AFS Server Performance Comparisons Bo Tretta Kim Kimball Jet - - PowerPoint PPT Presentation

▶

Feb 05, 2023 485 likes •701 views

AFS Server Performance Comparisons Bo Tretta Kim Kimball Jet Propulsion Laboratory Information Services - FIL Service http://fil.jpl.nasa.gov SLAC AFS Best Practices Workshop March 24, 2004 JPLIS-FIL Server Performance Comparisons 1 n

SLIDE 1

JPLIS-FIL Server Performance Comparisons 1

AFS Server Performance Comparisons

Bo Tretta Kim Kimball Jet Propulsion Laboratory Information Services - FIL Service http://fil.jpl.nasa.gov SLAC AFS Best Practices Workshop March 24, 2004

SLIDE 2

2 JPLIS-FIL Server Performance Comparisons 2

n Performance benchmarks of various hardware configurations n Range of ages for hardware n Different AFS versions n Most hardware is already in operation and limits the testing we can perform n Partitioned network n Testing was performed on both sides of the network firewall. n Operational Implications

SLIDE 3

3 JPLIS-FIL Server Performance Comparisons 3

Can we use less expensive hardware and still meet performance goals?

SLIDE 4

4 JPLIS-FIL Server Performance Comparisons 4

Cell Configuration

n At the Jet Propulsion Laboratory in Pasadena, the AFS cell is configured with a firewall that splits the database servers as well as the fileservers.

SLIDE 5

5 JPLIS-FIL Server Performance Comparisons 5

Internet

File Servers Database Servers Firewall Flight Ops Firewall Clients File Servers Database Servers Clients Clients Database Servers File Servers Test Cell

afs06, afs07, afs20

SLIDE 6

6 JPLIS-FIL Server Performance Comparisons 6

Benchmarking

n First assessment of OpenAFS fileserver hardware using Andrew Benchmark. n Initial goal: Determine if further assessment of inexpensive fileservers is warranted – without wasting time and resources in the initial trials.

SLIDE 7

7 JPLIS-FIL Server Performance Comparisons 7

Methodology

n Compare Andrew Benchmark results from inexpensive Intel-based fileservers with results from existing Sun Solaris fileservers.

SLIDE 8

8 JPLIS-FIL Server Performance Comparisons 8

Host O.S. AFS software Hardware Memory (MB) Proc 1 (MHz) Proc 1 (MHz) Storage Device afs06 Solaris8 3.6 2.45 Ultra - 1 256 167 N/A Sparc Array afs07 Solaris8 3.6 2.45 Ultra - 1 256 167 N/A Sparc Array afs12 Solaris8 3.6 2.45 Ultra - 2 256 N/A 2 * Sparc Array afs15 Solaris8 3.6 2.45 Ultra - 2 256 200 200 A 3500 afs16 Solaris8 3.6 2.45 Ultra -2 256 296 N/A 2 * A5000 afs17 Solaris8 3.6 2.45 Ultra - 60 256 450 450 A3500 afs18 Solaris8 3.6 2.45 Ultra - 60 1536 450 450 A3500 afs19 Solaris8 3.6 2.45 E 420 R 2048 450 450 A3500 afs20 Solaris8 3.6 2.45 280 R 2048 900 900 2 * T3 afslinux01 RH Enterprise 3 1.2.11 Compaq ML330 640 P3 1.4 GHz N/A HP Storageworks Smart Array afslinux02 RH Enterprise 3 1.2.11 Compaq ML330 640 P3 1.4 GHz N/A HP Storageworks Smart Array afslinux03 RH Enterprise 3 1.2.11 Aberdeen 845 PE 512 P4 2.4 GHz N/A ATA 100 Internal afstest03 Solaris8 3.6 2.45 Ultra - 1 256 167 N/A Sparc Array afstest05 Solaris8 3.6 2.51 Ultra - 1 256 167 N/A Sparc Array

SLIDE 9

9 JPLIS-FIL Server Performance Comparisons 9

Performance test from a client outside of the firewall

SLIDE 10

10 JPLIS-FIL Server Performance Comparisons 10

Observations

n afslinux02 was not performing the same as afslinux01 which is identical hardware. n After examining afslinux02, it was found that the L2 cache module was not installed. n The subsequent tests were performed with the L2 cache module installed.

SLIDE 11

11 JPLIS-FIL Server Performance Comparisons 11

Performance test from a client outside of the firewall with L2 cache installed

SLIDE 12

12 JPLIS-FIL Server Performance Comparisons 12

Performance test from a client inside of the firewall. Did not test to the test cell systems because the production servers can not be modified.

SLIDE 13

13 JPLIS-FIL Server Performance Comparisons 13

Conclusions

n Inexpensive hardware for OpenAFS fileservers is not ruled out. n Follow on: Proceed to stress testing to determine feasible transaction rates.

SLIDE 14

14 JPLIS-FIL Server Performance Comparisons 14

Examining a Myth

n Expensive “big iron” is frequently purchased because “we know it will get the job done.” n The result can be a collection of a small number of expensive fileservers.

SLIDE 15

15 JPLIS-FIL Server Performance Comparisons 15

But…

n A larger number of inexpensive fileservers may provide equivalent performance. n It may be that the initial investment in the larger number of less expensive fileservers does not significantly differ from the smaller collection of “big iron” for a given aggregate transaction load.

SLIDE 16

16 JPLIS-FIL Server Performance Comparisons 16

Benefits

n Initial cost aside, the “more and cheaper fileserver” approach offers:

w Inexpensive incremental increase in capacity. w Cost effective redundancy. w Better manageability.

SLIDE 17

17 JPLIS-FIL Server Performance Comparisons 17

Cost effective redundancy

n If a fileserver fails, it can be immediately replaced, with similar (or identical) hardware kept for this purpose.

SLIDE 18

18 JPLIS-FIL Server Performance Comparisons 18

Better manageability

n We theorize that it will be easier to take a machine

ut of service when it houses less RW data.

SLIDE 19

19 JPLIS-FIL Server Performance Comparisons 19

Assumption

n That the time required to move the Read/Write volumes in the OpenAFS namei implementation does not increase to a point that volume moves are truly abysmal.

SLIDE 20

20 JPLIS-FIL Server Performance Comparisons 20

Benefit

n It will be easier to recover from unexpected hardware failure. It’s easier to justify a “spare server” at $10,000 than at $100,000.

SLIDE 21

21 JPLIS-FIL Server Performance Comparisons 21

AFS Server Performance Comparisons Bo Tretta Kim Kimball Jet - - PowerPoint PPT Presentation

AFS Server Performance Comparisons

Can we use less expensive hardware and still meet performance goals?

Cell Configuration

n At the Jet Propulsion Laboratory in Pasadena, the AFS cell is configured with a firewall that splits the database servers as well as the fileservers.

Internet

File Servers Database Servers Firewall Flight Ops Firewall Clients File Servers Database Servers Clients Clients Database Servers File Servers Test Cell

Benchmarking

n First assessment of OpenAFS fileserver hardware using Andrew Benchmark. n Initial goal: Determine if further assessment of inexpensive fileservers is warranted – without wasting time and resources in the initial trials.

Methodology

n Compare Andrew Benchmark results from inexpensive Intel-based fileservers with results from existing Sun Solaris fileservers.

Performance test from a client outside of the firewall

Observations

n afslinux02 was not performing the same as afslinux01 which is identical hardware. n After examining afslinux02, it was found that the L2 cache module was not installed. n The subsequent tests were performed with the L2 cache module installed.

Performance test from a client outside of the firewall with L2 cache installed

Performance test from a client inside of the firewall. Did not test to the test cell systems because the production servers can not be modified.

Conclusions

n Inexpensive hardware for OpenAFS fileservers is not ruled out. n Follow on: Proceed to stress testing to determine feasible transaction rates.

Examining a Myth

n Expensive “big iron” is frequently purchased because “we know it will get the job done.” n The result can be a collection of a small number of expensive fileservers.

But…

n A larger number of inexpensive fileservers may provide equivalent performance. n It may be that the initial investment in the larger number of less expensive fileservers does not significantly differ from the smaller collection of “big iron” for a given aggregate transaction load.

Benefits

n Initial cost aside, the “more and cheaper fileserver” approach offers:

w Inexpensive incremental increase in capacity. w Cost effective redundancy. w Better manageability.

Cost effective redundancy

n If a fileserver fails, it can be immediately replaced, with similar (or identical) hardware kept for this purpose.

Better manageability

n We theorize that it will be easier to take a machine

Assumption

n That the time required to move the Read/Write volumes in the OpenAFS namei implementation does not increase to a point that volume moves are truly abysmal.

Benefit

n It will be easier to recover from unexpected hardware failure. It’s easier to justify a “spare server” at $10,000 than at $100,000.

Contact Information

Bo Tretta – botretta@jpl.nasa.gov Kim Kimball – Kim.Kimball@jpl.nasa.gov