicated on microengines for higher speed
ucbuild/WWB_dl.uc
…
DL_Init[]
EthernetIngress_Init[]
WWBumpInit[]
Dispatch
loop
.while(1)
Top_Of_Loop#:
DL_SASource[ ]
alu[--, dl_buffer_handle, -, IX_BUFFER_NULL]
br=0[Test_Ingress#], guess_branch
br[Send_MB#]
WWBump
microblock
SlowIngress
microblock
Test_Ingress#:
EthernetIngress[ ]
alu[--, dl_buffer_handle, -, IX_BUFFER_NULL]
br=0[Top_Of_Loop#]
…
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Microblock structure
Asynchronous model
Neither core ACE nor MicroACE may block
Programmer creates
Initialization macro
Dispatch loop
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Event loop
Central to asynchronous programming model
Uses polling
Repeatedly checks for presense of event(s) and calls
appropriate handler
Can be hidden from programmer
In ACE model
Explicit
Programmer can modify/extend
Note
Event loops are also used when programming Berkeley
sockets with select()
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Event loop processing
If event loop stops
All processing stops
The arrival of a new event will not trigger invocation of an
event handler
Conclusion: the event loop must go on
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Crosscall mechanism
Used between
Core component of one ACE and another
ACE core component and non-ACE application
Not intended for packet transfer
Operates like Remote Procedure Call
Mechanism know as crosscall
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Crosscall implementation
 Both caller and callee programmed
to use crosscall
 Not dynamic
 Not at run-time
include/idl/wwbump.idl
 Declaration given in Interface
Definition Language (IDL)
 IDL compiler
 Read specification
 Generates stubs that handle calling
details
INF5060 – multimedia communication using network processors
interface wwbump
{
twoway long getcnt();
};
include/wwbump_cb_c.h
include/wwbump_cc_c.h
include/wwbump_import.h
include/wwbump_sk_c.h
include/wwbump_stub_c.h
2004 Carsten Griwodz & Pål Halvorsen
Crosscall implementation
Three types of crosscalls
Deferred: caller does not block; return notification
asynchronously
Oneway: caller does not block; no value returned
Twoway: caller blocks; callee returns a value
Twoway call corresponds to traditional RPC
Type of call determined at compile time
Note
ACEs are not allowed to block
Therefore an ACE can not call a twoway crosscall (only
answer one)
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Packet queues
Packet queues
Generated by the SDK when ACEs communication
asynchronously
Buffer packets
Placed between ACE components
Permit asynchronous operation
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Illustration of packet queues
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Packet queues
Operation
 ix_target refers to the ring used to transfer packets from the current
ACE to another ACE
 The target may be bound or not
 If bound, buffers to the target (using ix_target_take) are placed in the
ring
 If unbound, buffers to the target are discarded
 ix_res_bind allocates a new ring for a target
 ix_res_unbind disassociates current packet processing queue from
the target
 all previously taken packets will still processed by the target ACE
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen
Summary
Intel SDK uses ACE programming model
ACE
Basic unit of computation
Can include code for StrongARM (core) and microengines
(microblock)
Packet queues used to pass packets between ACEs
Crosscall mechanism used for nonpacket
communication
INF5060 – multimedia communication using network processors
2004 Carsten Griwodz & Pål Halvorsen