Chinaunix首页 | 论坛 | 博客
  • 博客访问: 661798
  • 博文数量: 291
  • 博客积分: 10025
  • 博客等级: 上将
  • 技术积分: 2400
  • 用 户 组: 普通用户
  • 注册时间: 2004-12-04 12:04
文章分类

全部博文(291)

文章存档

2008年(102)

2007年(112)

2006年(75)

2004年(2)

我的朋友

分类: 服务器与存储

2008-03-10 14:02:13

The three boxes are indeed different in their architecture.

IBM DS8000 is basically a modified pSeries cluster with a dedicated
operating system to manage IO, with all the defects and the limits of
such configuration (for istance processing power does not scale with
configuration and so you have performances degrading as the
configuration grows). The cache is the two servers memory, with a
portion of it backed up with batteries for mirrored writes.

EMC DMX4 and HDS USP/USPV have specifically designed architectures for
managing IO's somewhat similar.

The core of DMX4 is a 128 path (1 GBytes/sec each) point to point matrix
connecting up to 8 Channel Directors (CD) for front end and up to 8 Disk
Directors (DD) for backend to a maximum of 8 Memory Directors (MD)
granting access to the cache. Every CD and DD has 8 processors, so the
box in the maximum configuration has 128 processors. The point to point
architecture makes somewhat rigid the access to the cache that in fact
is divided in 32 segments (not a global cache really). Every cache
segment has a single access from MD (there is an arbitration circuit
every 4 point to point connections), so the parallelism to the cache is
32 ways for 32 GBytes/sec of throughput.

HDS uses instead an architecture based on up to 4 T-Bar Cross switches
(Hitachi Ltd. Design) connecting up to 16 Directors to the cache. USPV
directors configurations are flexible in the sense you can have up to 14
Front End Directors (FED) and up to 8 Back End Directors configurable in
groups of two, given that sum can not exceed 16. So you can start froma
2 FED + 2 BED config up to a 8 FED + 8 BED configuration or a 14 FED + 2
BED configuration. Being accessed through the switches the cache is not
segmented and has a throughput of 68 GBytes/sec (64 paths by 1.062
GB/sec each). As EMC each Director has 8 processors, so in the max
config USPV has 128 processors.

Very recently HDS has publicly released the results of SPC-1 benchmark
on USP V that put the box at the top of the performances for this class
of storage, almost doubling IBM DS8300. By policy EMC does not gives any
performance benchmark results so it is not possible a real comparison.
阅读(594) | 评论(0) | 转发(0) |
给主人留下些什么吧!~~