欢迎访问ic37.com |
会员登录 免费注册
发布采购

PEX8532-BB25BI 参数 Datasheet PDF下载

PEX8532-BB25BI图片预览
型号: PEX8532-BB25BI
PDF下载: 下载PDF文件 查看货源
内容描述: [PCI Bus Controller, CMOS, PBGA680, 35 X 35 MM, 2.23 MM HEIGHT, 1 MM PITCH, BGA-680]
分类和应用: 时钟数据传输PC外围集成电路
文件页数/大小: 512 页 / 4374 K
品牌: PLX [ PLX TECHNOLOGY ]
 浏览型号PEX8532-BB25BI的Datasheet PDF文件第132页浏览型号PEX8532-BB25BI的Datasheet PDF文件第133页浏览型号PEX8532-BB25BI的Datasheet PDF文件第134页浏览型号PEX8532-BB25BI的Datasheet PDF文件第135页浏览型号PEX8532-BB25BI的Datasheet PDF文件第137页浏览型号PEX8532-BB25BI的Datasheet PDF文件第138页浏览型号PEX8532-BB25BI的Datasheet PDF文件第139页浏览型号PEX8532-BB25BI的Datasheet PDF文件第140页  
Performance Metrics  
PLX Technology, Inc.  
8.5  
Latency  
8.5.1  
Queuing Effect  
In switches with large internal buffers, the latency increases once internal queuing is developed. The  
packet at the end of an egress VC&T queue does not transmit until all packets in front of it are  
transmitted. Assume the egress RAM of a x4 port is packed with packets of the same VC&T, draining  
the entire egress RAM takes 2,560 clocks (512 beats x 20B per beat / 4B/clock). Worst-case packet  
latency can be as long as 10 µs.  
To overcome the queuing effect, attempt the following:  
Avoid creating hot spots. In particular, ensure that the upstream port width in a host-centric  
application matches the sum width of all active downstream ports. Otherwise, the upstream port  
can easily become a hot spot when all downstream ports are attempting to transmit packets to it.  
Program a small Egress queue packet upper and lower limit, to avoid packet accumulation in an  
egress port. Section 8.3.2.4 describes how to program these thresholds. Lower latency is achieved  
at the cost of reducing the PEX 8532’s capability to buffer transient congestion.  
Reduce traffic load. Lighter traffic is less likely to experience congestion and can drain relatively  
faster, as the egress links can drain at the full link rate.  
8.5.2  
Time Division Multiplex Effect  
As previously illustrated, the PEX 8532 source station employs port-to-station aggregation, and the  
destination station employs station-to-port de-aggregation. Time Division Multiplex (TDM) controls  
aggregation and de-aggregation. Usually, waiting for a proper TDM slot to process packet coming from  
or going to a particular port increases the latency. The wider the port, the more TDM slots that port  
owns; therefore, the less latency contributed by TDM.  
Within a station, only a subset of 16 lanes are connected to SerDes. One approach to reduce latency is to  
strap the port as a wider port and allow it to negotiate down to the expected link width. For example, if  
there is only a x2 port owned by a station, the port can be strapped as x16, and allowed to become x2  
later through the normal link training process. As a result, all TDM slots in this station are acquired by  
the x2 port. The worst case TDM effect for a x1 or x2 port is 14 symbol times = 57 ns.  
114  
ExpressLane PEX 8532AA/BA/BB/BC 8-Port/32-Lane Versatile PCI Express Switch Data Book  
Copyright © 2007 by PLX Technology, Inc. All Rights Reserved – Version 1.6  
 复制成功!