CMSC 411 Computer Systems Architecture Lecture 18 Storage Systems 2 I O performance measures I O performance measures Throughput vs latency diversity which I O devices can connect to the system capacity how many I O devices can connect to the system bandwidth throughput or how much data can be moved per unit time latency response time the interval between a request and its completion High throughput usually means slow response time CMSC 411 18 some from Patterson Sussman others Fig 6 9 3 CMSC 411 18 some from Patterson Sussman others Improving performance cont Modeling computer performance Adding another server can decrease response time if workload is held constant but keeping work balanced between servers is difficult To design a responsive system must understand what the typical user wants to do with it Each transaction consists of three parts entry time the time for the user to make the request system response time the latency think time the time between system response and the next entry Key observation is that a faster system produces a lower think time see Fig 6 10 The usual way to model computer performance is using queuing theory mathematics again Unfortunately even queuing theory does not provide a very good model so more complicated mathematics is now being applied e g stochastic differential equations But H P only consider queuing models and we don t even have time to go into that now maybe later CMSC 411 18 some from Patterson Sussman others CS252 S05 Fig 6 8 5 CMSC 411 18 some from Patterson Sussman others 4 6 Data Management Issues Two concerns we ll talk about stale data DMA design And the book has short discussions of several more including asynchronous I O through the OS file systems server manages blocks and maintains metadata vs disks doing it and server uses file system protocol such as NFS also called NAS network attached storage Data Management Issues 8 CMSC 411 18 some from Patterson Sussman others Stale Data Stale Data cont May have copies of data in cache memory disk Need to make sure that always use the most recent version for use in the CPU for output Two approaches to the problem both having disadvantages Approach 1 Attach the I O bus to the cache Advantage No problem of stale data since CPU and I O devices all see the copy in the cache Disadvantage All I O data must go through the cache even if the CPU doesn t need it so performance is reduced CPU and I O bus must take turns accessing the cache so arbitration hardware required CMSC 411 18 some from Patterson Sussman others 9 CMSC 411 18 some from Patterson Sussman others Stale Data cont Stale Data cont Approach 2 Attach the I O bus to the memory Advantage I O does not slow down the CPU Disadvantage The I O system may see stale data unless we do write through CPU might see stale data if the I O system modifies memory after the cache copied it Extra hardware is required to check whether I O data is currently held in cache 10 Fig 7 15 of H P 3ed CMSC 411 18 some from Patterson Sussman others CS252 S05 11 CMSC 411 18 some from Patterson Sussman others 12 DMA design Direct memory access hardware needs to use either virtual addresses or physical addresses Using physical addresses If the data is longer than a page then several addresses need to be passed The data may be relocated by the operating system changing the physical address Virtual addresses gives a cleaner design CMSC 411 18 some from Patterson Sussman others 13 Designing an I O System I O Systems Price performance and capacity issues Need to choose which I O devices to connect how to connect them Example The CPU is seldom the limiting factor for I O performance Suppose the CPU can handle 10 000 I O operations per second IOPS And suppose the average I O size is 16 KB The other links in the I O chain are the I O controller suppose it adds 1 ms overhead per I O operation the I O bus suppose it is a bus that can transfer 20 MB sec 20 KB ms the disk suppose it rotates at 7200 RPM with 8 ms average seek time and 6 MB sec transfer rate CMSC 411 19 some from Patterson Sussman others 15 CMSC 411 19 some from Patterson Sussman others 16 I O System Performance Storage Example Internet Archive Consider the disk time first 7200 RPM 7200 60 103 12 revolutions per ms 6 MB sec 6 KB ms So the average disk time is seek rotational latency transfer 8 ms 5 12 ms 16 6 14 9 ms So the average time per transfer is I O controller time bus time disk time 1 ms 16 20 ms 14 9 ms 16 7 ms So with one controller one bus and one disk can do at most 1 16 7 10 3 60 IOPS If this is not good enough should analyze to see whether it is better to add more controllers more buses or more disks Another more complex performance analysis in Section 6 7 for the Internet Archive Cluster Goal of making a historical record of the Internet Internet Archive began in 1996 Wayback Machine interface performs time travel to see what the website at a URL looked like in the past It contains over a petabyte 1015 bytes and is growing by 20 terabytes 1012 bytes of new data per month In addition to storing the historical record the same hardware is used to crawl the Web every few months to get snapshots of the Internet CMSC 411 19 some from Patterson Sussman others CS252 S05 Designing an I O System 17 12 1 2009 some from Patterson Sussman others 18 Internet Archive Cluster Estimated Cost VIA processor 512 MB of DDR266 DRAM ATA disk controller power supply fans and enclosure 500 7200 RPM Parallel ATA drive holds 500 GB 375 48 port 10 100 1000 Ethernet switch and all cables for a rack 3000 Cost 84 500 for a 80 TB rack 160 Disks are 60 of the cost 1U storage node PetaBox GB2000 from Capricorn Technologies Contains 4 500 GB Parallel ATA PATA disk drives 512 MB of DDR266 DRAM one 10 100 1000 Ethernet interface and a 1 GHz C3 Processor from VIA 80x86 Node dissipates 80 watts 40 GB2000s in a standard VME rack 80 TB of raw storage capacity 40 nodes are connected with a 48 port 10 100 or 10 100 1000 Ethernet switch 1 PetaByte 12 racks 12 1 2009 some from Patterson Sussman others 19 some from Patterson Sussman others Estimated Performance Estimated Reliability 7200 RPM Parallel ATA drive holds 500 GB has an average time seek of 8 5 ms transfers at 50 MB second from the disk The PATA link speed is 133 MB second performance of the VIA processor is 1000 MIPS operating system uses 50 000 CPU instructions for a disk I O network protocol stack uses 100 000 …