1

Important Announcements

Midterm 3 is on Wednesday, April 20 from 7pm to 8:30pm— Practice Midterm 1 released tonight— Please email me ASAP in case you need a conflict

Final Exam (cumulative) is on Monday, May 9 from 1:30pm to 4:30pm— Please email me ASAP in case you need a conflict

2

Virtual Memory system

virtual address

data

physical address

TLB

page table

memory

cachedisk

page offset

page offsetvirtual page number (VPN)

PPN

tag index blockoffset

3

Hard drives

The textbook shows the ugly guts of a hard disk—Data is stored on double-sided magnetic disks called platters—Each platter is arranged like a record, with many concentric

tracks—Tracks are further divided into individual sectors, which are the

basic unit of data transfer—Each surface has a read/write head like the arm on a record

player, but all the heads are connected and move together

A 75GB IBM Deskstar has roughly:—5 platters (10 surfaces),—27,000 tracks per surface,—512 bytes per sector,—~512 sectors per track…

…but this number increases goingfrom the center to the rim

Platter

Track

Platters

Sectors

Tracks

4

There are two fundamental performance metrics for I/O systems:

1. Latency Time to initiate data-transfer (units = sec)

2. Bandwidth Rate of initiated data-transfer (units = bytes/sec)

Time = latency + transfer_size / bandwidth

sec bytes / (bytes/sec)

I/O Performance

Dominant term forsmall transfers

Dominant term forlarge transfers

5

Accessing data on a hard disk Factors affecting latency:

—Seek time measures the delay for the disk head to reach the track—A rotational delay accounts for the time to get to the right sector

Factors affecting bandwidth:—Usually the disk can read/write as fast as it can spin—Bandwidth is determined by the rotational speed, which also

determines the rotational delay

We can compute average seek time/rotational delay/etc. but careful placement of data on the disk can speed things up considerably:—head is already on or near the desired track—data in contiguous sectors—in other words, locality!

Even so, loading a page from the hard-disk can take tens of milliseconds

6

Parallel I/O Many hardware systems use parallelism for increased speed

A redundant array of inexpensive disks or RAID system allows access to several hard drives at once, for increased bandwidth—similar to interleaved memories from last week

7

MP6: Due Friday 4/29 Input: A string s Output: Maximum number of non-overlapping increasing substrings in s

A string x is larger than string y if— length(x) length(y) (e.g. “aa” “z”); or— length(x) = length(y) but x is lexicographically bigger (e.g. “z” “a”)

Example 1: s = aaa splits into two increasing substrings “a” and “aa”

Example 2: s = abc splits into three increasing substrings “a”, “b”, “c”

Example 2: s = aaaa splits into two increasing substrings “a” and “aa”; or “a” and “aaa”

Example 3: s = aaaaaa splits into three increasing substrings “a”, “aa” and “aaa”

8

MP6: Algorithm

Recurrence:

C++ code: Uses a hashtable (STL map) for best(index, substr) values

Optimization techniques:— Compiler optimizations— Better (more cache-friendly) data structure— Better algorithm— Exploit parallelism— Manual optimizations

best(i, last)

next

i j

= 1 + best(j, next) best(i+1, last)