Network writes contantly spiking in throughput


 
Thread Tools Search this Thread
Operating Systems Solaris Network writes contantly spiking in throughput
# 1  
Old 12-06-2009
Network writes contantly spiking in throughput

Hey guys

First post... and im not exactly a solaris guru but here goes

Ive setup a solaris 10 box with a raidz2 set of 6 disks...

I have also setup Samba with open shares for some CIFs access...

now my issue is that when i transfer large files to it the network performance contantly spikes at about 40% of the GB connection and then plummets to 0% and keeps doing that contantly ( goes up then down every 4 seconds or so ) - This is from a windows box and ive tried multiple boxes

I did some read tests from it and i dont see the same problem... guessing a buffer problem or something?

Im running an ABIT il-90mv MB which has an integrated Intel 82573L GB network adapter... its running the normal e1000g driver...

any help would be massively appreciated as im trying to get some decent centralized storage at home...
# 2  
Old 12-06-2009
have you tuned the mtu packet size? what kind of switch (managed/unmanaged) do you use? can you connect the boxes with a cross cabel and check if the problem persists? is autoneg used?
# 3  
Old 12-06-2009
The writes from CIFS are probably synchronous to ensure reliable data transfer all the way from the client to disk platters. If so, what you're seeing is a feature of ZFS. Read the "ZFS Evil Tuning Guide" and pay particular attention to the ARC and ZIL sections. You can try disabling your ZIL and see if your jumpiness disappears.

FWIW, unless you have a need to have deterministic performance at all times, I wouldn't worry too much about what you're seeing. Especially if the jumpiness in IO throughput disappears if you test with the ZIL disabled.

In my opinion, ZFS has a weakness in its ability to guarantee deterministic performance. Similar to what you're seeing, in my experience ZFS tends to be very episodic in the way it writes data to disk. Most of the time it's pretty fast, but every few seconds I've noted that when it is under a continuous write load, ZFS tends to grind to a short halt.
# 4  
Old 12-07-2009
Quote:
Originally Posted by silicoon
now my issue is that when i transfer large files to it the network performance contantly spikes at about 40% of the GB connection and then plummets to 0% and keeps doing that contantly ( goes up then down every 4 seconds or so ) - This is from a windows box and ive tried multiple boxes
Doesn't looks like a problem to me as long as the disks are ~100% busy, which you you don't tell. Gigabit Ethernet is simply not the bottleneck here.
What says:
Code:
iostat -xtc 1 10
zpool iostat 1 10

during a transfer ?
# 5  
Old 12-08-2009
Cheers guys - im looking into the other posts but here are the iostats etc

Code:
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.1    0.0    0.6    0.3  0.0  0.0   11.8   0   0    0    0   0  1  0 99
sd1       1.1    2.2   65.1  134.6  0.4  0.0  112.3   1   1
sd2       1.1    2.2   65.1  134.6  0.3  0.0  108.0   1   1
sd3       1.1    2.2   65.1  134.6  0.3  0.0   99.3   1   1
sd4       1.1    2.2   65.1  134.6  0.3  0.0   96.5   1   1
sd5       1.1    2.2   65.1  134.6  0.4  0.0  114.9   1   1
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  617   3 12  0 85
sd1       0.0  184.3    0.0 9849.2 18.3  0.6  102.4  58  58
sd2       0.0  170.9    0.0 8396.1 15.6  0.5   94.5  50  50
sd3       0.0  132.8    0.0 6108.0 11.1  0.4   86.5  38  38
sd4       0.0  123.5    0.0 5579.4  9.4  0.4   78.6  35  35
sd5       0.0  179.1    0.0 9499.7 17.7  0.6  101.8  57  58
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  623   7 29  0 64
sd1       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd2       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd3       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd4       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd5       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    2.0    0.0   28.0  0.0  0.0    0.4   0   0    0  620   4 30  0 66
sd1       0.0   53.0    0.0 3607.8  6.7  0.3  131.8  26  26
sd2       0.0   59.0    0.0 3906.8  7.1  0.3  124.0  26  26
sd3       0.0   69.0    0.0 4255.8  7.6  0.3  114.5  26  26
sd4       0.0   64.0    0.0 4250.8  7.2  0.3  116.4  26  26
sd5       0.0   59.0    0.0 3571.8  7.4  0.3  129.5  26  26
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  623   0  6  0 94
sd1       0.0  191.0    0.0 13612.1 34.0  1.0  183.1 100 100
sd2       0.0  206.0    0.0 14725.2 34.0  1.0  169.8 100 100
sd3       0.0  230.0    0.0 16393.9 34.0  1.0  152.1 100 100
sd4       0.0  229.0    0.0 16351.4 34.0  1.0  152.7 100 100
sd5       0.0  194.0    0.0 13869.1 34.0  1.0  180.3 100 100
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  625   2  8  0 90
sd1       0.0  219.0    0.0 12037.3 23.8  0.7  112.0  74  74
sd2       0.0  206.0    0.0 10624.7 21.3  0.7  106.5  66  66
sd3       0.0  163.0    0.0 8611.5 16.2  0.5  102.7  54  55
sd4       0.0  163.0    0.0 8664.0 16.0  0.5  101.7  53  53
sd5       0.0  205.0    0.0 11820.7 23.3  0.7  117.5  74  74
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  623   7 29  0 64
sd1       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd2       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd3       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd4       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
sd5       0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  617   5 33  0 62
sd1       0.0   39.0    0.0 2158.8  3.8  0.2  100.8  15  15
sd2       0.0   39.0    0.0 2288.8  3.7  0.2   99.6  15  15
sd3       0.0   38.0    0.0 2581.3  3.4  0.2   92.3  15  15
sd4       0.0   42.0    0.0 2331.8  3.9  0.2   96.6  15  15
sd5       0.0   38.0    0.0 2073.8  3.7  0.2  102.7  15  15
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  620   0  6  0 94
sd1       0.0  193.0    0.0 13781.0 34.0  1.0  181.3 100 100
sd2       0.0  208.0    0.0 14853.0 34.0  1.0  168.2 100 100
sd3       0.0  230.0    0.0 16393.9 34.0  1.0  152.1 100 100
sd4       0.0  226.0    0.0 16093.9 34.0  1.0  154.8 100 100
sd5       0.0  196.0    0.0 13997.5 34.0  1.0  178.5 100 100
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0
                 extended device statistics                    tty         cpu
device    r/s    w/s   kr/s   kw/s wait actv  svc_t  %w  %b  tin tout  us sy wt id
cmdk0     0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0    0  625   2  8  0 90
sd1       0.0  225.0    0.0 13319.0 27.0  0.8  123.9  84  84
sd2       0.0  212.0    0.0 12122.2 24.5  0.8  119.1  77  77
sd3       0.0  183.0    0.0 10291.8 19.5  0.6  110.0  64  64
sd4       0.0  191.0    0.0 10832.4 20.4  0.7  110.2  67  67
sd5       0.0  229.1    0.0 13188.4 27.0  0.8  121.4  85  85
nfs1      0.0    0.0    0.0    0.0  0.0  0.0    0.0   0   0


               capacity     operations    bandwidth
pool         used  avail   read  write   read  write
----------  -----  -----  -----  -----  -----  -----
cifs        2.50T  2.05T      2      3   323K   408K
cifs        2.50T  2.05T      0    318      0  39.8M
cifs        2.50T  2.05T      1    374   1014  36.9M
cifs        2.50T  2.05T      0      0      0      0
cifs        2.50T  2.05T      0      8      0   535K
cifs        2.50T  2.05T      0    318      0  39.9M
cifs        2.50T  2.05T      0    432      0  44.2M
cifs        2.50T  2.05T      0      0      0      0
cifs        2.50T  2.05T      0     71      0  8.08M
cifs        2.50T  2.05T      0    320      0  40.1M

as you can see the through put hits around 40MB/s and then goes back down... wait times hit 100% :S seems like its hitting a cache wall or something :S not too familure with solaris but i work with Netapps and its basically the same filesystem... difference is obviously a decent amount of cache infront of the disk :P

Last edited by silicoon; 12-08-2009 at 08:17 AM..
# 6  
Old 12-10-2009
any one?
# 7  
Old 12-10-2009
Quote:
Originally Posted by silicoon
as you can see the through put hits around 40MB/s and then goes back down... wait times hit 100% :S seems like its hitting a cache wall or something :S
The OS is busy doing something when no IOs are going on. Might want to look what it is using dtrace.
How much free space is available on the pool ?
Have you enabled data compression ?
Are you sending multiple large files in parallel ?
Service time doesn't look great.
Quote:
not too familure with solaris but i work with Netapps and its basically the same filesystem...
I don't believe Netapps uses ZFS.
Login or Register to Ask a Question

Previous Thread | Next Thread

9 More Discussions You Might Find Interesting

1. IP Networking

OID for Bandwith and Throughput Measurement

Hey Guys, Does anybody know, which OID's of Net-SNMP is used to collect throughput and bandwith usage of machine?? I got these OID's ..iso.org.dod.internet.mgmt.mib-2.interfaces.ifTable.ifEntry.ifOutOctets ..1.3.6.1.2.1.2.2.1.16 ... (1 Reply)
Discussion started by: franzramadhan
1 Replies

2. IP Networking

Issue with ns2 - no throughput data

Hello, First time poster here hoping to get some help with ns2. I've recently started using ns2(first time user) but I'm having difficulty getting the results I'm after. I am trying to set up a network with wireless nodes(5-15 nodes) and then use xgraph to display a timing diagram,... (0 Replies)
Discussion started by: UnicksMan
0 Replies

3. Solaris

Throughput problems with solaris aggregation

Hello gurus, I have the following configuration in the server side: # dladm show-aggr key: 33 (0x0021) policy: L4 address: 0:14:4f:6c:11:8 (auto) device address speed duplex link state nxge0 0:14:4f:6c:11:8 1000 Mbps ... (3 Replies)
Discussion started by: FERCA
3 Replies

4. HP-UX

Performance - CPU spiking

We have a DB server which is constantly utilised above 95% above. This is becoming nuisance when the monitoring team frequently calls to check on it. Frankly I do not know what to tweak or even interpret the outputs. I noticed constant 30 to 60% in wio column of the cpu utilisation. There... (1 Reply)
Discussion started by: sundar63
1 Replies

5. IP Networking

How to improve throughput?

I have a 10Gbps network link connecting two machines A and B. I want to transfer 20GB data from A to B using TCP. With default setting, I can use 50% bandwidth. How to improve the throughput? Is there any way to make throughput as close to 10Gbps as possible? thanks~ :) (3 Replies)
Discussion started by: andrewust
3 Replies

6. IP Networking

Gigabit Link throughput

As a rule of thumb in doing calculations, what figure would you use in Mbytes/sec? I know the answer varies grealty on the topolgy of the network but I wonde what newteok engineers use a rough rule of thumb? Many thanks. (1 Reply)
Discussion started by: debd
1 Replies

7. UNIX for Advanced & Expert Users

tool to monitor throughput

Was wonder if there was a tool or program I could run to measure throughput on our CentoS 4.x server. Our current dedicated host provider is charging us by how much throughput we are using and I just want to see if their numbers add up to whatever I get using a throughput tool of some kind. ... (6 Replies)
Discussion started by: mcraul
6 Replies

8. Shell Programming and Scripting

Contantly check for file at the remote server and ftp when avaliable

Hello, I used the search engine but could not find the solution I am looking for. Probably its simple but I do not know the solution. My requirement is I ftp files every day in the morning from the remote server to the local machine. Now if the files are not there at the time I schedule the ftp... (2 Replies)
Discussion started by: kiran_418
2 Replies

9. UNIX for Advanced & Expert Users

how can i test my tape throughput in Mb/sec?

Is there a tool or cmd line program i can use to test my tape throughput in Mb/sec? thank you (2 Replies)
Discussion started by: progressdll
2 Replies
Login or Register to Ask a Question